The msot important question about corel duo, is if dual processors Apple Machines with ATI X1900XT woul be a better machine than AMD and Pentium with one processor at same clocks. Better when core duo would arrive to 3 Mhz or nearby. Becouse hard gamers, and renderers would buy this (expending a few more bucks) and have the choice of use MAC OS, Win XP, an Linux on the same machine with opengl games probably going better in Leopard than in XP becouse of the better networking of UNIX and Linux over TCP/IP. The test of packets losed with Quake3 XP vs Leopard would be a great test, becouse probably AMD, and Pentium with same clocks and ATI would have similar preformance in games that are not prepared for two processors, but Photoshop CS2, blender And other CAD/CAM apps would run better. Perhaps The future new market of Apple machines are hard gamers, and hard users like architects, renderers, animators and so one. But it must be tested. And I want to know if MAC mainboards are better than ASUS and Gigabyte ones (or other better if them exists).
Congrat AT commenters! I have to say, this is the new HIGH SCORE for useless, incorrect, biased, self important posts in the history of AT!
Really, i was going to respond to each one in turn, but I think its far easier just to make this one post where i point out that many, many, many of you should likely try lurking a bit instead of instantly hitting the reply button and spouting off about latencys, bus widths, and other thing your Toms Hardware Education degree has certified you an expert at. We will all be more intelligent if you didnt post.
They should have used low latency DDR2-533 rather than the normal JEDEC specified 4-4-4-12 latency for their reviews. It might be faster then :)).
Also, testing Sonoma notebooks have shown that it likes single channel DDR2-533 better than DDR2-400, like how it doesn't benefit from dual channel. I would also like to see DDR2-667 results(over dual channel DDR2-533), as few % here and there will really show Yonah's potential.
If any of you actually care to search for transistor performance of Intel's and AMD/IBM, you can see that AMD/IBM's the newest 65nm process is only 2-3% faster but Intel is providing the numbers at HALF the leakage.
There WILL be X2 clock speed like versions of Yonah with higher TDP and being graded as EE.
-Equal platform comparisons are never possible.
-DDR2's power advantage isn't as great as you think.
-We don't know if Turion would benefit at all from DDR2 in performance, the claimed 15% or so is at best case, aka single benchmark. It always happens, companies say some wonder number and in reality its even worse than the previous one.
Correct me if I'm wrong. Doesn't the AMD architecture have a 3 cycle L1 latency due to an exclusive L1 L2 cache relationship. While Intel uses a 2 cycle L1 inclusive L1 L2 cache relationship. With the larger cache sizes now, the more costly exclusive set seems to be holding AMD back. However, this higher latency could be the reason AMD is able to reach higher speeds using a lower process.
As for the power consumption, I wonder if the board design had anything to do with the X2 being 30% higher. Chime in here
On die memory controller advantage AMD.
DDR2 lower power consumption advantage Intel
65nm process advantage Intel
Mature SOI advantage AMD.
You are wrong about the cache of Yonah. The mobile CPU is different with the Netburst.Yonah's L1 latency is 3 cycles, and it is a kind of write-back cache, which needs not always copy the data to L2. L2 latency is 14 cycles(AT said), which is the same with AthonX2. And Yonah's number of pipeline stages is 11,12, or 13. The AthonX2 is 12-stage. So, (Include AT)we believe that Yonah can reach the high frequency. The real reason of Yonah only max 2.16GHz -- for the moblie applications ... to control the power sum.
Yup... as I posted above... the rumor is that Yonah was designed for 2.5GHz (maybe even 2.6GHz or so) and can easily be overclocked to those speeds but is being launched at the speeds it is for exactly what you say... to fit in a certain power envelope.
We'll, that's definitely a logical argument, but I'd say only marginally relavent to this discussion. If a laptop with X% better power consumption comes out, does it really matter in the end where those X% came from. I can see the validity in saying that maybe some of those optimizations are on the board and thus the CPU shouldn't be given all the credit, but lower consumption is lower consumption. But that's all hypothetical. The reality is that even without optimizations, the Pentium M would have enjoyed lower power consumption just by moving to the 65nm proccess, as would any proccessor. that's where most of that 30% came from. I'd venture to say that the board design provides 1 or 2 % max, and definitely within the margin of error.
><eno
Um, i got confused there a bit. I'm basically saying that the new "Core Duo" derives the majority of it's efficiency not from any architectual changes (although they do help a lot), but from the move to 65nm. When AMD makes that jump to (and if Intel hasn't already gone to 45nm) then you'll see AMD pull ahead in power consumption on all 65nm parts, if not performance as well. Remember, aside from the on-die controller, this new architecture is very similar to what AMD is doing. Like the author said, if they wanted to copy something, they should at least copy the serial bus!
All one know that Dothan is 90nm.But it is very cool. hot == Intel's 90nm? No!
The CPU's architecture of low power consume is key. We must understand that IC design is very complex, many advanced tech are applied to keep CPU cool, include the CPU's architecture(such as micro-ops fusion).
Yonah is better? It seems to be like its a pretty even match clock-for-clock but if you consider that X2s max out at 2.4GHz (currently) and Yonah will launch in around 15 days at 2.13GHz (the FX-60 is supposed to launch around the same time, so the X2 max clock may hit 2.6GHz when Yonah arrives) then it is clearly inferior in the performance department. Granted, the power consumption is quite a bit better but I'd hardly say that the CPU is better, just better suited for low-power applications.
Yonah(2GHz/L2 2M) is better than AthonX2(2GHz/L2 2M) in the many cases.
Everyone also believe that Yonah is able to reach the very high frequency.
In the past, We also known that PM can be overclocked better than A64. The PM's best record of Super Pi is much fast than A64/FX.
Yonah Yonah Yonah... Nah it was hype as I suspected... these procesors are vitually identical to AMD's old A-64's.. AMD supposed to get signifigant bumps w. DDR2/3 and PCIe onboard plus I read about some technology AMD and IBM made to increase performance by a whopping 40%!!
http://www.guru3d.com/newsitem.php?id=3370">http://www.guru3d.com/newsitem.php?id=3370
AMD has nothing to fear other than in power arena when longer pipe Conroe comes and even then that advantage may very well disappear.
The 40% advantage is TRANSISTOR performance. Even then, its 40% advantage when no straining is included at all. AMD has strained silicon in their current process. Which means real advantage is FAR less than 40%.
PCIe onboard is not a speed advantage. It´s more a tactic move towards the the computer on a chip solution.
AMD is keen on ondie memory controller and in some years PCIe aswell due to the fact they suck at chipsets. Sure nVidia is making some good ones now. But it´s always been Intels extra strong point.
PCIe would also gives less traces on the motherboard that gives lower cost. But your CPU will break the 1300pin mark atleast and will be unable to scale in any way. Like AMDs backwards setting with DDR. When some new PCIe or more PCIe lanes comes out. You would need a new socket.
However personally I´m in favour of it. I just think AMDs ondie solutions is the wrong way. On package seems alot better and more flexible. Until we have an even more stagnant development cycle for external parts.
DDR2 advantage? I wouldn't count on it. The latency on DDR2 still sucks compared to DDR. I wouldn't be so quick to scoff - Yonah performs as well as the X2 while consuming only a fraction of the electrical power.
I don't see why people thought Yonah was going to be some sort of X2 destroyer. Its execution core has far fewer units than K8 and it lacks an on-die memory controler. The fact that it can match an X2 with slimmer cores makes it all the more impressive. The only hype was in your mind.
The AMD/IBM technology is not in processors that we can benchmark today -- they have yet to bring this process to mass manufacturing.
quote: Its execution core has far fewer units than K8 and it lacks an on-die memory controler.
Which should make you think about the supposed "magic" of the on-die memory controller that everyone constantly raves about as being *the* reason why the Athlon64 is as good as it is (hint: there are other reasons, the IMC is just one piece of why).
I don't get it why everyone expects a mobile chip to out-perform a desktop chip.
Yonah/Core Duo is specifically a low power optimised dual core processor for laptops. Its fine to use it as an indication of the future conroe/merom, but bear in mind that these will see faster clocks, further optimisations for their specific use (desktop/mobile respectively).
Next month, Intel will have no competition for dual core laptops untill the Turion X2 comes out. Will you think that the mobile version of a desktop processor will be able to perform the same? No, because of optimisations for the all-important mobility factor (battery-life and weight).
If you want an uber-powerful desktop replacement X2 processor-powered laptop, you can do that now. But expect no battery life and an aching back or a soar arm from carrying around a 4kg+ brick. So, Intel is now offering a laptop processor, capable of performing similarly to a desktop processor, but with low power consumption, long battery life and low weight.
What am I trying to say? A Turion X2 may be slower than current Athlon X2s and yes I know 65nm vs 90nm. It probably won't be faster anyway, so Intel have a fast, efficient and cheap processor - hey you expect Turion X2 to be as cheap as the single core version?
Apples to oranges? Get over it! The Core Duo uses DDR2, AMD X2 DDR. Intel FSB, AMD Mem controller. You will never get an apples to apples comparison - its business, competition, differentiated products, call it what you like, you buy AMD, you use different RAM, different Mobos, different drivers to Intel, end of story, so its a comparison of platforms, not CPU.
how do you know its going to be cheap? As far as I have seen, there are no prices released for these processors yet. And do you wanna know why we're somewhat dissapointed? Intel (and its fanbois) have been talking about how this new arc. is supposed to be the next big thing. In its current form, its not really.
With AMD announcing the use of SiGe in processors coming up very soon, it looks like they could still get a decent drop in power consumption at 90nm. I'd guess that's what they're waiting for to release the Turion x2. I doubt that it will be quite enough to match the power consumption of Yonah, but should mostly suffice until the 65nm shrink. It will also apparently serve to greatly reduce the area required for L2 cache by shrinking the cells. Once that happens I imagine that AMD will hold a small power/performance lead over Merom/Conroe/Woodcrest since adding 64 bit will tend to up their power consumption somewhat. At that point, AMD will be using SOI and SiGe, while Intel will only be using SiGe.
As for continuing the ramp of frequencies, I don't think that it's going to play a hugely significant role going forward. That's why everyone is going multi-core these days. It looks like rough performance parity is here to stay between the two. What I think AMD is aiming at is a duopoly in the CPU market. That is the point of the lawsuit, and the greatly expanding capacity that AMD is targeting (Fab36, Chartered, Fab38, etc...). What Intel really needs to worry about is AMD's ability to increase their brand awareness and supply large vendors reliably. Once that happens, Intel will need to compete on price. While AMD is very comfortable living on $100 ASP, that would be catastrophic to Intel's bottom line. I think over the next few years (barring any catastrophic screw-ups by AMD) we are going to see alot of cost-cutting at Intel.
Sorry for the off-topic ramblings, but I tend to look at things from an investors' POV.
So I guess the biggest question for these (at least to me, since I don't have a need for an uber-lappy) is price and a converter for a mobo, and I guess OC-abilty. If they can match clock speeds of the x2s (2.6GHz+) and come out to be cheaper, then I might consider buying one of these, but I don't see AMD's desktop market being threatened, considering Intel's last gasps with netburst (let it die already!).
If AMD can milk another 400MHz or so out of its K8 arctechture in its current form, they'll stay competitive with mermom it seems to me. Yonah certainly isn't the AMD killer the intel fanboi's have been heralding it as though.
I've seen a number of "reports" of these Yonahs (Core Duo) hitting 2.5GHz quite easily without even raising the voltage. It's "as if" it were designed to run in the 2.5-2.6GHz range but not released that way... yet.
That's a system power consumption, I was referring to CPU only. And in a laptop environment, the power usage of the other components are much smaller so the impact of the CPU portion is greater. 90nm single Turions are already uncompetitive with Yonah, making them dual core will just make it worse.
Dothan vs Turion ML has a 2X or greater edge under load. Yonah has comparable power consumption to Dothan while 90nm dual core Turions will clearly go up. 3X is not out of the question for non-undervolted DC Turions.
Ye I would guess on something like 35W will be 40-45W with 65nm Turion X2, unless you sacrifice speeds. The 25W part might be 35-40W.
But Turion really never had a chance, since it´s not designed for low power. And as we already saw in the benchies. Who want a dualcore Turion running at 1.8Ghz or less against a 2.13Ghz Yonah when the yonah uses less power.
Intel briliant move so Dothan->Yonah only gave 9% more transistors. Turion->Turion X2 will add 110-120% more transistors (Over 100% due to crossbar between CPUs).
So Yonah will also be cheaper to make than a dualcore Turion.
Intel did slice the "cache per core" in half with Yonah, so AMD could conceivably make Dual-core Turions have 512k per core, which would make the die-size increase around 50%, though this will probably have a greater impact on performance on the AMD side, since Dothan's cache was insanely huge to begin with.
About the price: AMD Turions will always be cheaper than their direct analogs from Intel because AMD needs to perform the same AND have a better price for people to use it, otherwise they'll go with the market leader, so I dont think we'll ever be faced with chosing between a 1.8GHz Turion and a 2.13GHz Yonah. This is regardless of the production cost, though AMD's margins may take a big hit if Intel pushes prices hard enough.
I doubt a a laptop uses 108W and got a x800XT. That powerchart is close to useless.
A Core Duo laptop is 31 peak for CPU and 2.5 for the 945GM. A single core tution is 35W. unless you want 25W part, but then we gotta use the lower intel part aswell.
Dualcore Turion wont even stand a chance until it get 65nm.
The 25W Turion parts are $5 more expensive than the 35W parts, hardly comparable with the price difference between regular and LV Intel CPUs. I agree with you in that I dont expect 90nm Turions to match Yonah's power consumption but I must say that what I've heard about the Yamato platform (which, to tell the truth, is almost nothing) makes me think that it'll be a closer match than we think.
Do you actually use 64-bit or is it just a feature checkbox? I bought some Athlon64s because they were 64-bit and allowed me to test the software I write on a 64-bit platform, for example... not that the codes I actually write require many of the features of 64-bit, though. In the past, I have written and worked on a number of projects that actually need 64-bit (typically just very large datasets which can just be used flat in 64-bit as opposed to all sorts of tiling and paging on a 32-bit machine).
Did anyone else notice that the numbers in the graph for Black and White did not match up with the spreadsheet below it? According to the graph, the Core Duo hit 44.5 FPS, the same as the X2 2ghz w/ 1mbx2, but in the sheet below, it is reported to only do 40.1 FPS, lower than even the Pentium M. In fact, it performs the worst out of all the CPU's there, at every resolution, even though this is not reported, nor mentioned.
I also find it interesting that Intel's new line is only able to match X2's, not beat them. Either way, lets see some new stuff from AMD!
I enjoyed reading this 2nd article on the next Intel platform.
Although I am basically in favor of AMD since Athlon platform, I am quite glad that Intel is closing the performance gap. It will make AMD move further ;-)
Anyway, it should be quite interesting to see how these 2 platforms compare in 64-bits tasks (Windows XP-64, Vista (is a 64-bits version available yet?) and linux based).
A 3rd opus of this article maybe? ;-)
I thought this review was much better than the first. Thanks guys. It pretty much illustrates the same thing, but it is interesting to read about new CPU's none the less.
I don't mind the name personally, I think that we uber-nerds don't like it because we actually know what a core is in regards to a CPU. Your typical idiot customer doesn't know what a Core, or Processor die is, and Core is just another brand name like Pentium.
While these chips look freaking awesome for laptop use, I must admit I will ver very dissapointed it Conroe launches, and only brings Intel back to equality with AMD. When a new generation of CPU launches, I generally like to think that they are releaseing something to regain their status as performance leader, not just level the playing field. (Though I guess PII/PIII/P4 al launched with worse performance than the high end of their predecessors)
If memory serves, the only one to launch with a loss in performance was the Pentium 4. The Pentium II was a great advancement over the Pentium it replaced (not the Pro) and the Pentium III performed equally, though was available at higher clock speeds. That said, I don't know where you got to grouping the PII and PIII in with the P4's poor launch status. Willamette was a joke.
Personally I wouldn't touch any of the compressed domain transcoders like DVD Shrink as they sacrifice quality for speed. Something like DVD Rebuilder combined with the excellent CCE SP encoder provide the best possible quality and are just as easy to use as DVD Shrink. It's nowhere near as fast as DVD Shrink, but I'll take better picture quality over saving a few minutes any day.
Anyone else notice that the game benches seem horribly gpu-bound? How about giving us some lower resolutions in testing?
Look at FEAR in particular: X2-4200 and X2-3800 getting exactly the same score? HELLO?!?!?!? That's not a cpu-benchmark. Gimme a break.
That's not the point. The point is to relieve the GPU so regardless of what GPUs come out a year or two from now that alieviate this bottleneck we'll have known beforehand how the CPU handles things.
Further, this is a CPU preview and as such we don't care about system performance or GPU performance; that shouldn't be the focus or included in the article. Instead, every benchmark should serve the articles purpose of comparing the CPUs - GPU bound benchmarks do NOT serve that purpose and should have their resolution lowered to serve that purpose, otherwise they should not be included because what point do they serve except fluff?
I wouldn't call it horrible.
There are some weak points, but it does give a good idea about how the Core Duo performs.
Mainly:
- It's not as good as the A64 when it comes to games
- FP is much improved but not there yet
The article does do a good job of reminding us that Yonah is just the dress rehersal for the real deal. Conroe is supposed to be faster, wider, and full of 64-bit goodness. Think about it - with Yonah's die size, it could be the Celeron-M in just 6 months!
If Yonah can compete with the A64, then unless AMD pull a fast one they're probably heading for the underdog position.
...and I'm sure some gillible sheep will buy into it.
Since the "review" tested apples to oranges AGAIN, it's not of much value for anyone looking to purchase a notebook PC because you used a desktop X2 for comparison to Intel's YAWNER -- a dual core laptop chippie.
To quote this story:
"Intel’s Core Duo launches in January at CES, so if you’ve been thinking about buying a new laptop, we’d suggest waiting at least another month or so. You won’t be disappointed. "
-- Now if that ain't fanboy, what is???
Obviously with Turion stealing a lot of sales from Centrino, it's no surprise Intel is stroking the media to gain as much positive hype on uncompetitive products as it can since it knows it will be at least '07 if not later before it can compete with AMD in any market segment based on performance, value and power consumption. That however won't stop the Intel shilling.
AnandTech is one of the most fair review sites on the net and has been one of the biggest champions of AMD products for years. Your "comments" are little more than pathetic insults against Anand, who is and always will be more credible than you.
I think that he is right to some extent, though. I was hoping for a power consumption comparison between Dothan and Yonah to see which one is better for battery life but it never materialized. I would not say that AMD has a Turion that can compete with Yonah but testing Yonah in a desktop setting and then concluding that it's a heck of a laptop chip without comparing it to other laptop chips leaves a bit to be desired. Also, the tone of this review seemed a quite a bit more Intel-appeasing, if you please, as there was nothing in this review that we didn't see before except for flowery praise about how Yonah does very well without an on-die mem controller.
That said I must say that all the asking for a 2GHz 1MB L2/core A64 was pretty retarded. There is no 2GHz 1MB/core SKU so including a fake one just for comparison does not really help since there's no way to get something even similar (the 4400+ is the lowest-clocked 1MB L2 X2). It would have been nice to see an X2 4600+ (the second-best A64 SKU) compared to this Yonah (the second-best one) but I guess the 4200+ is more inline with its price.
quote: That said I must say that all the asking for a 2GHz 1MB L2/core A64 was pretty retarded. There is no 2GHz 1MB/core SKU so including a fake one just for comparison does not really help since there's no way to get something even similar (the 4400+ is the lowest-clocked 1MB L2 X2). It would have been nice to see an X2 4600+ (the second-best A64 SKU) compared to this Yonah (the second-best one) but I guess the 4200+ is more inline with its price
An overclocked Opteron 165 or underclocked Opteron 175 might have been an idea? forget about prices.
I think AMD still has some tricks up there sleeves regardless of what there roadmaps might say.
Eliminating clockspeed and using the best cache available sets the baseline for comparison, by keeping as few items of difference between the chips as possible. I fully agree with the choice, and, moreover, am quite positive there will be something out in the Turion line that will be quite similar to the 2ghz, dual channel, 1m/core cache that was used for testing.
I think it's kind of funny to see us finally returning to tests where comparing close to exactly the same clockspeed produces even mildly comparable results. I say that because years ago that's what we used to do all the time, and finally things have come nearly full circle.
Ah, but we still don't know how DDR2 will affect K8 (or K9, as AMD likes to call the dual-cores) performance. Maybe AMD will increase the L2 cache data width from 128bits to 256bits (the Pentium M has a 256bit interface) to make up for the additional latency, though I doubt it.
It's called giving good advice. Not an Intel owner myself, but even I would appreciate this info as AT obviously has more info on this. Would you rather buy a laptop now and then regret the purchase when something much better comes along in just a month from now? AT is not telling you to buy an Intel based machine, just to wait a month to get a better idea of what your options are. If you have read AT for a while, you should know that they are definitely not biased towards Intel...
Am I the only one recalling the old Athlon days?
AMD smashes ahead with a great CPU, until around what, 2600+ or so, where they run out of steam, letting Intel overtake them.
Looks like the same might happen again...
Not sure if it's just AMD screwing up, or if it's really a question of resources.
For AMD, Athlon 64 was really a last-ditch gamble. They had to do something big, or they wouldn't exist 5 years from now.
Well, they did, and enjoyed a lot of success, but they might just not have the resources to follow up on it. Instead, their only option might be to milk the A64 for all it's worth, and then take a beating for a year or two, until they're ready with a next-gen architecture
A question of availability needs to be raised. Are these new duo Intel processors available now? Can anyone say paper launch. How long has the dual Athlon64 been out? I thought so.
I dont see where you're coming from. The A64 is a K7 on steroids. The P-M is a P3 on steroids and Conroe will also just be a P-M on steroids. Both companies didnt come up with flashy, brandnew architectures over the last few years, they only added flashy things to existing designs. Here and now, it's still Intel playing the catchup game until we see Conroe in stores. And then you always have to keep in mind that AMD is yet to move to 65nm. It will certainly give them some more clockspeed headroom.
I agree with the conclusion of the article, we'll see a neck to neck race in a year from now where the better price will make the difference. AMD really doesnt have to flex its muscles now, they can milk the crowd with a superior product - and people will pay whatever to get it, or so it seems.
Several of the flashy brand new architectures of the last few years had any kind of success. Itanium/EPIC, Transmeta, Netburst... As long as Transmeta was supported by a company with very little financial power (so their loss was somewhat expected), Itanium is crawling ahead on life support, and Netburst (while being king of the hill for a good time) will be discontinued.
The point is: Netburst is certainly a design that could be considered newer or more radical compared to the P3->P-M->Yonah/Conroe/Merom route. In the end netburst was still a technical failure, but certainly not a financial mistake.
Well and Epic, thats much older than you think. But yes, it never ended up where Intel wanted it to be: on the desktop.
It seems like with Intel's new chips, we're going to be back to a point where neither company has a true lead on the other. AMD is in great position at the moment, but they need to get their Turion X2 out to combat Intel's Centrino Duo. Otherwise, they'll find themselves losing laptop sales, an area which they've come a ways in.
Right now, although clockspeed isn't increasing as fast as it once was, is a very exciting time for computer technologies. Parallelism, die shrinking, heat reduction, among other strides are paving the way to the future.
I was thinking the same, considering his thoughts are pure speculation on his part. So, until Anand can provide data to back that up, he should stick to commending AMD's architecture rather than trying to credit all their success to the on-die memory controller (which surely helps, but it's only one part in the formula).
I would doubt AMD is sitting idle though, and as they work on their next architecture will keep their lips sealed in order to maintain an element of surprise. Surprise like when they added SSE support to the Palomino and SSE3 support to Venice - both unexpected additions.
The K8s execution core is pretty much the same as the K7s. If you compare performance (in games) between the two you'll notice that the K8 performs much better. This is due to the fact that it removes the FSB bottleneck (by integrating the memory controller), increases the width of the L2 cache (and the size) and dramatically drops the memory access latency. Sure there are other minor differences but they're mostly minor improvements, like better branch prediction, etc.
I don't see where the two of you are going with this and only serve my point that pinning the successes of the architecture on the memory controller is only speculation.
Well, the memory controller is the major difference between the K7 and the K8 but if you compare performance between the two the K8 performs much better. This means that the memory controller directly lead to this increase of performance.
In truth the K8's performance is a combination of its micro-architecture and the low-latency access to the memory but since the microarchitecture came first (and was insanely bottlenecked by FSB at higher clocks) the one improvement that lead to the performance difference between the K7 and the K8 was the memory controller. In fact when AMD launched its K8 it said that there would be a 20-30% performance improvement because of the on-die memory controller.
The K8 saw a new instruction set, a slightly lengthened pipeline, SSE2 extensions, SSE3 extensions (eventually), dual-core/multi-cpu design strategies, etc. Oh, and it got an on-die memory controller among other architectural tweaks.
I don't think it's valid to attribute so many factors that could have benefitted the architecture to just the memory controller. A lot of small differences add up to a lot.
Longer pipelines lead to lower performance, the "dual-core design strategies" have nothing to do with a single-core K8's performance benefits over a K7, SSE3 is useless even now and, of course, AMD64 does not benefit 32-bit execution. The only thing that you mentioned that makes a difference is SSE2 and it doesn't really make as much of a difference on A64s as it does on P4s since SIMD vector instructions require multiple passes on the A64. The deeper buffers help, as do the increased L2 cache bandwidth and the increase in L2, but the biggest benefit does come from the integrated memory controller. Cutting access latency is insanely important but having a faster frontside bus (the bus that connects the execution core/cache to the memory controller) is probably what makes A64s perform how they perform.
quote: Longer pipelines lead to lower performance,
This is not always the case. On branchy code, it is typically true. On non-branchy code, longer pipelines can be very efficient. The problem is that typical codes on the x86 are very branchy so longer pipelines aren't that good on typical x86 codes.
As far as latency numbers and the like, you should do the math to understand why the latency helps. For large cache sizes (512M and larger), the L2 should get above 96% hit rate typically. For 1M L2, hit rates should be 98% or more. Obviously, the application you have will govern these hit rates but this is for "typical" codes. Some applications will see almost no benefit from having an L2 cache at all, for example. The latency of the main memory accesses are felt in the misses (that other 4% or 2%). If the L1 is pretty good (1 cycle penalty), you can zero that out for the calculation. Use some numbers on L2 and main memory access times to get an idea of how it really helps.
So many people just chant "integrated memory controller" as some kind of mantra without even knowing how much it *really* effects memory access times.
Longer pipelines do not help non-branchy code, higher clock speeds do. Longer pipelines allow you to raise clock speeds but if you compare two equally clocked CPUs with similar architectures but different pipeline lenghts then the longer-pipelined one will ALWAYS be slower, since both will eventually mispredict a branch and the penalty on the longer-pipelined one will take a greater hit. In the case of the K8 compared to the K7, however, the branch predictor was improved, the TLBs increased and so on, so you probably end up having the same performance per clock.
"Typical" code is code that operates on very small data sets, like a word processor. This is not what I'm talking about, however, I'm referring to code that handles massive data sets that cannot fit inside the L2 cache. These include games and streaming media. A K7 performs pretty much the same as a K8 (clock for clock) in office applications and the like, but once you have data traveling down the frontside bus (the K8s frontside bus equivalent is the link between the execution core and the memory controller, which runs at CPU clock) then the performance differences are massive. It may be true that most of the code we execute on a PC does not even touch the main memory to a significant degree but it is also true that we perceive the times when it does as a massive drop in performance. Saying that memory bandwidth (and latency, as the two are directly related) is useless is like saying that a P3 is enough for everyone.
quote: Longer pipelines do not help non-branchy code, higher clock speeds do.
Yes... and longer pipelines is one of the design parameters to achieve higher clock speeds.
quote: "Typical" code is code that operates on very small data sets, like a word processor. This is not what I'm talking about, however, I'm referring to code that handles massive data sets that cannot fit inside the L2 cache.
Yes, which is why I used "typical" there with a caveat that some workflows do not match that pattern. The math that I mentioned is not difficult to do and the percentages for hit/miss are simply parameters into the equation. You can take any instruction mix and data access pattern, analyze it, and plug the newly found percentages into the equation for a comparison. And... I never said that memory bandwidth is useless. However, I would be inclined into discussion about your bandwidth and latency being directly related (in the general form). Quite obviously, satellite communication has high bandwidth and it is equally obvious that satellite communication has a very high latency, for example.
So, your post confirms exactly what I have said and that AnandTech's benchmarks show (and the conclusions stated in the article). For the majority of applications, since data locality is high, the IMC doesn't do all that much (simply because it isn't used that much). For applications such as games and other applications with data access patterns that do not have a high degree of data locality, the IMC starts to shine. I would also argue that streaming does not fall into that category unless you have poorly optimized code. Intelligent use of prefetching, for example, can hide most of the latency penalties of main memory. I guess we could discuss what "majority of things" means and whether or not games fall into that category. ;)
[quote] Yes... and longer pipelines is one of the design parameters to achieve higher clock speeds. [/quote]
That's exactly what i said in the line that followed what you quoted. When I said that longer pipelines themselves dont help performance I meant that the clock-for-clock performance benefits of the K8 over the K7 can be mostly attributed to its on-die memory controller. Of course the bigger caches help, as do SSE2 and the other improvements, but the lion's share of the improvement comes from the integrated northbridge (the FSB was a horrible choke point in the K7).
[quote] I would be inclined into discussion about your bandwidth and latency being directly related (in the general form). Quite obviously, satellite communication has high bandwidth and it is equally obvious that satellite communication has a very high latency, for example. [/quote]
Sorry, let me clarify that a bit. When dealing with DRAM (at a set frequency) in a computer system the usable memory bandwidth is directly related the latency. It is not directly proportional but a higher latency will mean a lower usable bandwidth. This is because the memory subsystem in a PC is not just a data transport mechanism but also functions as a data storage array, which gives latency more importance (Satellite communication, on the other hand, only moves data from point to point, it does not store it or modify it in any way, it's just a conduit, which makes its bandwidth somewhat independent of the latency). Now, remember that I'm talking about usable memory bandwidth, not peak bandwidth (which is what manufacturers love to quote). Peak bandwidth is pretty much unrealizable when doing anything useful.
Anyway, I agree with you on the caches, I wanted to point out that the pipeline length itself provides no performance improvements whatsoever, and wanted to say that an integrated memory controller is a wonderful thing. Now, I say that an IMC is wonderful but it does have huge drawbacks, the main one being what AMD is currently dealing with, having to change sockets in order to update memory technology. The thing is, Intel needs flexibility because it is always updating to the newest technologies out there but AMD, on the other hand, actually gained control over the part of the traditional northbridge that affects performance the most without having to go all out and design its own chipsets like Intel does, which is why pretty much all AMD chipsets perform very similarly.
Considering that bus was developed with dual core and multi-cpu design in thought, I'd say that "dual-core design strategies" had a lot to do with the increase in performance of K8 oer K7. AMD's technical director said something to that effect in so many words in an interview here a few years back; he said they'd built K8 from the ground up for dual core, multi-cpu applications.
Actually, last time I checked, AMDs FPU (since K7) has had 3 execution units, while Intels has had 2 execution units (since pentium..2, or the original pentium, can't remember).
The P6 has two FP units: An FADD unit and an FMUL unit. One of the big weaknesses of the P6 is the fact that the FMUL unit is not fully-pipelined but instead uses part of the FADD unit for FMUL operations. The K7, on the other hand, has three fully-pipelined units, an FADD, an FMUL and an FSTORE.
No he's not. AT's hardware reviewers are nublets. One thing to note though, Dothan FPU is better than the P4's hence its gaming performance advantage over the P4 in the old tests that everyone saw. It's likely that Yonah FPU is still the same as Dothan (similar to P3) and inferior to AMD's.
Actually the FPU on the P4 was tremendously more powerful then Dothan or Yonah. While games do use the FPU, they're not that bottlenecked by it on modern systems. The reason Dothan did so well was because of its large, very low latency L2 cache. This is roughly equivilent to the primary advantage of the K8, a very low latency memory controller.
Youre right about the low latency L1/L2 caches on the Dothan, but the P4 (Williamette/Northwood) has those as well. But the P4 FPU is only powerful in SSE2 mode where it can load store larger chunks of data. Not all games use that unfortunately.
So Dothan's L2 cache is roughly 2x as fast and 4x as large. If you compare prescott with its amazingly slow L2, the situation is even more biased towards Dothan. Clearly, in terms of cache performance Dothan has a massive advantage, at least once you're out of the L1.
Second, you're confusing SSE2 and vector processing. While SSE2 can perform vector ops, it also handles plan scaler as well. In x86-64 SSE actually replaces the traditional x87 unit. The relative performance of the two is irrelevent however, the P4 was faster in both.
Dothan I agree with as having 10 Cycle Cache.
Northwood has 16 Cycle Cache.
Well you also got to keep in mind northwoods clock frequency plays a role in speeding up the cache, accces latencies for Dothan @ 2.0GHZ vs Northwood @ 3.2GHZ are basically equivalent. Though the 2.26GHZ Dothan has the fastest cache of all.
Although, "Core Duo" is a stupid name. Why does Intel have to be so different? "Core Duo" is a little confusing. Is Duo a code name? What?
However, despite the stupid name, we've really turned a corner in performance. Intel can make a good CPU when they realized speed isn't the future. Looks like I should start considering replacing my old Pentium-M IBM T40p with the awesome battery life.
AMD needs to respond in kind with a great new CPU. The future looks bright. Competition is once again is good for everyone.
Personally, i dont like that AMD is just sitting back, seemingly waiting for Intel to catch up... They need to kick Intel while their down. these new Processors from intel look really nice and i am likely to buy one, but in a mactel laptop. I am happy for INtel that they are catching up, but AMD really NEEDS to step up and do soemthing new.
AMD's income is lower than Intel uses for R&D. You really can't expect from AMD to develop something faster than Intel can.
For AMD, to have an processor they could improve a step at a time since the introduction of the Athlon64/Opteron was a need - Intel is able to mantain several teams for microprocessor development, but AMD only has money for one. And AMD will milk the market for as much as possible, selling processors that are easy to make for prices that market will accept. If AMD will start selling a higher processor grade, they would need to reduce the price for lower speed processors. This is why the 2800+ and 3000+ are discontinued - they would have to sell them too cheap.
I swear this is the best first post ive ever seen. Good Post Alex. Ya competition is very good for the market. I think intel is starting to get back on track where they need to be. It all comes down to clock speed and cost at launch. What improvements will we see with the launch of amd's next chip other than ddr2 which right now i don't really care about and possibly more cores (at least for the opteron). I am not dogging amd because for about 3-4 years ive only used AMD chips but i think amd has to raise the bar even more.
DDR2 for AMD would be great at least for a few things:
moving to single channel DDR2 memory would decrease costs (in mainboards and a bit in processors)
moving integrated graphics to single or dual channel DDR2 would increase graphic performance and overall system performance in relation to single or dual channel DDR
As for the high end, I really don't think an increase in memory bandwidth will help - not even for dual core processors. Maybe for a quad core, but quad cores are certainly for servers, and I don't know about registered DDR2 memory to be used in them.
Hmmm, you could try an Opteron Dual Core with single channel DDR memory, to see how much performance would be lost by going quad core, dual DDR.
I agree with both Alex and ncage. I really disliked Intel all through out the Pentium4/Net-burst days. They were just releasing marchitecture with no improvements whatsoever. I loved AMD for their innovation and performance/watt.
Now both companies are equal but I don't think we will see the huge fall AMD suffered from when Intel released the Pentium 4 to compete with the Athlon/K7 architecture. The beauty of competition is showing its bright colors right now. If we only had Intel, we would have a very hot/power consuming inefficient Pentium 4 based on net-burst to play Quake 4 at 5fps right now.
Its time for the fanboys to turnover a new leaf. Go Intel and AMD!!! We love both you guys.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
103 Comments
Back to Article
mitcoes - Saturday, April 22, 2006 - link
The msot important question about corel duo, is if dual processors Apple Machines with ATI X1900XT woul be a better machine than AMD and Pentium with one processor at same clocks. Better when core duo would arrive to 3 Mhz or nearby. Becouse hard gamers, and renderers would buy this (expending a few more bucks) and have the choice of use MAC OS, Win XP, an Linux on the same machine with opengl games probably going better in Leopard than in XP becouse of the better networking of UNIX and Linux over TCP/IP. The test of packets losed with Quake3 XP vs Leopard would be a great test, becouse probably AMD, and Pentium with same clocks and ATI would have similar preformance in games that are not prepared for two processors, but Photoshop CS2, blender And other CAD/CAM apps would run better. Perhaps The future new market of Apple machines are hard gamers, and hard users like architects, renderers, animators and so one. But it must be tested. And I want to know if MAC mainboards are better than ASUS and Gigabyte ones (or other better if them exists).Cygni - Wednesday, December 21, 2005 - link
Congrat AT commenters! I have to say, this is the new HIGH SCORE for useless, incorrect, biased, self important posts in the history of AT!Really, i was going to respond to each one in turn, but I think its far easier just to make this one post where i point out that many, many, many of you should likely try lurking a bit instead of instantly hitting the reply button and spouting off about latencys, bus widths, and other thing your Toms Hardware Education degree has certified you an expert at. We will all be more intelligent if you didnt post.
IntelUser2000 - Tuesday, December 20, 2005 - link
They should have used low latency DDR2-533 rather than the normal JEDEC specified 4-4-4-12 latency for their reviews. It might be faster then :)).Also, testing Sonoma notebooks have shown that it likes single channel DDR2-533 better than DDR2-400, like how it doesn't benefit from dual channel. I would also like to see DDR2-667 results(over dual channel DDR2-533), as few % here and there will really show Yonah's potential.
coldpower27 - Tuesday, December 20, 2005 - link
Yes that would be interesting, as Anantech does have Corsair DDR2-667 3-2-2-8 available in their repitoire.StuckMojo - Tuesday, December 20, 2005 - link
I'd like to see compilation benchmarks. Lots of us use our laptops for software development.
Betwon - Tuesday, December 20, 2005 - link
Compilation?At the recent Spec CPU Cint2000 test--The most fast x86 CPU about compiler is P4 670.
176.gcc 2195/2195 ponits
PM@2.26GHz(1995/1994) is fast than FX-55@2.6GHz(1931/1933).
IntelUser2000 - Tuesday, December 20, 2005 - link
If any of you actually care to search for transistor performance of Intel's and AMD/IBM, you can see that AMD/IBM's the newest 65nm process is only 2-3% faster but Intel is providing the numbers at HALF the leakage.There WILL be X2 clock speed like versions of Yonah with higher TDP and being graded as EE.
-Equal platform comparisons are never possible.
-DDR2's power advantage isn't as great as you think.
-We don't know if Turion would benefit at all from DDR2 in performance, the claimed 15% or so is at best case, aka single benchmark. It always happens, companies say some wonder number and in reality its even worse than the previous one.
Betwon - Tuesday, December 20, 2005 - link
The very low latency of L2 cache is the main real reason? AT may be foreget that L2 can be shared, which is different with AMD.Schmide - Tuesday, December 20, 2005 - link
Correct me if I'm wrong. Doesn't the AMD architecture have a 3 cycle L1 latency due to an exclusive L1 L2 cache relationship. While Intel uses a 2 cycle L1 inclusive L1 L2 cache relationship. With the larger cache sizes now, the more costly exclusive set seems to be holding AMD back. However, this higher latency could be the reason AMD is able to reach higher speeds using a lower process.As for the power consumption, I wonder if the board design had anything to do with the X2 being 30% higher. Chime in here
On die memory controller advantage AMD.
DDR2 lower power consumption advantage Intel
65nm process advantage Intel
Mature SOI advantage AMD.
Betwon - Tuesday, December 20, 2005 - link
You are wrong about the cache of Yonah. The mobile CPU is different with the Netburst.Yonah's L1 latency is 3 cycles, and it is a kind of write-back cache, which needs not always copy the data to L2. L2 latency is 14 cycles(AT said), which is the same with AthonX2. And Yonah's number of pipeline stages is 11,12, or 13. The AthonX2 is 12-stage. So, (Include AT)we believe that Yonah can reach the high frequency. The real reason of Yonah only max 2.16GHz -- for the moblie applications ... to control the power sum.fitten - Tuesday, December 20, 2005 - link
Yup... as I posted above... the rumor is that Yonah was designed for 2.5GHz (maybe even 2.6GHz or so) and can easily be overclocked to those speeds but is being launched at the speeds it is for exactly what you say... to fit in a certain power envelope.Xenoterranos - Tuesday, December 20, 2005 - link
We'll, that's definitely a logical argument, but I'd say only marginally relavent to this discussion. If a laptop with X% better power consumption comes out, does it really matter in the end where those X% came from. I can see the validity in saying that maybe some of those optimizations are on the board and thus the CPU shouldn't be given all the credit, but lower consumption is lower consumption. But that's all hypothetical. The reality is that even without optimizations, the Pentium M would have enjoyed lower power consumption just by moving to the 65nm proccess, as would any proccessor. that's where most of that 30% came from. I'd venture to say that the board design provides 1 or 2 % max, and definitely within the margin of error.><eno
Xenoterranos - Tuesday, December 20, 2005 - link
Um, i got confused there a bit. I'm basically saying that the new "Core Duo" derives the majority of it's efficiency not from any architectual changes (although they do help a lot), but from the move to 65nm. When AMD makes that jump to (and if Intel hasn't already gone to 45nm) then you'll see AMD pull ahead in power consumption on all 65nm parts, if not performance as well. Remember, aside from the on-die controller, this new architecture is very similar to what AMD is doing. Like the author said, if they wanted to copy something, they should at least copy the serial bus!Betwon - Tuesday, December 20, 2005 - link
All one know that Dothan is 90nm.But it is very cool. hot == Intel's 90nm? No!The CPU's architecture of low power consume is key. We must understand that IC design is very complex, many advanced tech are applied to keep CPU cool, include the CPU's architecture(such as micro-ops fusion).
Furen - Tuesday, December 20, 2005 - link
Yonah is better? It seems to be like its a pretty even match clock-for-clock but if you consider that X2s max out at 2.4GHz (currently) and Yonah will launch in around 15 days at 2.13GHz (the FX-60 is supposed to launch around the same time, so the X2 max clock may hit 2.6GHz when Yonah arrives) then it is clearly inferior in the performance department. Granted, the power consumption is quite a bit better but I'd hardly say that the CPU is better, just better suited for low-power applications.Betwon - Tuesday, December 20, 2005 - link
Yonah(2GHz/L2 2M) is better than AthonX2(2GHz/L2 2M) in the many cases.Everyone also believe that Yonah is able to reach the very high frequency.
In the past, We also known that PM can be overclocked better than A64. The PM's best record of Super Pi is much fast than A64/FX.
Zebo - Monday, December 19, 2005 - link
Yonah Yonah Yonah... Nah it was hype as I suspected... these procesors are vitually identical to AMD's old A-64's.. AMD supposed to get signifigant bumps w. DDR2/3 and PCIe onboard plus I read about some technology AMD and IBM made to increase performance by a whopping 40%!!http://www.guru3d.com/newsitem.php?id=3370">http://www.guru3d.com/newsitem.php?id=3370
AMD has nothing to fear other than in power arena when longer pipe Conroe comes and even then that advantage may very well disappear.
IntelUser2000 - Tuesday, December 20, 2005 - link
The 40% advantage is TRANSISTOR performance. Even then, its 40% advantage when no straining is included at all. AMD has strained silicon in their current process. Which means real advantage is FAR less than 40%.Shintai - Monday, December 19, 2005 - link
PCIe onboard is not a speed advantage. It´s more a tactic move towards the the computer on a chip solution.AMD is keen on ondie memory controller and in some years PCIe aswell due to the fact they suck at chipsets. Sure nVidia is making some good ones now. But it´s always been Intels extra strong point.
PCIe would also gives less traces on the motherboard that gives lower cost. But your CPU will break the 1300pin mark atleast and will be unable to scale in any way. Like AMDs backwards setting with DDR. When some new PCIe or more PCIe lanes comes out. You would need a new socket.
However personally I´m in favour of it. I just think AMDs ondie solutions is the wrong way. On package seems alot better and more flexible. Until we have an even more stagnant development cycle for external parts.
stateofbeasley - Monday, December 19, 2005 - link
DDR2 advantage? I wouldn't count on it. The latency on DDR2 still sucks compared to DDR. I wouldn't be so quick to scoff - Yonah performs as well as the X2 while consuming only a fraction of the electrical power.I don't see why people thought Yonah was going to be some sort of X2 destroyer. Its execution core has far fewer units than K8 and it lacks an on-die memory controler. The fact that it can match an X2 with slimmer cores makes it all the more impressive. The only hype was in your mind.
The AMD/IBM technology is not in processors that we can benchmark today -- they have yet to bring this process to mass manufacturing.
fitten - Tuesday, December 20, 2005 - link
Which should make you think about the supposed "magic" of the on-die memory controller that everyone constantly raves about as being *the* reason why the Athlon64 is as good as it is (hint: there are other reasons, the IMC is just one piece of why).
Marmion - Monday, December 19, 2005 - link
I don't get it why everyone expects a mobile chip to out-perform a desktop chip.Yonah/Core Duo is specifically a low power optimised dual core processor for laptops. Its fine to use it as an indication of the future conroe/merom, but bear in mind that these will see faster clocks, further optimisations for their specific use (desktop/mobile respectively).
Next month, Intel will have no competition for dual core laptops untill the Turion X2 comes out. Will you think that the mobile version of a desktop processor will be able to perform the same? No, because of optimisations for the all-important mobility factor (battery-life and weight).
If you want an uber-powerful desktop replacement X2 processor-powered laptop, you can do that now. But expect no battery life and an aching back or a soar arm from carrying around a 4kg+ brick. So, Intel is now offering a laptop processor, capable of performing similarly to a desktop processor, but with low power consumption, long battery life and low weight.
What am I trying to say? A Turion X2 may be slower than current Athlon X2s and yes I know 65nm vs 90nm. It probably won't be faster anyway, so Intel have a fast, efficient and cheap processor - hey you expect Turion X2 to be as cheap as the single core version?
Apples to oranges? Get over it! The Core Duo uses DDR2, AMD X2 DDR. Intel FSB, AMD Mem controller. You will never get an apples to apples comparison - its business, competition, differentiated products, call it what you like, you buy AMD, you use different RAM, different Mobos, different drivers to Intel, end of story, so its a comparison of platforms, not CPU.
*end rant*
fitten - Tuesday, December 20, 2005 - link
They don't... and that's the point and why many are impressed with Yonah... that it *does* perform favorably to a desktop platform.
Leper Messiah - Monday, December 19, 2005 - link
how do you know its going to be cheap? As far as I have seen, there are no prices released for these processors yet. And do you wanna know why we're somewhat dissapointed? Intel (and its fanbois) have been talking about how this new arc. is supposed to be the next big thing. In its current form, its not really.stateofbeasley - Monday, December 19, 2005 - link
http://www.theinquirer.net/?article=26062">http://www.theinquirer.net/?article=26062Price points are the same as the current platform.
Yonah is not NGMA. Yonah is a P6 family processor.
People must read the news before they post!!!!
coldpower27 - Monday, December 19, 2005 - link
This isn't really new architecture, the next big thing is supposed to be Intel's NGMA, where alot of changes are going to be made.KayKay - Monday, December 19, 2005 - link
I am an AMD fanboybut the results (power consumption esp.) are very impressive
guess intel wants to keep things competitive
hope amd ups the ante in response
eetnoyer - Monday, December 19, 2005 - link
With AMD announcing the use of SiGe in processors coming up very soon, it looks like they could still get a decent drop in power consumption at 90nm. I'd guess that's what they're waiting for to release the Turion x2. I doubt that it will be quite enough to match the power consumption of Yonah, but should mostly suffice until the 65nm shrink. It will also apparently serve to greatly reduce the area required for L2 cache by shrinking the cells. Once that happens I imagine that AMD will hold a small power/performance lead over Merom/Conroe/Woodcrest since adding 64 bit will tend to up their power consumption somewhat. At that point, AMD will be using SOI and SiGe, while Intel will only be using SiGe.As for continuing the ramp of frequencies, I don't think that it's going to play a hugely significant role going forward. That's why everyone is going multi-core these days. It looks like rough performance parity is here to stay between the two. What I think AMD is aiming at is a duopoly in the CPU market. That is the point of the lawsuit, and the greatly expanding capacity that AMD is targeting (Fab36, Chartered, Fab38, etc...). What Intel really needs to worry about is AMD's ability to increase their brand awareness and supply large vendors reliably. Once that happens, Intel will need to compete on price. While AMD is very comfortable living on $100 ASP, that would be catastrophic to Intel's bottom line. I think over the next few years (barring any catastrophic screw-ups by AMD) we are going to see alot of cost-cutting at Intel.
Sorry for the off-topic ramblings, but I tend to look at things from an investors' POV.
Leper Messiah - Monday, December 19, 2005 - link
So I guess the biggest question for these (at least to me, since I don't have a need for an uber-lappy) is price and a converter for a mobo, and I guess OC-abilty. If they can match clock speeds of the x2s (2.6GHz+) and come out to be cheaper, then I might consider buying one of these, but I don't see AMD's desktop market being threatened, considering Intel's last gasps with netburst (let it die already!).If AMD can milk another 400MHz or so out of its K8 arctechture in its current form, they'll stay competitive with mermom it seems to me. Yonah certainly isn't the AMD killer the intel fanboi's have been heralding it as though.
fitten - Tuesday, December 20, 2005 - link
I've seen a number of "reports" of these Yonahs (Core Duo) hitting 2.5GHz quite easily without even raising the voltage. It's "as if" it were designed to run in the 2.5-2.6GHz range but not released that way... yet.Hacp - Monday, December 19, 2005 - link
If you want performance of an AMD X2 in a notebook package, the Yonah duo is the way to go.Griswold - Monday, December 19, 2005 - link
Or you wait for dual core Turion. Same thing.Accord99 - Monday, December 19, 2005 - link
But at 2-3x the power consumption.Houdani - Monday, December 19, 2005 - link
http://www.anandtech.com/cpuchipsets/showdoc.aspx?...">http://www.anandtech.com/cpuchipsets/showdoc.aspx?...No need to exaggerate unnecessarily.
Accord99 - Monday, December 19, 2005 - link
That's a system power consumption, I was referring to CPU only. And in a laptop environment, the power usage of the other components are much smaller so the impact of the CPU portion is greater. 90nm single Turions are already uncompetitive with Yonah, making them dual core will just make it worse.saratoga - Monday, December 19, 2005 - link
The difference will most likely be less then 2x judging by the relative power consumption of Dothan and Venice @ 90nm, so you're still wrong.Also, Yonah is a 65nm chip. It should not be surprising that it has an edge over 90nm chips.
Accord99 - Monday, December 19, 2005 - link
Dothan vs Turion ML has a 2X or greater edge under load. Yonah has comparable power consumption to Dothan while 90nm dual core Turions will clearly go up. 3X is not out of the question for non-undervolted DC Turions.Shintai - Monday, December 19, 2005 - link
Ye I would guess on something like 35W will be 40-45W with 65nm Turion X2, unless you sacrifice speeds. The 25W part might be 35-40W.But Turion really never had a chance, since it´s not designed for low power. And as we already saw in the benchies. Who want a dualcore Turion running at 1.8Ghz or less against a 2.13Ghz Yonah when the yonah uses less power.
Intel briliant move so Dothan->Yonah only gave 9% more transistors. Turion->Turion X2 will add 110-120% more transistors (Over 100% due to crossbar between CPUs).
So Yonah will also be cheaper to make than a dualcore Turion.
Furen - Thursday, December 22, 2005 - link
Intel did slice the "cache per core" in half with Yonah, so AMD could conceivably make Dual-core Turions have 512k per core, which would make the die-size increase around 50%, though this will probably have a greater impact on performance on the AMD side, since Dothan's cache was insanely huge to begin with.About the price: AMD Turions will always be cheaper than their direct analogs from Intel because AMD needs to perform the same AND have a better price for people to use it, otherwise they'll go with the market leader, so I dont think we'll ever be faced with chosing between a 1.8GHz Turion and a 2.13GHz Yonah. This is regardless of the production cost, though AMD's margins may take a big hit if Intel pushes prices hard enough.
bob661 - Monday, December 19, 2005 - link
pnw3dShintai - Monday, December 19, 2005 - link
I doubt a a laptop uses 108W and got a x800XT. That powerchart is close to useless.A Core Duo laptop is 31 peak for CPU and 2.5 for the 945GM. A single core tution is 35W. unless you want 25W part, but then we gotta use the lower intel part aswell.
Dualcore Turion wont even stand a chance until it get 65nm.
Furen - Tuesday, December 20, 2005 - link
The 25W Turion parts are $5 more expensive than the 35W parts, hardly comparable with the price difference between regular and LV Intel CPUs. I agree with you in that I dont expect 90nm Turions to match Yonah's power consumption but I must say that what I've heard about the Yamato platform (which, to tell the truth, is almost nothing) makes me think that it'll be a closer match than we think.Missing Ghost - Monday, December 19, 2005 - link
What, no 64 bit?I won't buy this.
fitten - Tuesday, December 20, 2005 - link
Do you actually use 64-bit or is it just a feature checkbox? I bought some Athlon64s because they were 64-bit and allowed me to test the software I write on a 64-bit platform, for example... not that the codes I actually write require many of the features of 64-bit, though. In the past, I have written and worked on a number of projects that actually need 64-bit (typically just very large datasets which can just be used flat in 64-bit as opposed to all sorts of tiling and paging on a 32-bit machine).Chickan - Monday, December 19, 2005 - link
Did anyone else notice that the numbers in the graph for Black and White did not match up with the spreadsheet below it? According to the graph, the Core Duo hit 44.5 FPS, the same as the X2 2ghz w/ 1mbx2, but in the sheet below, it is reported to only do 40.1 FPS, lower than even the Pentium M. In fact, it performs the worst out of all the CPU's there, at every resolution, even though this is not reported, nor mentioned.I also find it interesting that Intel's new line is only able to match X2's, not beat them. Either way, lets see some new stuff from AMD!
KazenoKoe - Monday, December 19, 2005 - link
Odd, the article clearly has it at 42.3fps on both the graph and the table. The Pentium M got 40.1 and the X2 2GHz/1MB got 44.5.WitchKing - Monday, December 19, 2005 - link
I enjoyed reading this 2nd article on the next Intel platform.Although I am basically in favor of AMD since Athlon platform, I am quite glad that Intel is closing the performance gap. It will make AMD move further ;-)
Anyway, it should be quite interesting to see how these 2 platforms compare in 64-bits tasks (Windows XP-64, Vista (is a 64-bits version available yet?) and linux based).
A 3rd opus of this article maybe? ;-)
saratoga - Monday, December 19, 2005 - link
Yonah does not support x86-64, so doing 64 bit testing is not possible.phaxmohdem - Monday, December 19, 2005 - link
I thought this review was much better than the first. Thanks guys. It pretty much illustrates the same thing, but it is interesting to read about new CPU's none the less.I don't mind the name personally, I think that we uber-nerds don't like it because we actually know what a core is in regards to a CPU. Your typical idiot customer doesn't know what a Core, or Processor die is, and Core is just another brand name like Pentium.
While these chips look freaking awesome for laptop use, I must admit I will ver very dissapointed it Conroe launches, and only brings Intel back to equality with AMD. When a new generation of CPU launches, I generally like to think that they are releaseing something to regain their status as performance leader, not just level the playing field. (Though I guess PII/PIII/P4 al launched with worse performance than the high end of their predecessors)
tfranzese - Monday, December 19, 2005 - link
If memory serves, the only one to launch with a loss in performance was the Pentium 4. The Pentium II was a great advancement over the Pentium it replaced (not the Pro) and the Pentium III performed equally, though was available at higher clock speeds. That said, I don't know where you got to grouping the PII and PIII in with the P4's poor launch status. Willamette was a joke.vailr - Monday, December 19, 2005 - link
Re: Chipset Drivers used in the review:nForce4 6.66
Intel 7.0.0.25
Check here: http://www.fdrsoft.fr.fm/">http://www.fdrsoft.fr.fm/
Intel Version 7.2.2.1006
nForce4 Version 6.70
Marlin1975 - Monday, December 19, 2005 - link
Being that DVD shrink will not be updated anymore and the creator is now part of Nero, and the Recode program. Why not use Recode?PrinceGaz - Tuesday, December 20, 2005 - link
Personally I wouldn't touch any of the compressed domain transcoders like DVD Shrink as they sacrifice quality for speed. Something like DVD Rebuilder combined with the excellent CCE SP encoder provide the best possible quality and are just as easy to use as DVD Shrink. It's nowhere near as fast as DVD Shrink, but I'll take better picture quality over saving a few minutes any day.mrred - Monday, December 19, 2005 - link
Anyone else notice that the game benches seem horribly gpu-bound? How about giving us some lower resolutions in testing?Look at FEAR in particular: X2-4200 and X2-3800 getting exactly the same score? HELLO?!?!?!? That's not a cpu-benchmark. Gimme a break.
Anemone - Tuesday, December 20, 2005 - link
Frankly I appreciate testing that shows resolutions we actually play at. Now if only we'd see 1920x1200 :)blackbrrd - Monday, December 19, 2005 - link
The games are benchmarked at 1024x768.Personally I haven't played a game below 1024x768 since I got my GF2mx four or five years ago.
Most games look horrible below 1024x768, except the games ported from consoles ;)
saratoga - Monday, December 19, 2005 - link
Yeah and above 1024 you're GPU limited. Its almost like this was a CPU review.tfranzese - Monday, December 19, 2005 - link
That's not the point. The point is to relieve the GPU so regardless of what GPUs come out a year or two from now that alieviate this bottleneck we'll have known beforehand how the CPU handles things.Further, this is a CPU preview and as such we don't care about system performance or GPU performance; that shouldn't be the focus or included in the article. Instead, every benchmark should serve the articles purpose of comparing the CPUs - GPU bound benchmarks do NOT serve that purpose and should have their resolution lowered to serve that purpose, otherwise they should not be included because what point do they serve except fluff?
tayhimself - Monday, December 19, 2005 - link
Yeah this review is horrible, just like the first Yonah review.uop - Monday, December 19, 2005 - link
I wouldn't call it horrible.There are some weak points, but it does give a good idea about how the Core Duo performs.
Mainly:
- It's not as good as the A64 when it comes to games
- FP is much improved but not there yet
The article does do a good job of reminding us that Yonah is just the dress rehersal for the real deal. Conroe is supposed to be faster, wider, and full of 64-bit goodness. Think about it - with Yonah's die size, it could be the Celeron-M in just 6 months!
If Yonah can compete with the A64, then unless AMD pull a fast one they're probably heading for the underdog position.
ozzimark - Monday, December 19, 2005 - link
they've got some headroom with clocks to play with, as the recent opterons are showing ;)Beenthere - Monday, December 19, 2005 - link
...and I'm sure some gillible sheep will buy into it.Since the "review" tested apples to oranges AGAIN, it's not of much value for anyone looking to purchase a notebook PC because you used a desktop X2 for comparison to Intel's YAWNER -- a dual core laptop chippie.
To quote this story:
"Intel’s Core Duo launches in January at CES, so if you’ve been thinking about buying a new laptop, we’d suggest waiting at least another month or so. You won’t be disappointed. "
-- Now if that ain't fanboy, what is???
Obviously with Turion stealing a lot of sales from Centrino, it's no surprise Intel is stroking the media to gain as much positive hype on uncompetitive products as it can since it knows it will be at least '07 if not later before it can compete with AMD in any market segment based on performance, value and power consumption. That however won't stop the Intel shilling.
stateofbeasley - Monday, December 19, 2005 - link
AnandTech is one of the most fair review sites on the net and has been one of the biggest champions of AMD products for years. Your "comments" are little more than pathetic insults against Anand, who is and always will be more credible than you.Furen - Monday, December 19, 2005 - link
I think that he is right to some extent, though. I was hoping for a power consumption comparison between Dothan and Yonah to see which one is better for battery life but it never materialized. I would not say that AMD has a Turion that can compete with Yonah but testing Yonah in a desktop setting and then concluding that it's a heck of a laptop chip without comparing it to other laptop chips leaves a bit to be desired. Also, the tone of this review seemed a quite a bit more Intel-appeasing, if you please, as there was nothing in this review that we didn't see before except for flowery praise about how Yonah does very well without an on-die mem controller.That said I must say that all the asking for a 2GHz 1MB L2/core A64 was pretty retarded. There is no 2GHz 1MB/core SKU so including a fake one just for comparison does not really help since there's no way to get something even similar (the 4400+ is the lowest-clocked 1MB L2 X2). It would have been nice to see an X2 4600+ (the second-best A64 SKU) compared to this Yonah (the second-best one) but I guess the 4200+ is more inline with its price.
SpinJaunt - Wednesday, December 21, 2005 - link
An overclocked Opteron 165 or underclocked Opteron 175 might have been an idea? forget about prices.
I think AMD still has some tricks up there sleeves regardless of what there roadmaps might say.
Anemone - Tuesday, December 20, 2005 - link
Eliminating clockspeed and using the best cache available sets the baseline for comparison, by keeping as few items of difference between the chips as possible. I fully agree with the choice, and, moreover, am quite positive there will be something out in the Turion line that will be quite similar to the 2ghz, dual channel, 1m/core cache that was used for testing.I think it's kind of funny to see us finally returning to tests where comparing close to exactly the same clockspeed produces even mildly comparable results. I say that because years ago that's what we used to do all the time, and finally things have come nearly full circle.
:)
Furen - Tuesday, December 20, 2005 - link
Ah, but we still don't know how DDR2 will affect K8 (or K9, as AMD likes to call the dual-cores) performance. Maybe AMD will increase the L2 cache data width from 128bits to 256bits (the Pentium M has a 256bit interface) to make up for the additional latency, though I doubt it.vijay333 - Monday, December 19, 2005 - link
"Now if that ain't fanboy, what is???"It's called giving good advice. Not an Intel owner myself, but even I would appreciate this info as AT obviously has more info on this. Would you rather buy a laptop now and then regret the purchase when something much better comes along in just a month from now? AT is not telling you to buy an Intel based machine, just to wait a month to get a better idea of what your options are. If you have read AT for a while, you should know that they are definitely not biased towards Intel...
tfranzese - Monday, December 19, 2005 - link
Something new and better will always be out "a month from now". Get use to it.bob661 - Monday, December 19, 2005 - link
Looks like just a P-M with two cores to me. Whoop-de-doo.Spoonbender - Monday, December 19, 2005 - link
Am I the only one recalling the old Athlon days?AMD smashes ahead with a great CPU, until around what, 2600+ or so, where they run out of steam, letting Intel overtake them.
Looks like the same might happen again...
Not sure if it's just AMD screwing up, or if it's really a question of resources.
For AMD, Athlon 64 was really a last-ditch gamble. They had to do something big, or they wouldn't exist 5 years from now.
Well, they did, and enjoyed a lot of success, but they might just not have the resources to follow up on it. Instead, their only option might be to milk the A64 for all it's worth, and then take a beating for a year or two, until they're ready with a next-gen architecture
LuxFestinus - Tuesday, December 20, 2005 - link
A question of availability needs to be raised. Are these new duo Intel processors available now? Can anyone say paper launch. How long has the dual Athlon64 been out? I thought so.Griswold - Monday, December 19, 2005 - link
I dont see where you're coming from. The A64 is a K7 on steroids. The P-M is a P3 on steroids and Conroe will also just be a P-M on steroids. Both companies didnt come up with flashy, brandnew architectures over the last few years, they only added flashy things to existing designs. Here and now, it's still Intel playing the catchup game until we see Conroe in stores. And then you always have to keep in mind that AMD is yet to move to 65nm. It will certainly give them some more clockspeed headroom.I agree with the conclusion of the article, we'll see a neck to neck race in a year from now where the better price will make the difference. AMD really doesnt have to flex its muscles now, they can milk the crowd with a superior product - and people will pay whatever to get it, or so it seems.
Calin - Tuesday, December 20, 2005 - link
Several of the flashy brand new architectures of the last few years had any kind of success. Itanium/EPIC, Transmeta, Netburst... As long as Transmeta was supported by a company with very little financial power (so their loss was somewhat expected), Itanium is crawling ahead on life support, and Netburst (while being king of the hill for a good time) will be discontinued.Griswold - Tuesday, December 20, 2005 - link
The point is: Netburst is certainly a design that could be considered newer or more radical compared to the P3->P-M->Yonah/Conroe/Merom route. In the end netburst was still a technical failure, but certainly not a financial mistake.Well and Epic, thats much older than you think. But yes, it never ended up where Intel wanted it to be: on the desktop.
allies - Monday, December 19, 2005 - link
It seems like with Intel's new chips, we're going to be back to a point where neither company has a true lead on the other. AMD is in great position at the moment, but they need to get their Turion X2 out to combat Intel's Centrino Duo. Otherwise, they'll find themselves losing laptop sales, an area which they've come a ways in.Right now, although clockspeed isn't increasing as fast as it once was, is a very exciting time for computer technologies. Parallelism, die shrinking, heat reduction, among other strides are paving the way to the future.
ncage - Monday, December 19, 2005 - link
Are you sure the improvements in games are due to the different Memory Controller technoliges? Are you sure its not the FPU?tfranzese - Monday, December 19, 2005 - link
I was thinking the same, considering his thoughts are pure speculation on his part. So, until Anand can provide data to back that up, he should stick to commending AMD's architecture rather than trying to credit all their success to the on-die memory controller (which surely helps, but it's only one part in the formula).I would doubt AMD is sitting idle though, and as they work on their next architecture will keep their lips sealed in order to maintain an element of surprise. Surprise like when they added SSE support to the Palomino and SSE3 support to Venice - both unexpected additions.
Furen - Monday, December 19, 2005 - link
The K8s execution core is pretty much the same as the K7s. If you compare performance (in games) between the two you'll notice that the K8 performs much better. This is due to the fact that it removes the FSB bottleneck (by integrating the memory controller), increases the width of the L2 cache (and the size) and dramatically drops the memory access latency. Sure there are other minor differences but they're mostly minor improvements, like better branch prediction, etc.tfranzese - Monday, December 19, 2005 - link
I don't see where the two of you are going with this and only serve my point that pinning the successes of the architecture on the memory controller is only speculation.Furen - Monday, December 19, 2005 - link
Well, the memory controller is the major difference between the K7 and the K8 but if you compare performance between the two the K8 performs much better. This means that the memory controller directly lead to this increase of performance.In truth the K8's performance is a combination of its micro-architecture and the low-latency access to the memory but since the microarchitecture came first (and was insanely bottlenecked by FSB at higher clocks) the one improvement that lead to the performance difference between the K7 and the K8 was the memory controller. In fact when AMD launched its K8 it said that there would be a 20-30% performance improvement because of the on-die memory controller.
tfranzese - Monday, December 19, 2005 - link
The K8 saw a new instruction set, a slightly lengthened pipeline, SSE2 extensions, SSE3 extensions (eventually), dual-core/multi-cpu design strategies, etc. Oh, and it got an on-die memory controller among other architectural tweaks.I don't think it's valid to attribute so many factors that could have benefitted the architecture to just the memory controller. A lot of small differences add up to a lot.
Furen - Monday, December 19, 2005 - link
Longer pipelines lead to lower performance, the "dual-core design strategies" have nothing to do with a single-core K8's performance benefits over a K7, SSE3 is useless even now and, of course, AMD64 does not benefit 32-bit execution. The only thing that you mentioned that makes a difference is SSE2 and it doesn't really make as much of a difference on A64s as it does on P4s since SIMD vector instructions require multiple passes on the A64. The deeper buffers help, as do the increased L2 cache bandwidth and the increase in L2, but the biggest benefit does come from the integrated memory controller. Cutting access latency is insanely important but having a faster frontside bus (the bus that connects the execution core/cache to the memory controller) is probably what makes A64s perform how they perform.fitten - Tuesday, December 20, 2005 - link
This is not always the case. On branchy code, it is typically true. On non-branchy code, longer pipelines can be very efficient. The problem is that typical codes on the x86 are very branchy so longer pipelines aren't that good on typical x86 codes.
As far as latency numbers and the like, you should do the math to understand why the latency helps. For large cache sizes (512M and larger), the L2 should get above 96% hit rate typically. For 1M L2, hit rates should be 98% or more. Obviously, the application you have will govern these hit rates but this is for "typical" codes. Some applications will see almost no benefit from having an L2 cache at all, for example. The latency of the main memory accesses are felt in the misses (that other 4% or 2%). If the L1 is pretty good (1 cycle penalty), you can zero that out for the calculation. Use some numbers on L2 and main memory access times to get an idea of how it really helps.
So many people just chant "integrated memory controller" as some kind of mantra without even knowing how much it *really* effects memory access times.
Furen - Tuesday, December 20, 2005 - link
Longer pipelines do not help non-branchy code, higher clock speeds do. Longer pipelines allow you to raise clock speeds but if you compare two equally clocked CPUs with similar architectures but different pipeline lenghts then the longer-pipelined one will ALWAYS be slower, since both will eventually mispredict a branch and the penalty on the longer-pipelined one will take a greater hit. In the case of the K8 compared to the K7, however, the branch predictor was improved, the TLBs increased and so on, so you probably end up having the same performance per clock."Typical" code is code that operates on very small data sets, like a word processor. This is not what I'm talking about, however, I'm referring to code that handles massive data sets that cannot fit inside the L2 cache. These include games and streaming media. A K7 performs pretty much the same as a K8 (clock for clock) in office applications and the like, but once you have data traveling down the frontside bus (the K8s frontside bus equivalent is the link between the execution core and the memory controller, which runs at CPU clock) then the performance differences are massive. It may be true that most of the code we execute on a PC does not even touch the main memory to a significant degree but it is also true that we perceive the times when it does as a massive drop in performance. Saying that memory bandwidth (and latency, as the two are directly related) is useless is like saying that a P3 is enough for everyone.
fitten - Wednesday, December 21, 2005 - link
Yes... and longer pipelines is one of the design parameters to achieve higher clock speeds.
Yes, which is why I used "typical" there with a caveat that some workflows do not match that pattern. The math that I mentioned is not difficult to do and the percentages for hit/miss are simply parameters into the equation. You can take any instruction mix and data access pattern, analyze it, and plug the newly found percentages into the equation for a comparison. And... I never said that memory bandwidth is useless. However, I would be inclined into discussion about your bandwidth and latency being directly related (in the general form). Quite obviously, satellite communication has high bandwidth and it is equally obvious that satellite communication has a very high latency, for example.
So, your post confirms exactly what I have said and that AnandTech's benchmarks show (and the conclusions stated in the article). For the majority of applications, since data locality is high, the IMC doesn't do all that much (simply because it isn't used that much). For applications such as games and other applications with data access patterns that do not have a high degree of data locality, the IMC starts to shine. I would also argue that streaming does not fall into that category unless you have poorly optimized code. Intelligent use of prefetching, for example, can hide most of the latency penalties of main memory. I guess we could discuss what "majority of things" means and whether or not games fall into that category. ;)
Furen - Thursday, December 22, 2005 - link
[quote] Yes... and longer pipelines is one of the design parameters to achieve higher clock speeds. [/quote]That's exactly what i said in the line that followed what you quoted. When I said that longer pipelines themselves dont help performance I meant that the clock-for-clock performance benefits of the K8 over the K7 can be mostly attributed to its on-die memory controller. Of course the bigger caches help, as do SSE2 and the other improvements, but the lion's share of the improvement comes from the integrated northbridge (the FSB was a horrible choke point in the K7).
[quote] I would be inclined into discussion about your bandwidth and latency being directly related (in the general form). Quite obviously, satellite communication has high bandwidth and it is equally obvious that satellite communication has a very high latency, for example. [/quote]
Sorry, let me clarify that a bit. When dealing with DRAM (at a set frequency) in a computer system the usable memory bandwidth is directly related the latency. It is not directly proportional but a higher latency will mean a lower usable bandwidth. This is because the memory subsystem in a PC is not just a data transport mechanism but also functions as a data storage array, which gives latency more importance (Satellite communication, on the other hand, only moves data from point to point, it does not store it or modify it in any way, it's just a conduit, which makes its bandwidth somewhat independent of the latency). Now, remember that I'm talking about usable memory bandwidth, not peak bandwidth (which is what manufacturers love to quote). Peak bandwidth is pretty much unrealizable when doing anything useful.
Anyway, I agree with you on the caches, I wanted to point out that the pipeline length itself provides no performance improvements whatsoever, and wanted to say that an integrated memory controller is a wonderful thing. Now, I say that an IMC is wonderful but it does have huge drawbacks, the main one being what AMD is currently dealing with, having to change sockets in order to update memory technology. The thing is, Intel needs flexibility because it is always updating to the newest technologies out there but AMD, on the other hand, actually gained control over the part of the traditional northbridge that affects performance the most without having to go all out and design its own chipsets like Intel does, which is why pretty much all AMD chipsets perform very similarly.
Furen - Thursday, December 22, 2005 - link
Now, can someone tell me how to make decent looking quotes?!!Xenoterranos - Tuesday, December 20, 2005 - link
Considering that bus was developed with dual core and multi-cpu design in thought, I'd say that "dual-core design strategies" had a lot to do with the increase in performance of K8 oer K7. AMD's technical director said something to that effect in so many words in an interview here a few years back; he said they'd built K8 from the ground up for dual core, multi-cpu applications.blackbrrd - Monday, December 19, 2005 - link
Actually, last time I checked, AMDs FPU (since K7) has had 3 execution units, while Intels has had 2 execution units (since pentium..2, or the original pentium, can't remember).Furen - Monday, December 19, 2005 - link
The P6 has two FP units: An FADD unit and an FMUL unit. One of the big weaknesses of the P6 is the fact that the FMUL unit is not fully-pipelined but instead uses part of the FADD unit for FMUL operations. The K7, on the other hand, has three fully-pipelined units, an FADD, an FMUL and an FSTORE.tayhimself - Monday, December 19, 2005 - link
No he's not. AT's hardware reviewers are nublets. One thing to note though, Dothan FPU is better than the P4's hence its gaming performance advantage over the P4 in the old tests that everyone saw. It's likely that Yonah FPU is still the same as Dothan (similar to P3) and inferior to AMD's.saratoga - Monday, December 19, 2005 - link
Actually the FPU on the P4 was tremendously more powerful then Dothan or Yonah. While games do use the FPU, they're not that bottlenecked by it on modern systems. The reason Dothan did so well was because of its large, very low latency L2 cache. This is roughly equivilent to the primary advantage of the K8, a very low latency memory controller.tayhimself - Monday, December 19, 2005 - link
Youre right about the low latency L1/L2 caches on the Dothan, but the P4 (Williamette/Northwood) has those as well. But the P4 FPU is only powerful in SSE2 mode where it can load store larger chunks of data. Not all games use that unfortunately.saratoga - Monday, December 19, 2005 - link
You're wrong on several points.First:
Dothan L2 latency (clks): 10 clks
Northwood L2 latency (clks): 18 clks (approximately)
So Dothan's L2 cache is roughly 2x as fast and 4x as large. If you compare prescott with its amazingly slow L2, the situation is even more biased towards Dothan. Clearly, in terms of cache performance Dothan has a massive advantage, at least once you're out of the L1.
Second, you're confusing SSE2 and vector processing. While SSE2 can perform vector ops, it also handles plan scaler as well. In x86-64 SSE actually replaces the traditional x87 unit. The relative performance of the two is irrelevent however, the P4 was faster in both.
coldpower27 - Monday, December 19, 2005 - link
Dothan I agree with as having 10 Cycle Cache.
Northwood has 16 Cycle Cache.
Well you also got to keep in mind northwoods clock frequency plays a role in speeding up the cache, accces latencies for Dothan @ 2.0GHZ vs Northwood @ 3.2GHZ are basically equivalent. Though the 2.26GHZ Dothan has the fastest cache of all.
AlexWade - Monday, December 19, 2005 - link
Although, "Core Duo" is a stupid name. Why does Intel have to be so different? "Core Duo" is a little confusing. Is Duo a code name? What?However, despite the stupid name, we've really turned a corner in performance. Intel can make a good CPU when they realized speed isn't the future. Looks like I should start considering replacing my old Pentium-M IBM T40p with the awesome battery life.
AMD needs to respond in kind with a great new CPU. The future looks bright. Competition is once again is good for everyone.
LuxFestinus - Tuesday, December 20, 2005 - link
The ambiguously gay duo, with Ace and Gary.:) An old SNL skit.ksherman - Monday, December 19, 2005 - link
Personally, i dont like that AMD is just sitting back, seemingly waiting for Intel to catch up... They need to kick Intel while their down. these new Processors from intel look really nice and i am likely to buy one, but in a mactel laptop. I am happy for INtel that they are catching up, but AMD really NEEDS to step up and do soemthing new.Calin - Tuesday, December 20, 2005 - link
AMD's income is lower than Intel uses for R&D. You really can't expect from AMD to develop something faster than Intel can.For AMD, to have an processor they could improve a step at a time since the introduction of the Athlon64/Opteron was a need - Intel is able to mantain several teams for microprocessor development, but AMD only has money for one. And AMD will milk the market for as much as possible, selling processors that are easy to make for prices that market will accept. If AMD will start selling a higher processor grade, they would need to reduce the price for lower speed processors. This is why the 2800+ and 3000+ are discontinued - they would have to sell them too cheap.
ncage - Monday, December 19, 2005 - link
I swear this is the best first post ive ever seen. Good Post Alex. Ya competition is very good for the market. I think intel is starting to get back on track where they need to be. It all comes down to clock speed and cost at launch. What improvements will we see with the launch of amd's next chip other than ddr2 which right now i don't really care about and possibly more cores (at least for the opteron). I am not dogging amd because for about 3-4 years ive only used AMD chips but i think amd has to raise the bar even more.Calin - Tuesday, December 20, 2005 - link
DDR2 for AMD would be great at least for a few things:moving to single channel DDR2 memory would decrease costs (in mainboards and a bit in processors)
moving integrated graphics to single or dual channel DDR2 would increase graphic performance and overall system performance in relation to single or dual channel DDR
As for the high end, I really don't think an increase in memory bandwidth will help - not even for dual core processors. Maybe for a quad core, but quad cores are certainly for servers, and I don't know about registered DDR2 memory to be used in them.
Hmmm, you could try an Opteron Dual Core with single channel DDR memory, to see how much performance would be lost by going quad core, dual DDR.
mlittl3 - Monday, December 19, 2005 - link
I agree with both Alex and ncage. I really disliked Intel all through out the Pentium4/Net-burst days. They were just releasing marchitecture with no improvements whatsoever. I loved AMD for their innovation and performance/watt.Now both companies are equal but I don't think we will see the huge fall AMD suffered from when Intel released the Pentium 4 to compete with the Athlon/K7 architecture. The beauty of competition is showing its bright colors right now. If we only had Intel, we would have a very hot/power consuming inefficient Pentium 4 based on net-burst to play Quake 4 at 5fps right now.
Its time for the fanboys to turnover a new leaf. Go Intel and AMD!!! We love both you guys.