It's been such a long time since we've had as exciting a product as Intel's 925X platform arrive in our labs. The platform brings about a new CPU interface (LGA-775), a new graphics interface (PCI Express x16), a new memory interface (DDR2) and a slew of other tweaks that make for an impressive bundle of technology. Unfortunately for Intel, we are much more than just technologists - we are pragmatic technologists.

We have already shown how Intel's 925X and 915 platforms basically offer no performance increase over current generation 875P/865 platforms. The lack of performance improvement can essentially be attributed to the high latency of current DDR2 memory, combined with the lack of bandwidth utilization of DDR2-533. These two problems can and will be addressed in the future by lower latency DDR2 memory as well as Prescott's forthcoming 1066MHz FSB (which will be very well matched to a DDR2-533 memory bus). Once again, unfortunately for Intel, we are talking about present day performance and the situation isn't as perfect as it would be had we been given both of those things.

So, the launch of the 925X and 915 has come and gone, with very little excitement from the community in regards to platform performance - but are there any other diamonds in the rough to be discovered?

Alongside the LGA-775 socket interface, Intel gave Prescott a bit of a speed bump - taking it up to 3.6GHz, making it the highest clocked Intel processor available today. This article will be taking a look at the extra 200MHz and how it changes, if at all, the Prescott factor.

Then, we have this issue of PCI Express graphics; Intel has pretty much guaranteed a fast transition to PCI Express graphics cards by removing any AGP support from their 925X/915 chipsets. Intel is expecting that half of all Intel platforms will be 9xx based by the end of 2004, meaning that 50% of all of Intel's platforms shipped by the end of 2004 will not have AGP support. Like it or not, PCI Express as a graphics bus is here.

But what about performance? Both ATI and NVIDIA have been duking it out over the past several months about whose PCI Express solution is the best. And now, we're finally able to find out. Toning down the suspense a bit, you'll find that the whole PCI Express debate was really much ado about nothing, but we'll have some more explanation and benchmarks showing that in the coming pages.

With the above paragraphs, we've pretty much summed up what you can expect out of this article, but wait, there's more (cue TV salesman). This week, we will also take an in-depth look at one other feature offered by the platform and investigate the real world performance benefits of Native Command Queuing. NCQ is supported in the new ICH6 South Bridge and is claimed to improve performance significantly; we'll see what that means in the real world soon enough.

With that said and done, let's get to it.

LGA-775: Do we really need it?

The hot topic at Computex this year was Intel's new 775-pin LGA socket. Motherboard and memory manufacturers alike were complaining left and right about reliability issues with Intel's new socket, and with the launch less than a month away, we were obviously concerned as well.

First, why even bother with a new socket/interface? The Pentium 4 was doing just fine on Socket-478, and now with LGA-775, we're able to get another 200MHz? Seems silly, no?

While it's true that LGA-775 isn't necessary today, there are a couple of factors that dictate its necessity for tomorrow. As we've talked about before, a CPU's package can actually be a limiting factor when it comes to core clock speed as well as FSB frequency. Think about it this way; today's Pentium 4s run at a 800MHz FSB, meaning that the interface between the CPU and the chipset has to be capable of handling 800MHz signaling across a wide 64-bit bus. When we talk about the interface between the CPU and the chipset, physically, what are we talking about? We are talking about the traces on the motherboard going from the chipset to the CPU socket, the CPU socket itself and the pins on the CPU. Improvements in the pin interface are necessary in order to allow for higher FSB CPUs. While today's 800MHz FSB isn't really pushing the limits, remember that Prescott's replacement will be paired up with a 1.2GHz FSB.

The LGA-775 interface also packs the pins closer together, allowing 775 pins to exist in about the same space that 478 pins did on the old package. More tightly packed pins mean that we're dealing with shorter routes from the CPU's core to the external interface itself, also a desirable trait.

With more pins, we also have the ability to deliver more power, more reliably than before - which will also come in handy as Prescott and its successor ramp up to 5GHz. Intel banked quite heavily on clock speed with their NetBurst architecture, and part of the commitment was a commitment to new packaging technologies.

The other change that LGA-775 makes, and clearly the most controversial one, was to move the pins off of the CPU itself and onto the motherboard - meaning that the CPU itself is mostly flat underneath. The capacitors on the bottom of the CPU keep it from becoming perfectly flat, but the lack of pins definitely make it a flatter chip.

With no pins on the CPU, the first thing that should pop into your head is - "woohoo, no bent pins on your CPU." Unfortunately, with the pins on the motherboard, now you can have bent pins on your motherboard - and with the LGA-755 interface, they are much easier to bend, and much more difficult to unbend.

The problem with the LGA-775 socket is that you can't see the pins in the socket very well when you've got the CPU suspended in air above it. You have to be very gentle and very precise when placing the CPU on top of the pins because if you drop it on the pins or if you are just one pin off, you risk bending a lot of pins.

Click to enlarge.

Although we haven't tried it yet, Intel has told us that once you bend a number of the pins on the socket itself, in essence, it's pretty much impossible to unbend them. Remember that we're talking about 775 very fragile pins in that socket - and we thought Socket-478 was bad.

Click to enlarge.

There was another significant change in the socket structure with LGA-775: the locking lever. It used to be that the heatsink, not the socket's lever, was what provided the majority of force on the CPU itself to ensure secure installation in the socket (to prevent against package pullout and to guard against shearing the mechanical attach between socket and motherboard as well as the chipset and motherboard). Unfortunately, that meant that the heatsink had to supply upwards of 40 lbs of force to the CPU, which caused a lot of issues, the biggest of which was bending motherboards.

Click to enlarge.

While Intel insists that the amount of bending caused by their heatsinks on Socket-478 motherboards was within tolerances, the fact of the matter is that with enough heatsink installations and removals, you could do some serious trace damage to your motherboard, thanks to the bending of the board.

With LGA-775, Intel has addressed the pressure problem and now it's the socket lever and not the heatsink that supplies the ~40 lbs of force to not only secure the chip but also make adequate contact between the CPU and the pins. The heatsink now has to provide much less force and thus, doesn't bend the motherboard as much.

The old heatsink used to bend the motherboard quite a bit...

...but the new heatsink doesn't do so, not as much at least.

The socket's lever requires more force to engage, but it also puts much more force on the CPU than in the old socket.

Intel's new heatsink is actually a huge improvement over their old ones, thanks to the removal of heavy-duty downforce as one of its job requirements.

The four pegs of the heatsink basically plug into the holes in the motherboard, and you twist the tops of the pegs to have it locked. You have to make certain to twist the pegs securely; otherwise, the heatsink has a tendency to pop off. Luckily, Intel's thermal throttling prevents any serious damage from occurring as a result.

The bottom line is that a new CPU socket was necessary to continue with Intel's future CPU plans, but the fragility of LGA-775 definitely shows that Intel didn't take enthusiast users into consideration. Intel has neglected the enthusiast community in the past, and although we do make up a very small portion of the overall computing community, it's still an important group on which to focus. If you buy an LGA-775 board, just be very careful. We haven't had any problems yet, but we've been extremely careful with our installations.

It looks like LGA-775 will take us to the end of the Pentium 4 line, although we would have liked a more user-friendly socket to keep us company over the next couple of years.

Intel's Pentium 4 560 - The Model Numbers Continue...

Intel's LGA-775 Prescotts are architecturally the same as their Socket-478 Prescotts, so we'll direct you back to our Prescott article for more information on exactly what it means to be a Prescott. The only differences you'll find between LGA-775 Prescotts and their Socket-478 counterparts are that you can get a 3.6GHz LGA-775 CPU, whereas the fastest Socket-478 chip is still 3.4GHz, and all LGA-775 CPUs use Intel's new model numbering scheme.

As we've reported before, Intel's model numbering system basically uses arbitrary numbers to represent various CPUs. The numbers don't necessarily mean higher clock speeds; they just denote faster CPUs within a family.

All of the Prescott based Pentium 4s fall into the 5xx series:

- Intel Pentium 4 560 (3.6GHz)
- Intel Pentium 4 550 (3.4GHz)
- Intel Pentium 4 540 (3.2GHz)
- Intel Pentium 4 530 (3.0GHz)
- Intel Pentium 4 520 (2.8GHz)

Intel has been understandably quiet about their new model numbering scheme. After all, they were the ones who were so openly critical of AMD's model numbering system upon its release. Intel forced AMD down the road of model numbers, and it looks like they have actually painted themselves into a corner with requiring the use of model numbers as well.

AMD has made some mistakes with their model numbers in the past, and it will be interesting to see how Intel handles some of the same challenges that AMD has faced. For starters, by completely disconnecting the model numbers from clock speeds, Intel has avoided the issue of applying conservative or liberal ratings to processors. At the same time, you have to give credit where credit is due, and we must say that Intel's modeling system is strangely reminiscent of AMD's numbering systems.

First 64-bit x86 extensions and now model numbers, Intel has been enjoying the taste of shoe for a while now, it seems.

The Test - Intel's Pentium 4 560 (3.6GHz) vs. the World

In our internal comparison tests, we found that Intel's 925X motherboard performed within 0.5% of their 875P motherboard, meaning that we could compare Pentium 4 560 scores on the new 925X platform to our most recent CPU scores and get comparable results. So, we did just that.

We omitted gaming performance for now and we will focus on PCI Express gaming performance later on in this article, including a comparison to an Athlon 64 3400+.

Performance Test Configuration
Processor(s):	Intel 560 (3.6GHz) Socket 775 Intel 3.2GHz Northwood Socket 478
RAM:	2 x 512MB Micron DDR2 533 2 x 512MB Corsair 3200XL (Samsung 2-2-2-5)
Hard Drive(s):	Seagate Barracuda 7200.7
Video AGP & IDE Chipset Drivers:	Intel Chipset Driver 6.0.0.1014 Intel Application Accelerator 4.0.0.6211
Video Card(s):	nVidia GeForce 6800 Ultra PCI Express nVidia GeForce 6800 Ultra AGP 8X ATI Radeon X800 XT PCI Express ATI Radeon X800 XT AGP 8X
Video Drivers:	nVidia 61.45 Graphics Drivers ATI Catalyst 4.6 beta
Operating System(s):	Windows XP Professional SP1
Power Supply:	HiPro 470W (Intel) Vantec Stealth 470W Aluminum
Motherboards:	Intel 925XCV (Intel 925X) Socket 775 Intel D875PBZ (Intel 875P) Socket 478

General Application Usage Performance (Winstone)

We have two LGA-775 CPUs in our performance tests, as you can see below: the LGA-775 Pentium 4 3.4GHz Extreme Edition and the new Pentium 4 560 (3.6GHz). Looking at the comparison between the LGA-775 3.4EE and the Socket-478 3.4EE, there is basically no performance difference, which is to be expected.

What's interesting is that the new Pentium 4 560 (3.6GHz) manages to outperform even the Extreme Edition, making it Intel's fastest CPU. Although Intel comes in behind AMD in Content Creation Winstone 2004, these scores are actually based on the original benchmark and not the latest patched version. For whatever reason, the original benchmark disabled multi-threaded rendering in Lightwave, causing Hyper Threading enabled processors to be held back in performance.

Veritest has since addressed the issue and we will be updating our scores to reflect the new patch; until then, this benchmark serves more as a comparison within processor families and not necessarily as an AMD vs. Intel benchmark.


In Business Winstone 2004, we see that the higher clock speed of the Pentium 4 560 (3.6GHz) is not enough to overcome the 2MB L3 cache of the 3.4GHz Extreme Edition. The performance here isn't very spectacular for Intel to begin with; if you're a business user, your choice is clearly AMD.

General Application Usage Performance (SYSMark)

Using SYSMark's overall score, we see that the new 3.6GHz Pentium 4 basically provides the same performance as the 3.4EE and the Athlon 64 FX-53. Also note that there's no performance difference between the 925X LGA-775 platform and the older 875P platform under SYSMark 2004. Features have been improved, but performance has not.


Once again, the new Pentium 4 560 takes the crown away from the 3.4EE to come out on top in Internet Content Creation performance.


SYSMark's Office Productivity tests are much more evenly split than the Business Winstone scores, with the 3.4EE, Pentium 4 560 and Athlon 64 FX-53 all essentially performing the same at the top of the charts. In fact, you don't really see any difference in performance until you drop down to the 3GHz Pentium 4 and the Athlon 64 FX-51.

DivX Encoding Performance

Raw clock speed grants the new Pentium 4 560 the DivX Encoding crown, taking the lead away from the 3.4EE. Applications such as media encoding don't exactly benefit much from cache locality and thus, rarely benefit from increases in cache size or speed. Instead, we see that overall system bandwidth (FSB/memory) and CPU clock speed are much larger determinants of performance.

DivX encoding is one of the few remaining performance advantages that Intel holds over AMD.

3D Rendering Performance

Interestingly enough, although 3D rendering is not normally considered to be a cache size-dependant task, the 3.4EE cannot be touched by the new Pentium 4 560. Performance from the Intel platforms continues to be strong here as well.


The performance under Lightwave is similar to what we've seen under 3D Studio Max, with the Extreme Edition CPUs taking the crown.

Visual Studio Compile Time

The added clock speed still can't beat the added cache of the 3.4EE, so the Pentium 4 560 doesn't actually grant Intel any headway over AMD here. It looks like development workstations are still best when paired with an Athlon 64, especially considering that multi-threaded compiling isn't usually what's done on these workstations.

PCI Express Graphics

When the first 440LX motherboards hit the streets with AGP support, it was so exciting to finally have a new slot on motherboards that had been littered with PCI and ISA slots for so long. The excitement of having that new slot is once again duplicated with the new PCI Express x16 slots that have found their way onto 925X and 915 boards. So, what is the big deal behind PCI Express as a graphics bus?

Click to enlarge.

For starters, AGP/PCI are parallel and PCI Express is serial. What this means is that rather than sending multiple bits at a time, PCI Express only sends one bit per clock in each direction. Mildly confusing is the fact that multiple PCI Express lanes can be connected to one device (giving us PCI Express x4, x8, x16 and so on). Why is an array of serial interfaces different than a parallel interface? We're glad that you asked.

Signaling is generally more difficult using a parallel communication protocol. One of the problems is making sure that all the data being sent in parallel makes it to its destination in a timely fashion (along with all the signaling and control flow data that are included with each block of data sent). This makes circuit board layout a little tricky sometimes, and forces us to keep signaling over cables to relatively short distances using equal length lines (e.g. an IDE cable). The fact that so much care needs to be taken about getting all the bits to their destination intact and together also limits signaling speed. Standard 32bit PCI speed is 33MHz. DDR memory is connected to the rest of the system in parallel and runs at a few hundred MHz. On the other hand, one PCI Express lane is designed to scale well beyond 2GHz.

The downside of this enhanced speed and bandwidth is bus utilization. Obviously, if we are sending data serially, we are only sending one bit every clock cycle. This is 32 times less data per clock cycle than the current PCI bus. Add on to that, the fact that all low level signaling and control information need to come over the same single line (well, PCIe actually uses a differential signal - two lines for one bit - but who's counting). On top of that, serial communications don't really react well to long strings of ones or long strings of zeros, so extra signaling overhead is implemented to handle those situations better. Parallel signaling has its share of problems, but a serial bus will always have lower utilization. Even in cases where a serial bus has a bandwidth advantage over a parallel bus, latency may still be higher with the serial bus.

Fortunately, PCI Express is a very nice improvement over the current PCI bus. It's point to point, so we don't need to deal with bus arbitration; its serial, so it will be easy to route on a motherboard (with just four data wires for PCI Express x1) and will scale up in speed more easily. It's also backwards compatible with PCI from the software's perspective (which means developers will have an easier time porting their software).

Unfortunately, it will be harder for users to "feel" the advantages of PCI Express over PCI, especially while the transition is going on and motherboards will be supporting "legacy" PCI slots and busses, and companies will have to find the sweet spot between their PCI and PCI Express (or AGP and PCI Express) based cards. Software won't immediately take advantage of the added bandwidth because it is common practice (and simply common sense) to develop for the widest audience and highest level of compatibility when dealing with any type of computing.

Even after game developers make the most of PCI Express x16 in its current form, end users won't see that much benefit - there's a reason that high end GPUs have huge numbers of internal registers and a 35GB/s connection to hundreds of megs of local GDDR3. By the time games come out that would even think of using 4GB/s up from and 4GB/s down to main memory, we'll have even more massive amounts of still faster RAM strapped on to graphics boards. The bottom line is that the real benefit will present itself to applications that require communication with the rest of the system, like video streaming and editing, or offloading some other type of work from the CPU onto the graphics card.

PCI Express Graphics Cards

Both ATI and NVIDIA supplied cards for Intel's 925X/915 launch; NVIDIA provided NV45s (PCI Express GeForce 6800 Ultras) while ATI provided R423s (PCI Express Radeon X800 XTs).

There is a major difference between both ATI's and NVIDIA's approach to PCI Express graphics that has been the debate between the two companies for the past several months. ATI built brand new GPUs with a PCI Express x16 interface in the core to support the new standard, meaning that for every GPU that they want to have a PCI Express version of, ATI has to have a separate GPU design. Granted that the differences between a R420 (AGP 8X X800) and R423 (PCI Express X800) on a design level are minimal, they are still two separate chips. ATI's PCI Express strategy, basically, increases dramatically the number of GPUs that they have to manufacture, but it also guarantees the highest possible performance.

NVIDIA, on the other hand, takes a more economical approach from a manufacturing standpoint and will continue to only make AGP 8X compliant GPUs. In order to satisfy the PCI Express market, NVIDIA will outfit their PCI Express cards with an AGP-to-PCI Express bridge chip that NVIDIA calls their High Speed Interconnect (HSI). The bridge chip means that NVIDIA can still produce the same number of GPUs; just toss on a bridge chip whenever those GPUs need to be put on PCI Express cards.

NVIDIA will eventually offer native PCI Express GPUs, at which point, they will continue to use the HSI as necessary to bridge back to AGP for older platforms. Considering that it will take years for AGP to go away completely, this solution isn't a bad one at all.

The obvious downside to NVIDIA's approach is that the additional bridge could offer a performance penalty, but the question is how much? That is what we're going to answer now...

The Test - AGP vs. PCI Express Graphics Cards

The coming pages offer two performance comparisons; first, a comparison of NVIDIA's AGP 8X and PCI Express x16 solutions as well as a comparison of ATI's AGP 8X and PCI Express x16 solutions. We ran the PCI Express x16 solutions on Intel's 925X board and the AGP 8X solutions on Intel's 875P board. Since the performance between the two is basically negligible, thanks to high latency DDR2 memory, we have ourselves a pretty decent AGP vs. PCI Express performance comparison. We also kept GPU clock and memory clock speeds identical across both platforms, so that we truly have a useful performance comparison.

We would have rather had a solution with both AGP and PCI Express interfaces, but it seems that the only chipset which will offer native AGP and PCI Express interfaces is from VIA and it won't be ready until next month at the earliest.

The other comparison that we'll make in the coming pages is between the Pentium 4 3.4EE and our usual gaming platform, the Athlon 64 3400+, just to have a reference point to compare to our older gaming benchmarks.

X2 PCI Express Performance

Looking at the AGP vs. PCI Express comparison, we see that both AGP platforms outperform their PCI Express counterparts by a small margin under X2. The performance drop is not related to whether or not a bridge chip is used, as both ATI and NVIDIA suffer, so it's probably a PCI Express driver issue.

Halo PCI Express Performance

Under Halo, the performance drop with PCI Express is basically nothing. There's no advantage or disadvantage to moving to PCI Express.

Wolfenstein ET PCI Express Performance

Under Wolfenstein, there is a very slight performance improvement, thanks to PCI Express, but at less than a 1% performance difference, we'll just call this one equal.

Jedi Knight: Jedi Academy PCI Express Performance

Once again, the performance difference between PCI Express and AGP is basically nothing here. Although, ATI does suffer a slightly larger performance drop than NVIDIA does under Jedi Knight.

Unreal Tournament 2004 PCI Express Performance

Under UT2K4, the two graphics interfaces are identical once again.

Far Cry PCI Express Performance

Far Cry is actually the only game that we've tested where PCI Express has a significant performance penalty when compared to AGP 8X. ATI's performance actually doesn't drop much at all, but NVIDIA's PCI Express loses a whopping 30%; we're hoping that this is a driver bug, which it most likely is, since it's the only situation where such a large performance hit is seen.

Final Fantasy XI PCI Express Performance

Under Final Fantasy, the performance picture goes back to normal, with no performance difference between PCI Express and AGP 8X.

Homeworld 2 PCI Express Performance

Here, we have a slight performance improvement, thanks to PCI Express on NVIDIA's card, and a miniscule drop on ATI's card - nothing real world.

F1 Challenge PCI Express Performance

Our F1 Challenge benchmark shows a slight advantage for the ATI PCI Express solution and a slight disadvantage for NVIDIA's, but when you look at the actual numbers, you'll see that there's once again no real world performance difference between the two interfaces.

Neverwinter Nights PCI Express Performance

The performance between AGP and PCI Express is, once again, basically nothing here under Neverwinter Nights.

EVE: The Second Genesis PCI Express Performance

In EVE we see, once again, that performance doesn't really change between AGP and PCI Express. Here we have a situation where NVIDIA takes a performance hit with PCI Express while ATI gains a little bit. In the real world, the performance differences continue to be negligible.

Warcraft III: The Frozen Throne PCI Express Performance

Warcraft III is interesting because without AA and AF enabled, we see that NVIDIA takes a reasonable performance hit with PCI Express, while ATI stays exactly the same between the two interfaces. The performance drop isn't nearly as pronounced as in Far Cry, but we have to wonder what's causing the smaller but still significant performance hit. We will be looking into it more...

Final Words

We split this article into two sections, basically focusing on the performance of the new Pentium 4 560 (3.6GHz) as well as the performance impact of using PCI Express graphics, so we will conclude the article accordingly.

The Pentium 4 560 manages to offer performance better than that of Intel's fastest Extreme Edition P4, thanks to the additional 200MHz clock increase. The performance enhancement has nothing to do with any architectural changes, as there are none, but everything to do with clock speed. If Intel wants to continue to uphold that their Extreme Edition brand is truly extreme, then a higher frequency core is necessary.

Intel's move to the LGA-775 socket leaves a somewhat foul taste in our mouth, although we do understand the engineering need for such a move. The fragility of the socket requires us to caution our loyal readers once again. Remember that it's quite easy to damage these pins, so don't rush your CPU installation. Intel quotes the mean time between failures of the new LGA-775 socket at around 20 insertions, which they claim is similar to the current Socket-478 interface; only time will tell how reliable these things really are.

With this article, we were also trying to put an end to the ATI vs. NVIDIA PCI Express debate. Our conclusion? The debate was much ado about nothing - both solutions basically perform the same. ATI's native PCI Express offering does nothing to benefit performance and NVIDIA's bridged solution does nothing to hamper performance. The poor showing of NVIDIA under Far Cry and Warcraft III is some cause for concern, which we will be looking into going forward. We will keep you all updated on any and all findings with regards to that issue as we come across them.

Be sure to read our chipset coverage of the new 925X and 915 platforms and stay tuned for more coverage later this week, including integrated graphics performance and an investigation of the impacts of Native Command Queuing support on disk performance.