Comments Locked

50 Comments

Back to Article

  • Drumsticks - Thursday, July 26, 2018 - link

    On one hand, I think Intel has given away their lead unnecessarily; it looks like Zen 2 (and by extension, EPYC 2/Rome) will lead Icelake capable systems by at 6 months or so.

    On the other hand, Intel, caught in their worst rut likely ever, is only going to be six months behind Zen 2 at the end of the day. Even if they're in a self made quandary, I think they'll be able to keep pace once they finally get ice lake out. Here's to (hopefully) years of solid competition, given that both companies probably have roadmaps for years to come.
  • Alistair - Thursday, July 26, 2018 - link

    Intel uses a lot of tricks to increase their profits. No one should actually be worried about whether or not they have a competitive product. They can add more hyper-threading. They can unlock all CPUs. They could solder, they could include better coolers (quieter fans and larger heatsinks).

    Think about the i5-8400. They could just re-release it with 400mhz higher clocks and maybe a slightly higher TDP at the same price and be super competitive (Intel is efficient up to 4.6ghz or so). It just sucks to be an Intel consumer...
  • Samus - Thursday, July 26, 2018 - link

    All I can say is one thing: i5-8350U

    Seriously, a 4C/8T 15W CPU that's faster than most of the first-gen Ryzen 3's and 5's running at 65W, and as fast or nearly as fast as their second-gen Ryzen 3's and 5's running at 35W.

    And per usual, Intel is just icing AMD in single-threaded performance (mostly due to the lack of app optimizations, not CPU architecture) but the fact still stands: Intel will always have higher performance per watt in mainstream mobile and desktop.

    That's the hard truth. That doesn't mean AMD isn't competitive. Intel still outrageously overprices their products (because they can) and AMD is in a better position than they have been for nearly 20 years.

    AMD needs to continue focusing on GPU's too; nVidia isn't sitting idle.
  • t.s - Friday, July 27, 2018 - link

    Where's the link to that 'hard truth'? Hard truth for one could be bullshit for everybody else, just saying.
  • close - Friday, July 27, 2018 - link

    The Ryzen 2500U (also 15W) is indeed slightly slower on the CPU (not GPU) side.

    But just remember that the statements "hard truth" and "as fast or nearly as fast as their second-gen Ryzen 3's and 5's running at 35W" is coming from the same person who says that "AMD is in a better position than they have been for nearly 20 years". I assume he was thinking of the Athlon XP/64 period 12-15 years ago. So a 50% exaggeration is totally acceptable for him.
  • CBeddoe - Friday, July 27, 2018 - link

    close, I'm with you.
    The 2500U actually beats the i5-8350u in lots of multi-threaded loads but looses to it by a little bit in single threaded applications. Intel's integrated graphics are obsolete compared to AMD's which is why Intel paired with AMD on Devil's Canyon. IF you can't beat em, join em to make a great product.
  • yeeeeman - Friday, July 27, 2018 - link

    i5 8350U is the same as Ryzen 5 2500U on notebooks. The fact that "Intel will always have higher performance per watt in mainstream mobile and desktop" is just your opinion. History has long shown that everything can happen in this business. Also, this 10nm process that Intel has better be good, because if not (and it probably won't be very good, since they patched it up to work) 7nm products from TSMC (read AMD Zen) will exceed Intel's unbeaten performance.
  • Santoval - Monday, July 30, 2018 - link

    Intel cannot have too much confidence in Ice Lake and its successor (Tiger Lake), otherwise they wouldn't have hired Jim Keller to design Ocean Cove (presumably their first truly new architecture since the Core architecture was released back in ~2006, the fundamental design of which is still employed).
    I have no idea how Ocean Cove CPUs will be branded, but they will apparently be the first post-Core Intel CPUs in roughly 15 years when they are released.
  • Alistair - Friday, July 27, 2018 - link

    Actually the Ryzen and Intel chip in mobile are pretty similar in speed. Also Ryzen has better integrated graphics.

    User benchmark has 4 times the submissions for the 2500u vs the 8350u.
  • The Chill Blueberry - Friday, July 27, 2018 - link

    > Nvidia isn't sitting idle.
    Haha, how long ago was Pascal launch?
  • PeachNCream - Friday, July 27, 2018 - link

    Volta is out already isn't it? Not in consumer products, but if I'm not mistaken, Volta is what's powering the Titan V (at admittedly crazy prices) so it isn't like nvidia is doing no work on the next generation chips.
  • Spunjji - Friday, July 27, 2018 - link

    That CPU isn't faster than a 65W CPU when it's drawing 15W - it's only comparable to desktop chips when it's able to turbo up to 45W with 35W sustained. The fact that it can (in some tests) beat out those Ryzen CPUs isn't dreadfully surprising.
  • 1_rick - Friday, July 27, 2018 - link

    The 8250U, which is only 100MHz base/200MHz turbo slower, is a nice chip, if you're not doing anything that requires sustained performance. I have an Acer with the 8250 and I love it, but start transcoding a video and it's going to drop fast to around 2.0-2.1Ghz and stay there.
  • jospoortvliet - Friday, July 27, 2018 - link

    Well the gap is probably both bigger and smaller than that. AMD is starting with EPIC before rolling out consumer products so Intel might be only months behind there of at all. But in the server market the gap could be a year or more of they are slow with the big dies...
  • caqde - Friday, July 27, 2018 - link

    Exactly and for some reason I have my doubts about when the Icelake based XCC/HCC chips will be released. I'd imaging 2020 around the time Threadripper 4 gets released. Just shorltly before EPYC Milan's successor? If this is the case Intel is in a world of hurt when it comes to the server/HEDT sector's of the computing market.
  • duploxxx - Friday, July 27, 2018 - link

    the issue is that no OEM will dare to bring an AMD based version in a HW generation (e.g. HP gen11 - gen12) when there is nothing from Intel ready. they will just delay the AMD based solution till there is a intel counterpart. THis due to mass production of there base hw designs and volume. So it will be a lot of back and forth between these OEM to figure out weather they will just go for the same generation with CPU update or bring a full new generation. Knowing the dev and $$$ push from Intel to these OEM I am sure they will put pressure to delay the AMD as much as possible.

    "Cascade Lake-SP platform will bring support for 3D XPoint-based Optane DC Persistent Memory DIMMs". This must be marketing talk right? HP for example has already persistant memory on current Intel. It can only be used for very specific cases. Very limited use and not worth the investment because it also kills of a lot of possible normal memory slots. Intel will need to have a huge memory controller redesign to get it more wide open to choose. ANd way more dimms. THye are behind EPYC already.
  • Alex_Haddock - Friday, July 27, 2018 - link

    Though AMD has publicly stated Rome will be socket compatible with Naples so a Gen11 wouldn’t be needed. Naturally as an HPE employee I’m not publicly stating our roadmap ( :-p ) but available evidence.

    There is definitely a lot of interest in persistent memory use cases so I wouldn’t discount Optane; but there are also other routes to persistent memory or making DIMMs themselves effectively persistent. Regardless of technology we are definitely moving to a memory centric computing model so it’s all getting interesting again. Cheers, Alex
  • edzieba - Friday, July 27, 2018 - link

    Don't count your Zeppelins before they're fabbed. Intel are having problems with 10nm, but it seems foolish to expect that creating 7nm (very similar actual feature sizes barring the finer Cobalt layers at 10nm) large dies will be plain sailing for TSMC, GloFo or Samsung.
  • Cooe - Friday, July 27, 2018 - link

    Except Zeppelin isn't a large die... Not even close. (Vega 7nm, also samples this year otoh... THAT'S a large die [but even then, still much smaller than most large GPU's ala Fiji, P100, or God forbid, V100]). And a theoretical 16c/18t successor on 7nm would be of extremely similar size.

    Unless Intel suddenly goes modular, your argument is kinda nonsense.
  • edzieba - Friday, July 27, 2018 - link

    Zeppelin is a large die compared to the SoCs that are currently being fabbed on that process (and not even shipping yet). It is 212mm^2, where even Apple's gargantuan-for-an-SoC A12 is 88mm^2.
  • Spunjji - Friday, July 27, 2018 - link

    88mm^2 is larger than the 70mm^2 Cannon Lake 10nm die that Intel can't manufacture profitably in any reasonable volume. At this stage I'd expect to hear rumours if the other fabs were having similar issues, but instead we have roadmaps being stuck to and product releases being planned. This just does not look good for Intel in the near term.
  • .vodka - Friday, July 27, 2018 - link

    Let's not forget that 88mm² Cannonlake die in the form of the i3 8121u has the iGPU completely disabled, has worse clocks than 14nm++ parts (as expected) and its 15w TDP rating is horrible for a plain dual core. There are no reviews of that Lenovo notebook using this chip, and nobody really knows its performance or if there's an IPC improvement over Skylake. It's all smoke and mirrors.

    I'll be gentle and say that 10nm is a train wreck and a dumpster fire. Intel is in for a lot of pain next year unless there's a breakthrough on that money black hole of a process.
  • Santoval - Monday, July 30, 2018 - link

    It makes zero sense to compare x86 desktop and laptop CPUs with mobile ARM based CPUs. It is an apples to watermelons comparison. Compare instead x86 die sizes with the same number of cores, e.g. 8-core dies from Intel with 8-core dies from AMD.
  • PeachNCream - Thursday, July 26, 2018 - link

    Error...

    "...its products made using one of its 14nm process technology."

    The last word should be "technologies" in that sentence.
  • Flunk - Thursday, July 26, 2018 - link

    Sweet, only a whole year later than we were expecting at the beginning of 2018!
  • vFunct - Thursday, July 26, 2018 - link

    When can we expect updated 10nm MacBook Pro models? November/December 2019?
  • jospoortvliet - Friday, July 27, 2018 - link

    Given how long it historically takes Apple I would be extremely surprised if it would be before mid 2020.
  • danwat1234 - Friday, July 27, 2018 - link

    Since Apple is usually 6 months behind regular laptops... Mid 2020?
  • KAlmquist - Thursday, July 26, 2018 - link

    Intel 10nm is almost four years behind it's original target of the second half of 2015, unless you want to count the very limited 10nm production that Intel is currently doing. That's a really big failure. Intel used to have something on the order of a three year lead over most of the other semiconductor manufacturers, and they've blown it.
  • jjj - Friday, July 27, 2018 - link

    Don't forget that systems in stores is very vague, could be all segments or a few 5W systems.
    As for high volume production at 10nm, just the fab cycle is gonna be 3-4 months and takes a while to ramp output and get inventory out there so they can't blanket the entire market out of the sudden

    XPoint does not ramp to any real volumes before 2020 as second gen arrives in second half 2019 and needs to ramp. They can talk all they want but they got little output, no demand for even that output and they are losing money with it.Ofc that's before even considering how viable those DIMMs might be.

    Server is under pressure from AMD as they are finally starting to ramp Epyc, already sampling 7nm and by the time Intel goes 10nm, Epyc will be on N7+.
  • danwat1234 - Friday, July 27, 2018 - link

    So, basically 2020 and mostly efficiency gains. Would take some luck to get more than 10% more IPC than Intel 6th gen Skylake.. AMD and Intel IPC will be very close.
  • ikjadoon - Friday, July 27, 2018 - link

    That’s the craziest part: this will still be Skylake IPC in late 2019 and early 2020.

    Clocks are good, but piling on 300MHz more to a 4.5GHz part isn’t much. And will we get a Broadwell repeat where the first CPUs on a node are absolutely atrocious on voltage/frequency?

    I was hoping these five years might be worth the wait—I’m very excited about AMD getting serious into mobile. Maybe my next laptop will be Zen-based!
  • Santoval - Friday, August 10, 2018 - link

    "And will we get a Broadwell repeat where the first CPUs on a node are absolutely atrocious on voltage/frequency?"
    This is what was almost going to happen with 10nm and Cannon Lake, which is why Intel is effectively skipping both (barring.. rare 2-core samples with a disabled iGPU) and moving straight to 10nm+ and Ice Lake in late 2019. ASML helped them fix the issues they faced with 10nm but apparently they implemented the fixes in their second-gen 10nm+.
  • caqde - Friday, July 27, 2018 - link

    AMD should have better than Skylake IPC in 2019. In 2020 AMD will have a larger IPC lead with their Zen 3 chips. Currently AMD IPC is almost on par with Intel's. With AMD have between 15 slower to 8 faster than Intel in IPC depending on the task. Worst case is things like Handbrake. Zen 2 could certainly fix this to either be on par or faster if AMD focuses on their weaknesses. Most IPC improvements I have seen with improved Architectures have been dramatic in a few cases and 7-12% most other places and nothing to maybe 3% in the rest. I expect no different with Zen 2, but Zen 2 should have slightly better than Skylake IPC.

    The only website I've seen that did a Pinnacle ridge vs Coffeelake IPC shootout. So judge for yourself is Zen already at Skylake IPC levels? Personally based on what I have seen I think it effectively is.
    https://www.techspot.com/article/1616-4ghz-ryzen-2...
  • zodiacfml - Friday, July 27, 2018 - link

    Wow, more delays after another. They seem to prefer to pocket more cash than pushing technology.
    This is short term benefit.
    Intentionally creating an Intel Pentium4 scenario?
  • Duncan Macdonald - Friday, July 27, 2018 - link

    Intel thought that they could do 10nm without EUV and did not order the EUV machinery. With the long delays on the EUV kit (as the maker AMSL has a lot of orders), Intel will not have production EUV kit until well after its competitors. Expect the first 10nm products from Intel to be in small quantities until they have acquired the EUV equipment.
  • Amandtec - Friday, July 27, 2018 - link

    Ah - finally a proper explanation.
  • Death666Angel - Friday, July 27, 2018 - link

    Huh? Can you explain what you mean? No first gen 7nm node from the competitors will have EUV, as far as I know.
  • Duncan Macdonald - Friday, July 27, 2018 - link

    TSMC 7nm 2nd gen which does use EUV for some layers is scheduled for mid-2019 in volume production. GlobalFoundaries 2nd gen (again with some EUV) is expected in 2019.
    Doing 10nm or 7nm without EUV requires multi-patterning with as many as 4 exposures per step (vs 1 for EUV) which reduces yield and increases production time and complexity in the design process.
    Full EUV (no multipatterning needed) is expected in 2020/2021
    The first gen 7nm from both TSMC and GlobalFoundaries does not use EUV
  • DanNeely - Friday, July 27, 2018 - link

    What's the typical timespan from "we finally got it working right at the R&D scale" to volume production of parts? I know from reading elsewhere that a fresh attempt at the R&D scale takes a few months to get going to the point that they can tell if it's actually fixed the blocking problems or not. So I'm wondering if this is just another round of "we hope we'll have 10nm unfubarred soon enough to hit that deadline" or volume production a year from now means that their current attempt has finally gotten it to work reliably.
  • smilingcrow - Friday, July 27, 2018 - link

    The same question passed through my mind. Seems a long way off unless that is due to needing to wait for equipment to be delivered which has a long lead time as others have speculated.
  • James5mith - Friday, July 27, 2018 - link

    When are we going to see an AMD Ryzen CPU with on-package/on-die royalty free Thunderbolt 3? That's what I'm wanting to see.

    After the 2017 announcement that Intel was going to offer up TB3 royalty free and move the controller on-die for future Intel CPUs, I was hoping for a quicker adoption of the technology by AMD and Intel for that matter.

    https://www.tomshardware.com/news/intel-integrates...
  • OFelix - Friday, July 27, 2018 - link

    Me too!
    Although I don't have a burning need for TB, I'd like to have it in my next system which is currently preventing me moving to AMD.
  • Spunjji - Friday, July 27, 2018 - link

    I wouldn't have thought we'd see that any time soon. Zen 2 has taped out and Zen 3 is most of the way through its design process, too. It would make way more sense for them to integrate it into the chipset than to their CPU in the near-term.
  • Icehawk - Friday, July 27, 2018 - link

    TB of all flavors reminds me of FW, it may be good but adoption was verrrry low (basically just Apple and some camcorders). Hell I can’t get my work to stop buying monitors with VGA inputs.
  • CBeddoe - Friday, July 27, 2018 - link

    Intel seams to be at least a full year or more behind AMD in the server space. Parts on shelves probably means low core count consumer chips with enterprise products to follow... someday.
    AMD is launching 7nm EPYC in early 2019, they already started sampling them.
    AMD is doubling core counts from 32 to 64 (and can probably keep turning this knob pretty easily)
    Rumors are 10-15% IPC Gains, with TSMC boasting significant power efficiency gains, and clock speed increases.
    Intel is still selling or trying to sell ultra expensive monolythic dies in the server space. These chips are much more expensive to manufacture and achieve good yields with. Intel is significantly more impacted by the meltdown and spectre exploits that just seem to keep coming and their pathes have shown 10-25% performance hits in datacenters.
    AMD inovated significantly with Zen and have their foot planted solidly on the gas and their current tragectory is HUGE trouble for Intel. I hope Intel's new CEO can turn things around quickly to keep the competition up because it is great for advancing technology and great for consumers.
    At my place of work 3rd and 4th gen Intel systems are being replaced with AMD as needed. You just can't beat the current value in AM4 and future upgradability.
  • sa666666 - Friday, July 27, 2018 - link

    Where is our resident Intel shill HStewart now that Intel is not king of the 10nm hill anymore. His silence in thread speaks volumes.

    But of course he's in the other AMD thread stating that the 20% increase in AMD stock is nothing, and the drop in Intel stock has nothing to do with AMD :)
  • Maxiking - Saturday, July 28, 2018 - link

    7nm used by AMD is worse than 10nm by Intel.

    I really don't know what are you trying to imply here, but AMD is not king of anything either, so back to your cave.
  • Fulljack - Sunday, July 29, 2018 - link

    lol that's not true. Samsung and TSMC's 7nm are competitive against Intel's 10nm. if you're comparing the MTr/mm², then Intel is slightly worse. not to mention that those two fabs has already sampling their 7nm products, while Intel still couldn't get their 10nm working.

    Intel's the reigning king of manufacturing process? not anymore, bud.
  • iwod - Saturday, July 28, 2018 - link

    I know from people inside what those 10nm problem is, and they are not a big as the media have spin it, they will have 10nm in a few months time. - Said by someone

    Some of the 10nm problem are so big, but Intel are now planning to skip 10nm altogether and go to 7nm. - Said by someone.

    Cannonlake has been launched, and we are going straight to Icelake in 2019, - said by someone.

    I wonder where those people are?

Log in

Don't have an account? Sign up now