Comments Locked

99 Comments

Back to Article

  • milkywayer - Thursday, January 2, 2020 - link

    Intels reluctance to offer you guys a unit for windows testing shows their trust in their product 🤣

    Safe to say the beating AMD has been delivering has been too much. I don't see Intel getting up from this one anytime soon. 🤒
  • Gondalf - Thursday, January 2, 2020 - link

    https://www.eetimes.com/intel-shows-next-steps-in-...

    Sometime a great packaging can save a Company, doing even better than competitors.
    Chipslet saga is good, but there are other more advanced solution around in a short time.
  • Ian Cutress - Thursday, January 2, 2020 - link

    You mean https://www.anandtech.com/show/14211/intels-interc... ?

    Another interview soon on this.
  • Gondalf - Thursday, January 2, 2020 - link

    Thank you Ian 😊
  • bairlangga - Thursday, January 2, 2020 - link

    Are you saying that they did learned something from this seemingly mindless 'trial'?

    https://www.anandtech.com/show/12003/intel-to-crea...
  • Ian Cutress - Saturday, January 4, 2020 - link

    EMIB was in FPGAs before it was in Kaby G. There's more to the product stack beyond the consumer platforms. Intel has 10 year roadmaps for this sort of thing
  • Freeb!rd - Friday, January 3, 2020 - link

    The fact that the 9200 EVEN exists proves Intel knows they have no product in the near future that can match it... why would anyone in their right mind (apparently not Intel) pour the resources into a product like this if they had a product that could compete with the AMD (Rome) EYPC? or 2020 Milan? To me this only cements speculation as fact that Intel will not have a competitive product in the server market until 2022 at the EARLIEST.
  • patrickjp93 - Friday, January 3, 2020 - link

    Because even if the product is fully designed and taped in, the lag time to market is still another 8 months. Jesus Christ are you people actual tech enthusiasts, or are you just armchair analysts?
  • Korguz - Friday, January 3, 2020 - link

    patrickjp93 ok.. then explain how intel was able to release the initial response to 1st generation zen. oh wait.. they just took their xeon cpus, crippled them, disabled the server features, and relabled them core i7s and i9s... come on.. i think some details about epyc rome were out with plenty of time for intel to start to work on replies to it... and this xeon.. is just that... although some what barely...
  • Irata - Saturday, January 4, 2020 - link

    I think Intel’s Xeon 9200 processors are trying to solve one specific issue: To have a system with a higher core count to present in Benchmarks that they (Intel) run. It's a "look here CPU.
  • GreenReaper - Friday, January 3, 2020 - link

    I'd take them at their word: Windows is simply not the market for this CPU. They're not trying to sell it to that market, you don't want a ~400W server as your workstation, and you're not likely to run it in a regular server app for the typical Windows applications either. It's for scientific workloads, if anything. Most of those customers would rather pay for hardware than for an OS that gets in the way and which they can't as-easily customize.
  • Spunjji - Friday, January 3, 2020 - link

    "Windows isn't our market... so we're not going to let you do apples-to-apples testing of our hardware using your established test patterns."

    Nope, that logic does NOT check out. It only makes sense if there's something peculiar about the 9200 setup that is unfairly hobbled by Windows, a-la early Threadripper CPUs, yet even then AMD managed to let people test those on Windows.
  • patrickjp93 - Friday, January 3, 2020 - link

    Other than SQL Server, no, Windows is useless for this kind of platform.
  • evilpaul666 - Friday, January 3, 2020 - link

    Wendell from L1Techs (Dr. Ian was a guest on a recent stream) has been trying to get high core count stuff working right under Windows and is been an uphill battle. I'm pretty sure AMD would end up ahead, and pretty sure Intel knows that too, but saying Windows is awful isn't wrong.
  • Gondalf - Thursday, January 2, 2020 - link

    In my knowledge Intel will fix actual AP serie issues stacking two 28 cores dies via TSV. This solution apparently is for sometime 2020. There will be a power reduction and far better latencies between cores.
    So Intel has answered to Ian in advance finding new solutions.

    My bet they will do the same on Ice Lake SKUs for a more advanced AP lineup.
  • Ian Cutress - Thursday, January 2, 2020 - link

    I doubt you'll see XCC sized die stacked. Not only big dies, but therman density with any reasonable TDP would be crazy
  • Gondalf - Thursday, January 2, 2020 - link

    Depends on cooling solution adopted. One can cool from the up and from the down.......nice puzzle 😊
  • Operandi - Thursday, January 2, 2020 - link

    Yeah, if figuring out how to have such a CPU not melt itself down to a slag of glass sounds nice to you.

    All these high tech interconnects that Intel talks about are interesting but are all completely theoretical with their current CPU architecture and manufacturing process roadmaps for the foreseeable future. AMD and TSMC (and probably Samsung as well) are so far ahead right now and you can be certain they are not sitting still when it comes to next gen interconnects either.
  • Gondalf - Thursday, January 2, 2020 - link

    Basically the larger silicon company is done of a bunch of morons?? I can be true.
    Still i follow Intel since the beginning and i have read comment like yours many times. Sadly the writers of the comments have disappeared miserably.
    In the meantime Intel is here with record revenue on year basis.
  • Korguz - Thursday, January 2, 2020 - link

    and yet.. they stagnated the cpu industry for the last few years... didnt innovate.. didnt move the industry forward.. gave 10%or less performance increases year over year while over charging for that increase. all while lieing to its customers... give it a rest gondalf.. stop being such an intel fanboy for a moment.. and you will see intel... isnt as good as you make them out to be...
  • Spunjji - Friday, January 3, 2020 - link

    Open with a straw man, follow with an appeal to authority, end on some vague pontificating... this comment is not worth much.

    Intel have record revenue because they're *large* thanks to a history of executing better than their competitors, not because they are currently executing well.
  • Lord of the Bored - Friday, January 3, 2020 - link

    Problem with cooling "from the down" is that you lose the ability to put electrical connections on the bottom of your processor. While I admit a processor with no electricity in it stays VERY cool, I hesitate to call this an optimal solution.
  • GreenReaper - Sunday, January 5, 2020 - link

    Just need to make a new vertical package which you can attach a few fans to on either side. We could call it...oh, Slot Duo.
  • Ghan - Thursday, January 2, 2020 - link

    This whole product line seems to be a scramble to address the looming threat from AMD. When you start requiring liquid cooling in your deployments, you're not really serious about providing customers a solid all-round solution to their computing needs. I honestly wouldn't be surprised if the top end AP series CPU has a hard time competing with the newly released 3rd gen AMD Threadripper CPUs by simple fact of the clock speed difference.

    Supercomputer customers are going to start looking at AMD this year unless Intel can come up with a real answer. The work required to shift platforms isn't small, but AMD is making a very compelling argument for doing so right now.
  • Gondalf - Thursday, January 2, 2020 - link

    The lower single thread performance of Epyc line is a serious limiting factor. Now Amd have a lack of 20%, the thing will not change anytime soon. This the reason Epyc is right now an unbalanced solution under many aspect.
    An higher number of cores is not always a real advantage.
  • Ian Cutress - Thursday, January 2, 2020 - link

    Intel 9282: 56 core, 3.8G turbo
    AMD 7H12: 64 core, 3.3G turbo
    AMD Zen2 has IPC over Sky/Cascade.
  • Gondalf - Thursday, January 2, 2020 - link

    Spec 2017 best submissions on integer single thread Intel 11, Amd 9.
    (Intel a 4 Ghz turbo, so the advantage is a little lower on AP line).
    Looking all results available on Spec site, Skylake is 5% ahead in IPC. Spec bench is capable to bypass the gigantic L3.
  • schujj07 - Thursday, January 2, 2020 - link

    There are enough benchmarks out there that show that Zen2 has a higher IPC, 5-8%, than Sky/Cascade Lake CPUs. https://www.anandtech.com/show/14605/the-and-ryzen... Intel is able to counter this IPC disadvantage by having clock speeds over 10% higher. However, in the server world ST performance is not what we are looking for. Every review out there of the Epyc 7002 series against the 2nd Gen Xeon Scalable shows Epyc is faster than Xeon in the areas that they compete. This review is one of the most complete server CPU reviews out there and shows that the Epyc 7002 series dominates the Xeon https://www.servethehome.com/amd-epyc-7002-series-...
  • Gondalf - Thursday, January 2, 2020 - link

    I don't see your 5/8%, this is consumer core that is different from Xeon (1 MB L2), it is Spec the golden standard. A large L3 cache very easily hide weakness and increase the performance even on a weaker core, expecially in consumer (remember Intel EE edition??). Spec submission show clearly what is the truth. Moreover Intel turbo advantage (over 20%) is enough to beat badly Epyc in ST on server tasks.
    About ST, likely YOU are not interested, too bad the bulk of customers look at this result before to evaluate a cpu. Pretty certain your workload is not so stressful for single core or task.

    Amd have chosen a nice of the market with Epyc line. To pack 64 cores in the TDP budget they chosen the SOC process cutting severely the single thread performance.
  • Korguz - Thursday, January 2, 2020 - link

    gondalf, give it up already.. face it.. intel is at a disadvantage right now, no matter HOW you try to spin it, amd has the better performance cpus right now. if you THINK large caches make up for weak performance, then WHY didnt intel make the cache sizes of their cpus bigger ??? its plain and simple, CLOCK FOR CLOCK, amds IPC is currently higher then what intel has right now, the ONLY way intel has ANY performance lead, is ONLY because of the clock speed, AND they get that performance as MUCH higher power usage and thermals.
  • DigitalFreak - Thursday, January 2, 2020 - link

    I think Gondalf is Gandalf's mentally challenged troll half brother.
  • Freeb!rd - Friday, January 3, 2020 - link

    To be honest, Intel can't "afford" to match the L3 cache increase of Zen2 when comparing their 14nm to TSMC 7nm. Increase die size, decreased yield and maybe increase thermal output. If they are truly capacity limited as they claim adding to the L3 size would only exacerbate the problem and only get them a few % performance return.

    That said, Intel is in a "world of hurt" for the next several years, unless they can pull a rabbit out of their "fab" hat.
  • schujj07 - Thursday, January 2, 2020 - link

    The uArc of the Core i-9000 series and 2nd Gen Xeon Scalable are the same just like Ryzen 3000 series and Eypc 7002. The link to the Ryzen 3900X review has a chart that shows "Estimated Performance per GHz" for Spec 2017 and when you do a little math you find that Ryzen 3000 series is 5-8% faster that Core i-9000 series when running at the same GHz. Or you can just look at what Andrei said under that graph "Normalizing the scores for frequency, we see that AMD has achieved something that the company hasn’t been able to claim in over 15 years: It has beat Intel in terms of overall IPC."

    "About ST, likely YOU are not interested, too bad the bulk of customers look at this result before to evaluate a cpu. Pretty certain your workload is not so stressful for single core or task." That comment makes no sense at all. Almost all tasks on a modern server a multi-threaded even before you get into virtualization of the hardware. If you can have your host CPUs running at 50% all the time then you are in a very good position. That will affect the high turbo speeds of Intel to the point that the Intel and AMD CPUs will be running at around the same clock speed. Since Epyc 7002 series has higher IPC than Intel, that means the performance will be better on the VM with AMD. Just look at the review of the Epyc 7402P and compare results against the Xeon 8268. Both CPUs are 24c/48t but the Epyc runs a 2.8/3.35GHz and the Intel run at 2.9/3.9GHz https://www.servethehome.com/amd-epyc-7402p-review... The CPUs trade blows for who is on top in the benchmarks. Not to mention these are REAL USE CASE Scenarios for the server CPUs.
  • yeeeeman - Thursday, January 2, 2020 - link

    You guys seem to not understand what he said. He said that if you have workloads that don't fit into the l3 cache of epyc, then the ipc of Skylake is better than zen2 and I suspect that is the case given both cores specs.
  • Korguz - Thursday, January 2, 2020 - link

    yeeeeman remember.. gondolf is also an intel fan boy.. and is trying to spin this to make intel look better no matter what... the fact is.. intel is at a disadvantage right now.. and he thinks.. zen has higher ipc mostly because of the bigger caches.. but doesnt realize.. there is more to it then just the size of the cache
  • Gondalf - Thursday, January 2, 2020 - link

    Exactly, this is the goal of Spec benchmark, in fact the dataset of the recent revision is so gigantic to bypass the L3 absolutely. Obviolusly it is a bad idea utilize a single istance of Rate subset because the dataset is much smaller and is sometime unable to give a good indication of the IPC.
    Finally a man with a brain :).
  • Korguz - Thursday, January 2, 2020 - link

    ahh but i do. in another thread ( cant remember which one now ) he was adamant that zen 2s IPC increase, was pretty much due to the bigger caches. but he CAN'T explain why clock for clock, core for core, zen2 has a higher IPC then intel.

    from anandtechs year in review cpu article :
    " and matching if not beating Intel in raw clock-for-clock performance "
    " More importantly however, in terms of performance per clock against the Core i9-9900K, the new Ryzen 9 3900X on the industry standard SPEC2017 tests, was scoring higher for its frequency than Intel was. This was broadly interpreted as AMD taking the IPC crown from Intel on such a significant test, meaning that AMD doesn’t need to hit such a high frequency to get the same performance as Intel."
    there are a few other things in that article as well that dont paint a good picture for intel.
  • AshlayW - Saturday, January 18, 2020 - link

    Huh? Zen2 is a wider core, with a bigger uOp cache and a better front end branch predictor. Core to core, it's likely that Zen2 has higher IPC.
  • Gondalf - Thursday, January 2, 2020 - link

    Remember that Xeon have a far larger L2.
    Anyway your post is too long and with few relevance.
    I have only said the core IPC of Skylake Xeon is better than Zen 2. So under a single thread a Xeon shines and Epyc nope. Obviously the aggregate performance of an 64 cores Epyc is higher, still i want to remember that right now Intel is the absolute dominator of 32 core class SKUs, better suited to manage hard tasks with a large memory footprint. Basically AMD is hungry because nobody want their server line under 48 cores. This hurt the revenue a lot leaving to Intel the large part of the market.

    Anyway there are the official submissions on Spec site, enjoy, uilize the brain and take some medicine.
  • schujj07 - Thursday, January 2, 2020 - link

    Using your own logic then the i series CPUs and Ryzen CPUs are the perfect way to measure IPC of Zen and Core from Spec since they have less cache than Epyc and Xeon. When the scores were normalized for clock speed the Zen2 uArc was 5-8% faster than 9 series Core. You seem to equate ST performance and IPC being the same thing, but they aren't. ST performance is IPC * MHz. This is why the 9900k has better ST performance than the 3900X all while the 9900k has a LOWER IPC.
  • schujj07 - Thursday, January 2, 2020 - link

    "Still i want to remember that right now Intel is the absolute dominator of 32 core class SKUs, better suited to manage hard tasks with a large memory footprint"
    Intel outsells AMD in server CPUs that everyone knows. However, performance wise the CPUs trade blows with each other when at equal core counts. Epyc 7601, 32c/64t, was slower than Intel 8180, 28c/56t, in most applications. Now the 7402, 24c/48t, is equal in performance to the 8268, 24c/48t, but the 7502, 32c/64t is far faster than the 8280, 28c/56t, except in the small subsets of applications that use AVX512. Both AMD CPUs have about a 15% base clock speed disadvantage so the only way to make up for that loss of speed is by being faster per clock.
  • Korguz - Thursday, January 2, 2020 - link

    " i have only said the core IPC of Skylake Xeon is better than Zen 2 " to bad it DOESN'T have better IPC then zen 2 clock for clock, its SLOWER, it ONLY gets its performance, because it is clocked higher. when will you understand this ??

    " Anyway your post is too long and with few relevance." like your posts are just intel irrelevance??
  • AshlayW - Saturday, January 18, 2020 - link

    You are mistaken. Intel has a better memory controller in terms of raw latency, due to Zen2 chiplets design. That's it. Zen2 makes up for it with a bigger L3 cache. Core wise, Zen2 is wider, has a bigger uOp cache, a better branch predictor, better cache and memory parallelism, etc. Intel's only advantage in core performance is clock speed. Jeese, after reading your comments it's clear to me you're utterly delusional and/or financially invested in Intel stock. You have zero credibility and only seem to promote intel on every article, whilst attacking AMD products.

    It must really hurt you that Intel's skylake architecture is getting so utterly beaten in essentially all metric, yes?
  • martinw - Thursday, January 2, 2020 - link

    SPEC is not really the gold standard - it is a benchmark which is heavily optimized for by Intel using their own compiler. Few companies use the Intel compiler in the real world, so results based on Intel reported SPEC is not really that representative of real world performance.
  • Korguz - Thursday, January 2, 2020 - link

    martinw , gondalf doesn't care, he only cares about putting intel in the good light.. at all costs.
  • AshlayW - Saturday, January 18, 2020 - link

    I'm going to feature your comments on my website, I'm always looking to expose misinformation spreaders; especially those that zealously support disgusting, anti competitive and anti-consumer companies like Intel.
  • AshlayW - Saturday, January 18, 2020 - link

    Your misinformation spreading is disgusting. Stop it. Large L3 is part of the Zen2 architecture. Overall IPC is higher than Skylake, all included.

    Furthermore, your single threaded argument is grasping at straws and completely moot for almost all HPC workloads.

    Stop embarrassing yourself, hon.
  • Ghan - Thursday, January 2, 2020 - link

    Not to mention that AMD is providing this level of performance in a much more manageable power envelope due to their more efficient process node. And it's also worth noting that AMD's platform as a whole is much more capable. Where the Cascade Lake-AP line only has PCIe 3.0 (and a mere 40 lanes at that), AMD's 7H12 brings 128 lanes of PCIe 4.0 to the table.
  • prisonerX - Thursday, January 2, 2020 - link

    You seem to think that Epyc processors are for playing (single threaded) games. Four cores should be enough for anyone, right!? Well, at least Intel insisted that for a decade, until AMD ate its lunch. Now dumb Intel shills like you are trying to push that fiction along with the 20% difference lie. Shill harder, rube.
  • R7 - Thursday, January 2, 2020 - link

    EPYC Rome has all but closed the ST gap. Plus Intel has been hit hard by their security issues. That also means ST perf loss. Plus the TCO argument. You can get a 64c/128t Rome that uses 250-280W and is in mass production.

    Where as Intel's Xeon 9200 seems to be made for order only. Has 8c/16t less. Is BGA not LGA and uses 400W all while costing ~3 times as much as AMD's part.

    Even if this mythical unicorn exists somewhere and has higher performance than AMD's system the TCO argument blows it out of the water. For the same money you can get 2! AMD's top end EPYC Rome CPU's and have leftover for a 32c/64t one too. Plus an overall faster and more scaleable platform that is easier to deploy, maintain and upgrade.

    If i was running servers the only way i would consider Intel is if they gave away their CPU's at at half the price. Even that would be a hard sell for me.
  • Teckk - Thursday, January 2, 2020 - link

    AMD 64-core Epyc is 280W unlike the Intel's 56-core which is 400W too.
  • R7 - Thursday, January 2, 2020 - link

    Actually it's two 200W models. One 225W model and the top end 7H12 model that is 280W.

    Also in few months there will be first EPYC Milan CPU's that further lower the TDP from 120-155W on 16c/32t part to 100W. I would not be suprised if we see the top end Milan part being "only" 200W later this year when the production ramps up.
  • R7 - Thursday, January 2, 2020 - link

    https://en.wikipedia.org/wiki/Epyc#Second_generati...
  • Teckk - Thursday, January 2, 2020 - link

    The 2nd-gen 16/32 Threadripper is at 180W. I'm skeptical on this claim that they can get the Milan Epyc 16/32 part to 100W. Will be great if that happens but I'm not sure of that
  • schujj07 - Thursday, January 2, 2020 - link

    Ryzen 3950X is 105W and is 16c/32t. Going to the EUV 7nm node could make a 16c/32t Epyc go sub 100W.
  • zmatt - Thursday, January 2, 2020 - link

    It really isn't, except for some pretty niche use cases. When I look at these systems I see VM hosts. more cores for your money is far more important than a few percentage single threaded performance. With VMs, having huge single threaded performance isn't versatile. Its nice to have but its tied to the individual threads, you can only provision one thread to one VM at any given time. So it may make an individual VM run faster, but most servers don't need a lot of resources to begin with and most of the time the bottleneck isn't the cpu its storage or networking. So more threads means I can support more VMs on a given host. And if you are clustering then the benefits are even greater.
  • zmatt - Thursday, January 2, 2020 - link

    It's also worth noting that servers tend to have a life of around 5 years so the systems being replaced by Epyc today are going to be based on Haswell which not only has fewer cores but much worse single threaded performance both in terms of frequency and IPC.
  • schujj07 - Thursday, January 2, 2020 - link

    TCO benefits of Rome are so good that a company can save money on replacing Xeon V4 or 1st Gen Xeon Scalable.
    From page 10 of Patrick Kennedy's review of the Eypc Rome
    "If you are installing a new VMware ESXi cluster, you can get an AMD EPYC 7002 series CPU, like an AMD EPYC 7702P, and consolidate legacy servers at no worse than 2:1 ratios but in some cases 6:1 socket consolidation ratios even over early 2017 generation parts."
    https://www.servethehome.com/amd-epyc-7002-series-...
  • zmatt - Thursday, January 2, 2020 - link

    Very true although I'm not sure how many users will replace such new hardware that quickly. Unless your business lives and dies by getting absolutely the most out of its compute then I think its unlikely many will replace hardware that is still within its support contract.

    Doing some quick calculations Epyc 7702P would allow us to consolidate around 5:1. The number of PCIe lanes per thread is slightly worse but going from 3.0 to 4.0 means there is still more bandwidth overall available. It's a great time to be in the market for new servers. I have some old ESXi hosts to decommission.
  • diehardmacfan - Thursday, January 2, 2020 - link

    Unfortunately socket consolidation isn't quite what it used to when many licensing models switched from socket based to core based.
  • yetanotherhuman - Friday, January 3, 2020 - link

    Yeah, it reality, EPYC's density doesn't help. It means you're putting more VMs on fewer boxes, which reduces your overall reliability, all with high licencing costs. Ugh.
  • Spunjji - Friday, January 3, 2020 - link

    It's not exactly a reliability *reduction*. Having fewer servers could reduce fault tolerance, but it also leaves you with fewer things to go wrong in the first place. As long as you maintain the same level of overall failure tolerance, having fewer servers can increase reliability.
  • zmatt - Friday, January 3, 2020 - link

    As Spuniji showed, that's not how fault tolerance works. As long as you maintain fault tolerance at the same level it isn't a problem and is still an improvement.

    Lets take my 5:1 example from earlier. If I have a single failover node for each that means I have 10 physical VM hosts running at any one time. By consolidating those systems into one I can reduce the number of physical hosts to two, one hot and one failover and have the same level of fault tolerance for any given VM.

    Of course the reality would be that you wouldn't let that freed up rack space, power budget and cooling capacity go to waste. So I would implement a failover cluster of at least three machines, but I could use more to have even higher levels of fault tolerance than before.
  • Spunjji - Friday, January 3, 2020 - link

    you're discussing last-gen Epyc performance in a discussion about current chips. Anandtech's own tests ought to disabuse you of this illusion, but it looks like you're just here to shill.
  • Soda - Thursday, January 2, 2020 - link

    I believe AMD is recommending the new 3rd gen Threadripper CPU's to be watercooled as well. Not required but recommended.
  • PeachNCream - Thursday, January 2, 2020 - link

    Threadripper isn't EPYC.
  • Xyler94 - Thursday, January 2, 2020 - link

    Threadripper 3 tops out at 280W of TDP, while these top out at 400W.

    TDP is measured differently between AMD and Intel. AMD typically measures TDP as the typical sustained heat output when the processor is running pretty hard, while Intel measures it at the CPU's guaranteed base speed (Base clock). So yeah, 400W is much harder to cool than 280W.
  • tamalero - Friday, January 3, 2020 - link

    So this means.. that if the XEONS go fully AV512 or all turbo, they will consume more than 400W just like the 9900K and KS do right?
  • Spunjji - Friday, January 3, 2020 - link

    I believe they do things differently with the servers. 400W is likely to be a hard cap for reliability reasons, so the outcome of running AVX-512 and/or all-core loads is lowered clock speeds (dramatically lowered for AVX-512).
  • azfacea - Thursday, January 2, 2020 - link

    intel is clearly getting murdered here and everyone knows that. but its TSMC not AMD thats doing the killing. the AMD hype is reaching fusion temperatures but maybe its misplaced. what does AMD have ? a uArch thats decently high perf for its time plus a few other less important things. thats it. So do at least another dozen companies. intel, IBM, Apple, Arm, many arm licensees, on top of that we are probably going to see a lot of ARM, RISC-V, and even x86 core designs from china soon enough.

    If intel doesnt fix their semi problems soon enough, they will be going up against dirt cheap ARM CPUs that make them wish for AMD's pricing
  • R7 - Thursday, January 2, 2020 - link

    It goes both ways. Without compelling products to manufacture TSMC might as well have the best process node in the world and it would not matter.

    Where as AMD could also go to Samsung if they really needed to.
  • azfacea - Thursday, January 2, 2020 - link

    what ? "having best process node in the world" dont matter if they don't get AMD's current 3 % maybe soon to be 10% server market. clearly you have no idea where the industry is and where most of silicon demands comes from these days. Apple/Huawei volume alone is got to be more than an order magnitude greater than AMD, if not two orders mag.

    TSMC only faces one competitor in samsung. For any one else to enter and compete it will take years if not decades, tens of billions of dollars if not hundreds. On the CPU design side there are dozens of companies already and as the world moves to ARM (look at AWS and Apple's plans) so will software. In any event, x86 patent has long expired, AMD64 patent will expire soon, and there is VIA. so x86 IP is not going to protect AMD.
  • Spunjji - Friday, January 3, 2020 - link

    "and there is VIA"

    Where? I've heard story after story about them making a comeback over the past 15 years. If it were as easy to do as you're implying, then they'd have done it by now. Their current best is supposedly equivalent to an i5-7400 - not a terrible attempt by any means, but not even close to proving your point about AMD not being the architects of their own success.
  • azfacea - Friday, January 3, 2020 - link

    zen 2 level perf will cost 1 billion to develop, much to less license from ARM. I never said VIA is making a comeback, I said x86 license is there and the patent will expire anyway so its an open ISA.
    TSMC's leading edge nodes however will cost tens of billions for someone to develop.

    AMD are in fact not the architect of their success. Intel and TSMC are.
  • Korguz - Friday, January 3, 2020 - link

    um yea ok... even though AMD has done more innovating then inte has...
  • Lord of the Bored - Friday, January 3, 2020 - link

    Google "amd64". Intel's been making Athlon clones for years.
  • TheinsanegamerN - Monday, January 6, 2020 - link

    amd64 is not an architecture. Architecture =! instruction set. Unless you are also claiming AMD is still making intel 8086 clones to this day.
  • Korguz - Monday, January 6, 2020 - link

    i think the point was.. that intel has copied more of amds innovation.. then intel has done their own innovating..
  • akvadrako - Tuesday, January 7, 2020 - link

    IA-64 (also called Intel Itanium architecture)
  • Spunjji - Friday, January 3, 2020 - link

    Why are you trying to claim that having a uArch that edges out Intel on performance isn't really a big deal? They're the world's largest single spender on microprocessor R&D, and their primary market is CPUs. AMD managing to scrape out a lead is huge. It's telling that you couldn't name a single other shipping product in the server market that meets this (actually much higher than you're letting on) standard.
  • prisonerX - Thursday, January 2, 2020 - link

    These desperate moves by Intel serve only to illustrate how deep in the poo they are. Is anyone still in charge over in Santa Clara?
  • PeachNCream - Thursday, January 2, 2020 - link

    I don't think Intel is making moves that might be called desperate. The company is not highly leveraged, has lots of cash on hand, and has its fingers in a number of non-CPU markets as well. Yes, CPUs are their bread and butter, but they are facing an AMD that has spent a rather long while accumulating debt and has a specialized rather than broad product portfolio with a lower volume of business and significantly less revenue with which to fund R&D.

    I don't mean to devalue the accomplishments or the impressive showing AMD has made with Zen, but Intel is not hurting badly yet. I view the 9200 as more of a freedom-to-experiment product that is fishing for possible unmet demand (which, of course, does not appear to exist at present) than the uncontrolled flailing of a company locked in an innovate or die struggle with a competitor. That may change quickly as things often do in the tech industry, but right now Intel is not in a state of desperation.
  • Spunjji - Friday, January 3, 2020 - link

    I feel the truth lies somewhere between these positions.

    I do think this product is a flail. It's clearly an attempt to save face and/or stymie a competitor, because it makes no sense from any other perspective.

    I agree that they're nowhere close to desperate yet, though - not from an existential perspective.
  • lefty2 - Thursday, January 2, 2020 - link

    Level1Techs had a video that mentioned the HP engineers that setup the Rome versus Platinum 9200 demo for Intel ... during the test they decided to overclock Rome just a little bit to see how it would perform and it beat the Intel 56 core system.
  • schujj07 - Thursday, January 2, 2020 - link

    Patrick Kennedy at servethehome.com had an article similar to this in June 2019. https://www.servethehome.com/intel-xeon-platinum-9...
    It is interesting to read that even 6 months later, vendors have no interest in the 9200 series.
  • yeeeeman - Thursday, January 2, 2020 - link

    Imo this was wasted money and resources...
  • Lord of the Bored - Friday, January 3, 2020 - link

    Nah. This is a marketing part, it exists to create a bullet point in the AMD/Intel argument. And based on the comments, it is wildly successful.
  • eSyr - Friday, January 3, 2020 - link

    "Because an individual package has two silicon dies on it, this hardware is limited to a dual socket configuration, which acts pretty much identically to a traditional quad-socket configuration."
    "a user can get dual-socket like performance with only a single socket"
    "For any system that uses a 350 W or above, either single or dual socket, then liquid cooling is required."

    But this is not a socketed CPU, it uses BGA packaging. Probably, "single/dual CPU" is a better fit here.
  • Arnulf - Saturday, January 4, 2020 - link

    There's a period missing after "or push Intel to actually put dollar amounts on its products".
  • MASSAMKULABOX - Saturday, January 4, 2020 - link

    Ive always hated INtels feet .. they are dirty and smelly ..
    "one of the criticisms leveled at Intel’s feet" .. too much xmas?
  • GreenReaper - Sunday, January 5, 2020 - link

    No... It's an idiom - to lay something at the feet of someone is to assign responsibility to them. https://en.wiktionary.org/wiki/lay_something_at_th...
  • SanX - Saturday, January 4, 2020 - link

    Noticed that no pricing was mentioned? That is a style of current AT to dance only for those who pay for that. What was not said is that these 500 cores Intel system will be 3-4x more expensive than an equivalent 500 cores AMD system
  • TheinsanegamerN - Monday, January 6, 2020 - link

    Meanwhile, back in reality, Intel likely has not released pricing information because they are stupid expensive for what they offer.
  • AshlayW - Saturday, January 18, 2020 - link

    Read the article you Muppet.

    Here, I'll make it easy for you:

    INTEL HAS NOT RELEASED LIST PRICES FOR THESE PROCESSORS.
  • peevee - Thursday, January 16, 2020 - link

    Do I understand correctly that those Xeons support 12 DIMMs per CPU, so have 6 memory channels? And normal Xeons have 3 memory channels?
  • peevee - Thursday, January 16, 2020 - link

    Or is it 6 channels/12 DIMMS normal Xeons and 12 channel/12 DIMMs for 92xx?
  • AshlayW - Saturday, January 18, 2020 - link

    Probably this. And probably for space reasons. 2DPC on 12ch is an awful lot of dimms
  • AshlayW - Saturday, January 18, 2020 - link

    XCC die has 6 channels. The dual-XCC package thus supports 12 channels.

Log in

Don't have an account? Sign up now