actually you are correct. well not just 28 core dies. but the whole xeon lineup. there are stories of them giving steep discounts to xeon customers who have gotten quotes on epyc systems. everybody is upgrading from their broadwell\haswell era xeons at once and its the main cause of this shortage. intel wont go down without a fight and even though it appears that epyc will wipe the floor with them as the current msrp's those prices arent the real prices. its actually much closer than it would appear
A bigger question is if and when one can actually get these. Besides Amazon and Google and the like I don't see the 24-28 cores in the wild, point me to a vendor selling them/with stock? Will these be any better?
And this design isn't going to be available until some time in 2019. Sure, it makes sense for Intel to compare against the 7601, as that's the best CPU AMD has to offer at the moment, but with Rome expected to be announced at CES 2019 (January), and rumours/speculations pointing towards an 8x Zen2 chiplet design, this is likely what Cascade Lake will end up competing against, even though Rome is supposedly designed to compete against Ice Lake (the generation after Cascade Lake from what I understand).
But - since neither of these chips are available, making claims about what is and isn't current design makes little sense.
Hey, look at that - AMD has announced 64 core Rome the very next day. So now Intel's newest architecture really IS behind by 16 cores. Oh, and it has PCIe 4.0 support out of the box.
Intel has been connecting dies together in a package since the days o Pentium D and just because they did today - one calls this infringe on patents from AMD and also glue, if that was truly the case then AMD has infringe on Intel patents
One thing I am deeply concern on these patents - if an Intel person says anything about AMD product - they get viciously attack - but then they is article like this where Intel has done something and it get polluted with discussion about AMD.
I have the right to like Intel, and so does users that like AMD, but would success some maturity in the forums and keep this Intel vs AMD stuff out of logical discussions.
This is rich; the resident Intel fanboi extraordinaire complaining about fanboyism. You're the one that constantly brings mention of Intel into AMD articles, as you just can't resist when Intel is being cast in a bad light. And now you want people to stop talking about AMD in an Intel article?
Try looking in the mirror and analyzing your own behaviour before you denounce it in others. And to be clear, it's not Intel favouritism from you that is annoying, it's the complete illogic of some of your statements. If you would just preface all your articles with "I am an Intel supporter, always will be, and don't accept that they can do any wrong", then maybe some of your comments would be more palatable. At least you'd be honestly sticking to your opinion without the passive-aggressive attitude.
I only mention Intel in AMD articles if some one else blabs it out.
I don't consider myself Intel Fanboy - I have other products include from AMD - I just prefer Intel, it seems that some uses are AMD Fanboy's - they are consider Intel Fanboy.
I honest did try to give AMD a test with my Dell XPS 15 2in1.
I would take a look in the mirror from AMD statements - at least Anandtech is not as bas WCCFTech or ExtremeTech which are bluntly AMD bias places.
I wrote this message, because I don't like Intel related article that keep pushing AMD stuff - if this was done on AMD article - one would be harass like no tomorrow. Only because of long history of techy geekness that I even care about forums.
But lets be honest, the average consume does not care if Intel has dual connect 24 Cores and AMD core is 32 core ( actually 8 connect 4 cores ) they want something to emails and spreadsheets.
Which to some people might be iPad or Samsung Tab Sx tablet - which is actually Intel and AMD's bigger threat.
Being a code monkey for the part 30 years hasn't helped your communication skills in the slightest. Beyond that, if you really think that everyone who believes you're an slobbering Intel shill _must_ be diametrically opposed to you and a slobbering AMD shill yet somehow refuse to also acknowledge your blind loyalty, then I'm glad I've never hired you to bang out code for any project I've led because you're truly lost in the sauce.
Yikes, your awful English is rubbing off on me. "Being a code monkey for 30 years hasn't helped your communication skills in the slightest."
Sheesh, I need a time machine to hop over to Anandtech of the future where we get edit capabilities. Hmm...I guess I'll set the time machine's dial to 1998. That sounds like a reasonable year to expect an edit button.
My English skills is because of Aspeger's - almost perfect SAT math in high school but was in with Football players for English in college. I believe my mind moves faster than I write. Unfortunate these forums do not allow to edit.
In case you will notice that I did not comment on any of AMD announcements today like 64 core stuff - all I wish is AMD fanboy's have the same respect for Intel threads.
Really, you didn't see that one coming? I can't know your real thoughts, but from what you're writing you come across as very pro-Intel to me. Actually to a point where I stop reading most opf your posts after a few lines and just think "oh no, he's again explaining why everything in the Intel world is great".
In my opinion Intel deserves jokes about this - it's been less than 1,5 years that they harshly criticized AMD for the "just glued together" approach.
Maybe more that I dislike AMD instead of Pro Intel. I have absolutely no stock or work for Intel, I just don't like what happen to Intel related to original IBM computer situation. IBM required a second source of products and that is why AMD is out there. If I spent my hard earn time investing in a new product and another company comes along and clones it and take away my product - would you be upset.
It not related to their GPU industry, that was from ATI - that does not matter.
Most people on this forum are probably to young to remember Intel and IBM stuff when PC were first created and would not understand.
But I would agree in the situation, that AMD has help the industry - help Intel - by keeping Intel making new products - just like this product but I still prefer the original CPU - vendor and not the clone AMD.
Part of this maybe when I got my first PC - it was 386 PC - It had an AMD clone CPU in it and yes it was a cheaper clone pc - but the make Magitronic had an AMD CPU.
And again, you should just post at the start of your reply: "I am an Intel supporter, always will be, and don't accept that they can do any wrong". Can you not see the rubbish you're posting? Complaining that AMD was a second-source to the original IBM computer! Jesus H Christ, this was over 30 years ago. If you're holding a grudge about a non-event from that long ago, you have major issues.
Can you not see how someone would interpret your rantings as that of a lunatic? Why would you care what happened to Intel vs. AMD from over 30 years ago? Were you part of the company and lost a lot of money or something??
Again, can you not see how many people would interpret what you're saying. This isn't pro-Intel or anti-AMD at this point. You seem to be severely delusional about an event that had nothing to do with you personally, and happened over 30 YEARS AGO. Let it go already.
Oh, and the appeal to seniority/authority, having done development for over 30 years, is useless. I also have 30 years experience in dealing with computers and professional development, on everything from an Atari 2600 to the latest mainframes. And probably many other people in these forums do too. I/they just don't feel the need to brag about it, as you seem to.
Just because you have experience doesn't make your opinion valid or even sane. And something is not true just because you REALLY WANT IT TO BE TRUE.
Not sure if this is trolling or you are autistic or something. Obviously this means that in the benchmarks, Intel disabled SMT in the BIOS for the AMD system. To make their system look better, of course.
Yup. Disappointed that it wasn't called out in the article. "We are going to be almost 4x faster if our imaginary numbers prove accurate AND you kneecap the Epyc."
It isn't a good look for Intel, but it does show they have some serious concerns about the competition.
They also used binaries in the benchmarks specifically compiled for Intel's processors. So they basically raced against AMD. Having removed half the tyres from the AMD vehicle as well as making AMD race on a dirt track whilst Intel were on the tarmac.
I expected simpler "glue 2x 28C parts", which would lead to 16 memory channels and a 56C monster with pretty decent number of PCIe as well - not quite as high as on AMD side, but still decent. Why would Intel make such a puzzling chip instead - what is the benefit here? Do they feel AMD isn't THAT much of a threat so they can afford offering something intermediate or what?
I'm guessing the benefit is winning some time while they solve their 10nm woes and have a really scalable architecture that allows them to compete with Epyc. I mean compete in economic terms, they clearly still have an IPC lead, although a shrinking one, but they need a way to make CPUs with far more than 28 cores at a reasonable price. For the time being they will glue together two XCC cores to at least remain competitive agains the coming Zen 2 Epyc CPUs in terms of core counts, while they figure out how to also compete in TCO, or at least to remain reasonably competitive.
I'd argue that the perceived lead in single threaded performance is largely based on clock rate and more aggressive power management. Apart from workloads which make use of AVX512 an such the clock-for-clock and core-for-core performance is quite similar in most tasks.
I think the salient comparison here was Zen+ against Intel 8th gen Core parts, and the IPC difference at iso-clock speeds was south of 10%, used to be near 20%. By the time Intel actually updates their u-arch for real, they will be at parity with AMD on IPC
Power consumption. Intel is stuck on 14nm and they've already reached the reasonable per-socket TDP limit with the 28C parts. Now they are sticking two of them together, disabling SMT (note that in their benchmarks they also disabled SMT on the Epyc parts for no appreciable reason apart from the obvious), and disabling 8 cores. I guess they want to stay under 350W TDP, and rely on high-end cooling solutions. They have one benchmark for HPC that they look good in (although perf/W will be interesting versus Epyc 2) even though any serious HPC customer would be looking at dedicated HPC accelerators.
I'm wondering about memory channels too. Each die has two three channel memory controllers, 12 channels in total. Did they deliberately not activate all controllers or they simply couldn't technically do that?
If AMD has 256 cores per CPU at half the price of Intel's 48 core with performance to match Intel still isnt done. That's just not how server sales work.
- Now if AMD had and maintained that lead for 5+ years with performance and reliability on the whole platform Intel would have some serious trouble.
Please explain fully why Intel is "done" on the server front after Epyc has 64 cores when the curent EPYC 32 core chip after 18 months, does not still make up more than 2% of shipped production x86 data center servers world wide!!!
UPI is 10.4GT/s and inter-socket Infinity Fabric is 42.6GT/s. So Infinity Fabric has over 4 times the bandwidth. As for latency it's hard to get solid confirmation on numbers but they appear about equal maybe with a slight advantage to Infinity Fabric.
Intel inter-socket latency is roughly equivalent to the intra-socket latency for AMD EPYC Infinity Fabric. Intel is still doing a bit better in its most complex (with full 3x UPI link direct connection) four NUMA node topology than AMD EPYC is.
GT/s does not take encoding into account. GB/s is always lower than GT/s. If for example Intel uses a 8b/10b encoding the transfer rate is 8.32GB/s. GT/s is the raw transfer rate while GB/s is the actually transferred date rate. I'm not aware of the Intel encoding for UPI so I took the raw numbers.
Wihtout Ryzen, Intel will still be trying to flog quads as more than enough for eveybody and milking the customer for cash. Looking forward to the day AMD is actually on a level playing field and kicking Intel in the cash bags, not that the resident AT fanbois will ever report on such a thing !
Not sure about that. Ryzen didn't make them kill off Xeon Phi - that was GPUs. This seems to be an attempt at a Phi successor, but GPUs (in the form of Turing's new inference-oriented tensor cores) have upped their game as well.
As usual Intel fudging the numbers in their favour by crippling competition while testing. Testing Epyc with Smt off, how much more of this bull@&#& are Shintel going to pull before they are called in by FTC.
Intel is showing signs of panic, and I think their decision-making has been adversely affected. A new socket? Isn't that a bad idea? They always piss off existing customers.
in the enterprise market nobody ever updates the CPU so it is irrelevant if the socket changes, in all my life of hundreds of servers in the datacenter I have never seen a cpu upgrade
You're missing the point. Socket changes mean more work for their partners and less opportunity to reap returns on their previous platform investments.
This x1000. The "customer" of Intel is not the end-user, it's system integrators. New gigantic socket for a comedy CPU = new boards with new layouts, new validation, new user guides and best practices, new diagnostics... plus this thing is gonna need god-knows how many layers in the motherboard's CB to accommodate all those traces to the RAM slots and cover that >300W power draw.
If I were a customer like HP then I'd tell them where to cram it, but then I'd also be getting the full-blooded benefit of their "marketing development funds" so who knows.
How is Intel panicking??? Intel is competing with a better architecture just like they did from 1974 with their first microprocessor 4004. Intel has always competed with better chips and they have nearly always won with some type of x86 derivative. I think it would be foolish to think that they are panicking because they just made another 16% increase in revenue last quarter to...$19B dollars!!!
I think it would be foolish in the extreme to look at their last quarter's results and use that to dismiss what is obviously a panic response to the imminent announcement of a new CPU architecture they will not be able to compete with for at least a year after its release, more likely 18 months.
I remember running a tual intel cpu 800Mhz intel setup back in the day, i ran this program called Crafty on a Chess server against humans, it was the best on the server for humans vs CPU. Man times have changed. It was before Xeons and the likes, i upgraded it from 8 gigs of ram to 16gigs of ram to take advantage over higher chess tables of moves and it jumped 200 points in rankings in a week.
Oh good times.
Now i can't do that anymore cause i can't afford to pay a cars worth salary every upgrade. lol
They did not have to resort to any tricks like that. They didn't need actual silicon - nor a desk - nor in fact anything physical. It's only our naive first impression that anything was *compared* to EPYC.
But at least they were "honest" enough to disclose it all officially in their "Configuration details" slide, which was alas not shown in the Anandtech article.
There, they go for almost 3 lines detailing very precisely AMD EPYC configuration, then they state: (**) compared to 1-node, 2-socket 48-core Cascade Lake Advanced Performance processor projections by Intel as of 10/3/2018.
For Stream Triad: they link actual PDF report by AMD from June 2017 and compare it to 1-node, 2-socket 48-core Cascade Lake Advanced Performance *PROCESSOR PROJECTIONS* by Intel as of 10/3/2018.
Exacly same for DL Inference: 8 lines of detailed descriptions vs. "CL Advanced Performance PROCESSOR PROJECTIONS by Intel as of 10/7/2018"
So they seem to be far from any engineering / laboratory samples.
If that's not a paper launch, I don't know what is.
So, Intel will get 33 Cascade Lake CPUs (~76 dies @ 698mm/2) from a single 300mm wafer (given 0 defects). AMD will get 72 Rome CPUs (286 dies @ 213mm/2) from the same size wafer (same number of defects). If you factor in Murphy’s Low model of Die Yield and Defect density parameter, then Intel will get 15 CPUs (30 good dies), AMD will get 52 CPUS (208 good dies). As you see the yield plummets with die size!! 15 CPUs vs 52? Wow there's a big difference right there. I can't wait to see what the prices of these things are. I used this: https://caly-technologies.com/die-yield-calculator...
Granted we don't know how 7nm performs on the die size vs. yield correlations just yet, that 3:1 die size is gonna kill Intel in volume, where they are already struggling
Using my best estimates for dimensions and what I suspect the edge exclusion will be for inking die, I got 218 viable die for EPYC with a D0=0.1 (industry standard). So pretty close
Any news on Cascade Lake-X? Do we have any guideline as to what core amounts will be in that category? Will 8 core 16 thread still even exist for this generation of HEDT X series processors?
The question I have.... Is Intel really going to force the OEM to offer 2 brands of dual socket systems? (yes i say force as Intel always does that by giving r&d incentives aka money)
We had this before Intel remember? there was the 2 socket system and we had the 4socket which could be loaded with only 2 sockets (different cpu) which had more but slower cores and more memory... as we all know that was not really adopted in server world and was ditched fast by OEM... will we see the same thing again? It is clear that Intel is quite desperate at the moment
Will be interesting to see how this competition shapes up next year. I feel Intel is at a disadvantage since they no longer have the advantage of a more advance fab. With 14nm vs 7nm, Intel needs a new CPU architecture and also 10nm to get ahead.
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.
Facebook
Twitter
RSS
91 Comments
Back to Article
sgeocla - Monday, November 5, 2018 - link
I hope they didn't infringe any glue related patents from AMD.svan1971 - Monday, November 5, 2018 - link
But they cant deliver 8 core cpus 1 month after launch .mode_13h - Monday, November 5, 2018 - link
Lol. Because they're too busy fabbing 28-core dies, silly!bobhumplick - Monday, November 5, 2018 - link
actually you are correct. well not just 28 core dies. but the whole xeon lineup. there are stories of them giving steep discounts to xeon customers who have gotten quotes on epyc systems. everybody is upgrading from their broadwell\haswell era xeons at once and its the main cause of this shortage. intel wont go down without a fight and even though it appears that epyc will wipe the floor with them as the current msrp's those prices arent the real prices. its actually much closer than it would appearSpunjji - Tuesday, November 6, 2018 - link
For Intel, "won't go down without a fight" always seems to involve predatory pricing somewhere.jospoortvliet - Sunday, November 11, 2018 - link
A bigger question is if and when one can actually get these. Besides Amazon and Google and the like I don't see the 24-28 cores in the wild, point me to a vendor selling them/with stock? Will these be any better?IGTrading - Monday, November 5, 2018 - link
AMD EPYC 64 cores max vs Intel Xeon 48 cores max ....I guess Intel uses low quality glue.
Hul8 - Monday, November 5, 2018 - link
Current EPYC designs only go up to 32 cores/package. (4 dies x 2 CCXes x 4 cores.)Martin_Schou - Monday, November 5, 2018 - link
And this design isn't going to be available until some time in 2019. Sure, it makes sense for Intel to compare against the 7601, as that's the best CPU AMD has to offer at the moment, but with Rome expected to be announced at CES 2019 (January), and rumours/speculations pointing towards an 8x Zen2 chiplet design, this is likely what Cascade Lake will end up competing against, even though Rome is supposedly designed to compete against Ice Lake (the generation after Cascade Lake from what I understand).But - since neither of these chips are available, making claims about what is and isn't current design makes little sense.
Revdutchie - Monday, November 5, 2018 - link
AMD is having their "New Horizon" conference tomorrow so I expect Rome to be announced there. That's why this Intel news came out today.The Benjamins - Wednesday, November 7, 2018 - link
Rome has now been announced to be available this year with 64c128tMartin_Schou - Tuesday, November 6, 2018 - link
Hey, look at that - AMD has announced 64 core Rome the very next day. So now Intel's newest architecture really IS behind by 16 cores. Oh, and it has PCIe 4.0 support out of the box.The Benjamins - Wednesday, November 7, 2018 - link
Intel only made this as a stop gap, but their 48c part is going to need a socket 50% bigger then Rome. does not look good for intel.HStewart - Monday, November 5, 2018 - link
Intel has been connecting dies together in a package since the days o Pentium D and just because they did today - one calls this infringe on patents from AMD and also glue, if that was truly the case then AMD has infringe on Intel patentsOne thing I am deeply concern on these patents - if an Intel person says anything about AMD product - they get viciously attack - but then they is article like this where Intel has done something and it get polluted with discussion about AMD.
I have the right to like Intel, and so does users that like AMD, but would success some maturity in the forums and keep this Intel vs AMD stuff out of logical discussions.
sa666666 - Monday, November 5, 2018 - link
This is rich; the resident Intel fanboi extraordinaire complaining about fanboyism. You're the one that constantly brings mention of Intel into AMD articles, as you just can't resist when Intel is being cast in a bad light. And now you want people to stop talking about AMD in an Intel article?Try looking in the mirror and analyzing your own behaviour before you denounce it in others. And to be clear, it's not Intel favouritism from you that is annoying, it's the complete illogic of some of your statements. If you would just preface all your articles with "I am an Intel supporter, always will be, and don't accept that they can do any wrong", then maybe some of your comments would be more palatable. At least you'd be honestly sticking to your opinion without the passive-aggressive attitude.
Talk about the pot calling the kettle black ...
HStewart - Monday, November 5, 2018 - link
I only mention Intel in AMD articles if some one else blabs it out.I don't consider myself Intel Fanboy - I have other products include from AMD - I just prefer Intel, it seems that some uses are AMD Fanboy's - they are consider Intel Fanboy.
I honest did try to give AMD a test with my Dell XPS 15 2in1.
I would take a look in the mirror from AMD statements - at least Anandtech is not as bas WCCFTech or ExtremeTech which are bluntly AMD bias places.
I wrote this message, because I don't like Intel related article that keep pushing AMD stuff - if this was done on AMD article - one would be harass like no tomorrow. Only because of long history of techy geekness that I even care about forums.
But lets be honest, the average consume does not care if Intel has dual connect 24 Cores and AMD core is 32 core ( actually 8 connect 4 cores ) they want something to emails and spreadsheets.
Which to some people might be iPad or Samsung Tab Sx tablet - which is actually Intel and AMD's bigger threat.
PeachNCream - Monday, November 5, 2018 - link
"I don't consider myself Intel Fanboy"Everyone else does.
HStewart - Monday, November 5, 2018 - link
Of course AMD fanboy's believe that - if not AMD you must be Intel Fanboy.I just have 30 years experience developing on Intel machines and trust them
PeachNCream - Tuesday, November 6, 2018 - link
Being a code monkey for the part 30 years hasn't helped your communication skills in the slightest. Beyond that, if you really think that everyone who believes you're an slobbering Intel shill _must_ be diametrically opposed to you and a slobbering AMD shill yet somehow refuse to also acknowledge your blind loyalty, then I'm glad I've never hired you to bang out code for any project I've led because you're truly lost in the sauce.PeachNCream - Tuesday, November 6, 2018 - link
Yikes, your awful English is rubbing off on me. "Being a code monkey for 30 years hasn't helped your communication skills in the slightest."Sheesh, I need a time machine to hop over to Anandtech of the future where we get edit capabilities. Hmm...I guess I'll set the time machine's dial to 1998. That sounds like a reasonable year to expect an edit button.
HStewart - Wednesday, November 7, 2018 - link
My English skills is because of Aspeger's - almost perfect SAT math in high school but was in with Football players for English in college. I believe my mind moves faster than I write. Unfortunate these forums do not allow to edit.In case you will notice that I did not comment on any of AMD announcements today like 64 core stuff - all I wish is AMD fanboy's have the same respect for Intel threads.
MrSpadge - Monday, November 5, 2018 - link
Really, you didn't see that one coming? I can't know your real thoughts, but from what you're writing you come across as very pro-Intel to me. Actually to a point where I stop reading most opf your posts after a few lines and just think "oh no, he's again explaining why everything in the Intel world is great".In my opinion Intel deserves jokes about this - it's been less than 1,5 years that they harshly criticized AMD for the "just glued together" approach.
HStewart - Monday, November 5, 2018 - link
Maybe more that I dislike AMD instead of Pro Intel. I have absolutely no stock or work for Intel, I just don't like what happen to Intel related to original IBM computer situation. IBM required a second source of products and that is why AMD is out there. If I spent my hard earn time investing in a new product and another company comes along and clones it and take away my product - would you be upset.It not related to their GPU industry, that was from ATI - that does not matter.
Most people on this forum are probably to young to remember Intel and IBM stuff when PC were first created and would not understand.
But I would agree in the situation, that AMD has help the industry - help Intel - by keeping Intel making new products - just like this product but I still prefer the original CPU - vendor and not the clone AMD.
Part of this maybe when I got my first PC - it was 386 PC - It had an AMD clone CPU in it and yes it was a cheaper clone pc - but the make Magitronic had an AMD CPU.
sa666666 - Monday, November 5, 2018 - link
And again, you should just post at the start of your reply: "I am an Intel supporter, always will be, and don't accept that they can do any wrong". Can you not see the rubbish you're posting? Complaining that AMD was a second-source to the original IBM computer! Jesus H Christ, this was over 30 years ago. If you're holding a grudge about a non-event from that long ago, you have major issues.Can you not see how someone would interpret your rantings as that of a lunatic? Why would you care what happened to Intel vs. AMD from over 30 years ago? Were you part of the company and lost a lot of money or something??
Again, can you not see how many people would interpret what you're saying. This isn't pro-Intel or anti-AMD at this point. You seem to be severely delusional about an event that had nothing to do with you personally, and happened over 30 YEARS AGO. Let it go already.
sa666666 - Monday, November 5, 2018 - link
Oh, and the appeal to seniority/authority, having done development for over 30 years, is useless. I also have 30 years experience in dealing with computers and professional development, on everything from an Atari 2600 to the latest mainframes. And probably many other people in these forums do too. I/they just don't feel the need to brag about it, as you seem to.Just because you have experience doesn't make your opinion valid or even sane. And something is not true just because you REALLY WANT IT TO BE TRUE.
Spunjji - Tuesday, November 6, 2018 - link
You really are an artist when it comes to not understanding jokes when that outcome allows you to be pissed off.Seriously though, I don't understand why you're not banned.
FullmetalTitan - Wednesday, November 7, 2018 - link
Look, you really need to start paying out portions of your Intel royalty checks to AT readers for having to put up with this drivelFreckledTrout - Monday, November 5, 2018 - link
Intel disabled SMT on the EPYC chip?HStewart - Monday, November 5, 2018 - link
"Intel disabled SMT on the EPYC chip?"That is news - did Intel purchase AMD - because they don't make EPYC chips
sa666666 - Monday, November 5, 2018 - link
Not sure if this is trolling or you are autistic or something. Obviously this means that in the benchmarks, Intel disabled SMT in the BIOS for the AMD system. To make their system look better, of course.Spunjji - Tuesday, November 6, 2018 - link
He is probably both trolling and thus, by definition, also being an idiot.Lord of the Bored - Monday, November 5, 2018 - link
Yup. Disappointed that it wasn't called out in the article. "We are going to be almost 4x faster if our imaginary numbers prove accurate AND you kneecap the Epyc."It isn't a good look for Intel, but it does show they have some serious concerns about the competition.
philehidiot - Tuesday, November 6, 2018 - link
They also used binaries in the benchmarks specifically compiled for Intel's processors. So they basically raced against AMD. Having removed half the tyres from the AMD vehicle as well as making AMD race on a dirt track whilst Intel were on the tarmac.Seems legit.
Zizy - Monday, November 5, 2018 - link
I expected simpler "glue 2x 28C parts", which would lead to 16 memory channels and a 56C monster with pretty decent number of PCIe as well - not quite as high as on AMD side, but still decent.Why would Intel make such a puzzling chip instead - what is the benefit here? Do they feel AMD isn't THAT much of a threat so they can afford offering something intermediate or what?
Supercomputing 2018 will be interesting :)
sgeocla - Monday, November 5, 2018 - link
It's not that they don't know how much of a threat AMD is, it's just that they need to keep juggling 28 die yields, power draw, sku stack and margins.ManuelDiego - Monday, November 5, 2018 - link
I'm guessing the benefit is winning some time while they solve their 10nm woes and have a really scalable architecture that allows them to compete with Epyc. I mean compete in economic terms, they clearly still have an IPC lead, although a shrinking one, but they need a way to make CPUs with far more than 28 cores at a reasonable price. For the time being they will glue together two XCC cores to at least remain competitive agains the coming Zen 2 Epyc CPUs in terms of core counts, while they figure out how to also compete in TCO, or at least to remain reasonably competitive.GruenSein - Monday, November 5, 2018 - link
I'd argue that the perceived lead in single threaded performance is largely based on clock rate and more aggressive power management. Apart from workloads which make use of AVX512 an such the clock-for-clock and core-for-core performance is quite similar in most tasks.FullmetalTitan - Wednesday, November 7, 2018 - link
I think the salient comparison here was Zen+ against Intel 8th gen Core parts, and the IPC difference at iso-clock speeds was south of 10%, used to be near 20%. By the time Intel actually updates their u-arch for real, they will be at parity with AMD on IPCpsychobriggsy - Monday, November 5, 2018 - link
Power consumption. Intel is stuck on 14nm and they've already reached the reasonable per-socket TDP limit with the 28C parts. Now they are sticking two of them together, disabling SMT (note that in their benchmarks they also disabled SMT on the Epyc parts for no appreciable reason apart from the obvious), and disabling 8 cores. I guess they want to stay under 350W TDP, and rely on high-end cooling solutions. They have one benchmark for HPC that they look good in (although perf/W will be interesting versus Epyc 2) even though any serious HPC customer would be looking at dedicated HPC accelerators.Spunjji - Tuesday, November 6, 2018 - link
This is what really gets me. Their biggest "win" is in an area where you'd be an absolute idiot to use a CPU for your primary processing requirements.What one can interpret from this is that they feel like they're up against a wall, big time.
mode_13h - Monday, November 5, 2018 - link
16 memory channels, only if you count the 4 Optanes. But I think it's only 12 channels of DDR4.eddman - Monday, November 5, 2018 - link
I'm wondering about memory channels too. Each die has two three channel memory controllers, 12 channels in total. Did they deliberately not activate all controllers or they simply couldn't technically do that?eddman - Monday, November 5, 2018 - link
Scratch that. I somehow missed the first memory part and misread the 4S system part.eddman - Monday, November 5, 2018 - link
Scratch that. I somehow missed the first memory part and misread the 4S system example.jjj - Monday, November 5, 2018 - link
Hey, they found the glue!SaturnusDK - Monday, November 5, 2018 - link
Is Intel sniffing glue? I know it's a last ditch attempt at staying relevant before the EPYC 2 reveal tomorrow but even then this is a poor attempt.yeeeeman - Monday, November 5, 2018 - link
If AMD Rome has 64 cores per CPU, then Intel is done on the server front. PeriodDigitalFreak - Monday, November 5, 2018 - link
I love these comments. Even if AMD's Rome blows the 48c Cascade Lake out of the water, Intel is far from "done" on any front.goatfajitas - Monday, November 5, 2018 - link
If AMD has 256 cores per CPU at half the price of Intel's 48 core with performance to match Intel still isnt done. That's just not how server sales work.- Now if AMD had and maintained that lead for 5+ years with performance and reliability on the whole platform Intel would have some serious trouble.
Icehawk - Monday, November 5, 2018 - link
Yup, my director won’t even consider a test Epyc box - “not proven”.fallaha56 - Monday, November 5, 2018 - link
or they own Intel shares? Rome is going to be beyond compelling in terms of price / performanceplus comes without your monthly performance-losing, customer-secrets-losing security holes
rscsr90 - Monday, November 5, 2018 - link
plus rome will be UMA.lemans24 - Monday, November 5, 2018 - link
Please explain fully why Intel is "done" on the server front after Epyc has 64 cores when the curent EPYC 32 core chip after 18 months, does not still make up more than 2% of shipped production x86 data center servers world wide!!!Spunjji - Tuesday, November 6, 2018 - link
Massive overstatement. Server sales do not work that way.eddman - Monday, November 5, 2018 - link
How does UPI compare to infinity fabric?SaturnusDK - Monday, November 5, 2018 - link
UPI is 10.4GT/s and inter-socket Infinity Fabric is 42.6GT/s. So Infinity Fabric has over 4 times the bandwidth. As for latency it's hard to get solid confirmation on numbers but they appear about equal maybe with a slight advantage to Infinity Fabric.SaturnusDK - Monday, November 5, 2018 - link
Correction. The intersocket infinity fabric bandwidth is 37.9GB/s, it's the intrasocket bandwidth that is 42.6GB/s.coschizza - Monday, November 5, 2018 - link
you are completely wrongGT/s is not GB/s
Intel inter-socket latency is roughly equivalent to the intra-socket latency for AMD EPYC Infinity Fabric. Intel is still doing a bit better in its most complex (with full 3x UPI link direct connection) four NUMA node topology than AMD EPYC is.
SaturnusDK - Monday, November 5, 2018 - link
GT/s does not take encoding into account. GB/s is always lower than GT/s. If for example Intel uses a 8b/10b encoding the transfer rate is 8.32GB/s. GT/s is the raw transfer rate while GB/s is the actually transferred date rate. I'm not aware of the Intel encoding for UPI so I took the raw numbers.coschizza - Monday, November 5, 2018 - link
"GB/s is always lower than GT/s." ????????????????example
QPI at 6.4GT/s transfer 25.6GB/s,
coschizza - Monday, November 5, 2018 - link
examplePCI Express 3 have 8.0 GT/s and a x16 line have 31.5 GB/s
coschizza - Monday, November 5, 2018 - link
Intel UPI at 10.4 GT/s have a speed of 41,6 GB/s like a intrasocket bandwidth of IF.mode_13h - Monday, November 5, 2018 - link
And still not nearly enough if app isn't NUMA-aware.Wolfclaw - Monday, November 5, 2018 - link
Wihtout Ryzen, Intel will still be trying to flog quads as more than enough for eveybody and milking the customer for cash. Looking forward to the day AMD is actually on a level playing field and kicking Intel in the cash bags, not that the resident AT fanbois will ever report on such a thing !mode_13h - Monday, November 5, 2018 - link
Not sure about that. Ryzen didn't make them kill off Xeon Phi - that was GPUs. This seems to be an attempt at a Phi successor, but GPUs (in the form of Turing's new inference-oriented tensor cores) have upped their game as well.Chaitanya - Monday, November 5, 2018 - link
As usual Intel fudging the numbers in their favour by crippling competition while testing. Testing Epyc with Smt off, how much more of this bull@&#& are Shintel going to pull before they are called in by FTC.Taiki - Monday, November 5, 2018 - link
14 nm is basically two 7nm glued together, right?shabby - Monday, November 5, 2018 - link
Mind blown...GreenReaper - Monday, November 5, 2018 - link
Nah, it's four of them, like the pictures! Well, until you start gluin' extra cores on top . . .colonelclaw - Monday, November 5, 2018 - link
Intel is showing signs of panic, and I think their decision-making has been adversely affected. A new socket? Isn't that a bad idea? They always piss off existing customers.coschizza - Monday, November 5, 2018 - link
in the enterprise market nobody ever updates the CPU so it is irrelevant if the socket changes, in all my life of hundreds of servers in the datacenter I have never seen a cpu upgrademode_13h - Monday, November 5, 2018 - link
You're missing the point. Socket changes mean more work for their partners and less opportunity to reap returns on their previous platform investments.Spunjji - Tuesday, November 6, 2018 - link
This x1000. The "customer" of Intel is not the end-user, it's system integrators. New gigantic socket for a comedy CPU = new boards with new layouts, new validation, new user guides and best practices, new diagnostics... plus this thing is gonna need god-knows how many layers in the motherboard's CB to accommodate all those traces to the RAM slots and cover that >300W power draw.If I were a customer like HP then I'd tell them where to cram it, but then I'd also be getting the full-blooded benefit of their "marketing development funds" so who knows.
iwod - Monday, November 5, 2018 - link
>Ahead of the annual Supercomputing 2018 conference next weekOr it should be Ahead of the AMD Next Horizon conference tomorrow.
nandnandnand - Monday, November 5, 2018 - link
You got that right. And there's nothing about AMD's conference in this article even though we are 90% sure they will announce 64-core Epyc tomorrow.benzosaurus - Monday, November 5, 2018 - link
Been a while since we’ve seen Intel friggin panic.I wonder how many team leads associated with their 10nm process are going to get fired the day after it finally ships.
lemans24 - Monday, November 5, 2018 - link
How is Intel panicking??? Intel is competing with a better architecture just like they did from 1974 with their first microprocessor 4004. Intel has always competed with better chips and they have nearly always won with some type of x86 derivative.I think it would be foolish to think that they are panicking because they just made another 16% increase in revenue last quarter to...$19B dollars!!!
Spunjji - Tuesday, November 6, 2018 - link
I think it would be foolish in the extreme to look at their last quarter's results and use that to dismiss what is obviously a panic response to the imminent announcement of a new CPU architecture they will not be able to compete with for at least a year after its release, more likely 18 months.imaheadcase - Monday, November 5, 2018 - link
I remember running a tual intel cpu 800Mhz intel setup back in the day, i ran this program called Crafty on a Chess server against humans, it was the best on the server for humans vs CPU. Man times have changed. It was before Xeons and the likes, i upgraded it from 8 gigs of ram to 16gigs of ram to take advantage over higher chess tables of moves and it jumped 200 points in rankings in a week.Oh good times.
Now i can't do that anymore cause i can't afford to pay a cars worth salary every upgrade. lol
Atari2600 - Monday, November 5, 2018 - link
"No information was given about TDP per processor, or pricing, PCIe lanes, memory capacity/support, frequencies, or variants."Did you "reach out" to Intel and ask had they an industrial water cooler under the desk* when they generated the performance comparisons to EPYC?
*That they conveniently forgot to mention - whoops - won't do it again - honest!
dlum - Monday, November 5, 2018 - link
They did not have to resort to any tricks like that.They didn't need actual silicon - nor a desk - nor in fact anything physical.
It's only our naive first impression that anything was *compared* to EPYC.
But at least they were "honest" enough to disclose it all officially in their "Configuration details" slide, which was alas not shown in the Anandtech article.
There, they go for almost 3 lines detailing very precisely AMD EPYC configuration, then they state:
(**) compared to 1-node, 2-socket 48-core Cascade Lake Advanced Performance processor projections by Intel as of 10/3/2018.
For Stream Triad: they link actual PDF report by AMD from June 2017 and compare it to 1-node, 2-socket 48-core Cascade Lake Advanced Performance *PROCESSOR PROJECTIONS* by Intel as of 10/3/2018.
Exacly same for DL Inference: 8 lines of detailed descriptions vs. "CL Advanced Performance PROCESSOR PROJECTIONS by Intel as of 10/7/2018"
So they seem to be far from any engineering / laboratory samples.
If that's not a paper launch, I don't know what is.
dlum - Monday, November 5, 2018 - link
A sample link with Intel Configuration Details slide you can see e.g. here:http://cdn.benchmark.pl/uploads/backend_img/c/news...
Intel999 - Monday, November 5, 2018 - link
Seems that in order to keep TDP down this 48 core kneejerk reaction will not have hyperthreading.Now we know why the 9700K lacks hyperthreading. They are the failed dice from the CLAP.
lemans24 - Monday, November 5, 2018 - link
Why don't you wait until we get complete details about the chip architecture before dissing why there is no hyperthreading!!!Spunjji - Tuesday, November 6, 2018 - link
Why don't Intel wait until they have actual CPUs before publishing "performance comparisons"?Oh, yeah. Panic.
Kango_V - Monday, November 5, 2018 - link
So, Intel will get 33 Cascade Lake CPUs (~76 dies @ 698mm/2) from a single 300mm wafer (given 0 defects). AMD will get 72 Rome CPUs (286 dies @ 213mm/2) from the same size wafer (same number of defects). If you factor in Murphy’s Low model of Die Yield and Defect density parameter, then Intel will get 15 CPUs (30 good dies), AMD will get 52 CPUS (208 good dies). As you see the yield plummets with die size!! 15 CPUs vs 52? Wow there's a big difference right there. I can't wait to see what the prices of these things are.I used this: https://caly-technologies.com/die-yield-calculator...
FullmetalTitan - Wednesday, November 7, 2018 - link
Granted we don't know how 7nm performs on the die size vs. yield correlations just yet, that 3:1 die size is gonna kill Intel in volume, where they are already strugglingFullmetalTitan - Wednesday, November 7, 2018 - link
Using my best estimates for dimensions and what I suspect the edge exclusion will be for inking die, I got 218 viable die for EPYC with a D0=0.1 (industry standard).So pretty close
Endelite - Monday, November 5, 2018 - link
Any news on Cascade Lake-X? Do we have any guideline as to what core amounts will be in that category? Will 8 core 16 thread still even exist for this generation of HEDT X series processors?duploxxx - Monday, November 5, 2018 - link
The question I have.... Is Intel really going to force the OEM to offer 2 brands of dual socket systems? (yes i say force as Intel always does that by giving r&d incentives aka money)We had this before Intel remember? there was the 2 socket system and we had the 4socket which could be loaded with only 2 sockets (different cpu) which had more but slower cores and more memory... as we all know that was not really adopted in server world and was ditched fast by OEM... will we see the same thing again? It is clear that Intel is quite desperate at the moment
watzupken - Tuesday, November 6, 2018 - link
Will be interesting to see how this competition shapes up next year. I feel Intel is at a disadvantage since they no longer have the advantage of a more advance fab. With 14nm vs 7nm, Intel needs a new CPU architecture and also 10nm to get ahead.