Intel Xeon Gold & Xeon Platinum (Skylake-SP) Lineup Leaked

Status
Not open for further replies.
Rumor says Skylake-SP is leading 32C Naples even in CB R15, where Zen has an unusual advantage against Broadwell-E and other Intel designs:

https://www.chiphell.com/thread-1737523-1-1.html

It will be massacre in AVX workloads, but we already knew that.

Zen performed better than Broadwell on CB15 thanks to a combination of 512KB of L2 cache plus higher stock clocks. Skylake-Xeon outperforms Naples thanks to the 1MB of L2 and the higher clocks.

A note: the 180W is a marketing label. The real TDP is >200W.
 
They've known AMD would be forcing core dense designs for some time, years really. I doubt you'll see Intel ceding even a single percent of the server market willingly. Should be good stuff. Hopefully a little price adjusting here and there.
 
Good to see these chips finally performing at (nearly) their full potential...I wonder if there'll be any workstation-esque boards that use Socket 3647 (that also come with external PCI-E clock generators)....;)

Would also be nice to see if these chips support Turbo Boost Max 3.0 and can hold a few cores at 3.8GHz, even when the rest of the cores load up and drop to 3.2GHz...
 
Yowza. I thought I've been rocking pretty good with dual E5-2643 V4's. Those 6134's would be badass crunchers and would still rock low thread count stuff nicely too!
 
I really want to know what the M models are and what's different with them. Not some guess or opinion but a provable explanation. Took a look at the Cinebench results from both the 8173M and the 8176 and the 8173M beats it in spite of lower base and turbo clocks.
 
So now that we have the specs we can compare Intel/AMD's best offerings:

- Xeon Platinum 8180: 28C/56T = Native Skylake-SP HCC die clocked at 2.5 GHz / 3.2 GHz / 3.8 GHz (base/all-core Turbo/max Turbo) + AVX-512 support
- EPYC 7601: 32C/64T = 4x Zeppelin dies in MCM configuration clocked at 2.2 GHz / ??? / 3.2 GHz (base/all-core Turbo/max Turbo)
 
Also one is rated 205W (including AVX512) whereas the other is rated at 180W (real >200W) without AVX512 support, which implies we will see on server space something similar to what we saw on desktop space: how a 140W BDW-E chip was more efficient than a '95W' RyZen chip.
 
I said before Naples is the biggest flop AMD have yet produced for the server. And now you got clocks and TDP as well. SKL-SP just utterly destroys it.
 
I said before Naples is the biggest flop AMD have yet produced for the server. And now you got clocks and TDP as well. SKL-SP just utterly destroys it.
Well, you must feel pretty smug now, considering that this quad MCM uncore really fucked Naples clocks.
 
If you think clocks are the only issue you are in for another surprise.
Well, if those SPEC ints are to be believed it is not particularly bad behaving in the case when you can just split workload on a bunch of VMs... Ah, right, virtualization and Ryzen, never mind.
 
Well, if those SPEC ints are to be believed it is not particularly bad behaving in the case when you can just split workload on a bunch of VMs... Ah, right, virtualization and Ryzen, never mind.

Bulldozer did quite well in the same, take a wild guess on why. And its specint_rate. Note the last part.

Another hint is that SKP-SP does incredible well vs Broadwell-EP in rate too.
 
*pats server*
2P Intel Gold 6150 on Lewisburg

Time to get aquinted with socket 3647...none-volatile memory looks like it will make an serious push this time ^^
I wonder if this turbo from 2.7 GHz to 3.7 GHz :D
 
It's fast, the 2P Intel 6150 Gold system beat the 4P Intel Xeon E7-8880v4 I tested earlier:

2P Intel 6150 Blender Ryzen benchmark:
6150.png


"Old" 4P Intel Xeon E7-8880v4:

i78880v4.jpg
 
Last edited:
Yup....we kinda looked at time, then each other...and then we dug up the old test....looked at each other...and then we laughed.
People saying the CPU progression has stagnated don't have a clue ;)

Peolle who are saying that CPU progression has stagnated are usually referring to the consumer processors and for good reason. I feel so good for you that Intel is diligently meeting your needs and aggressively innovating to make the products that you use (in the datacenter) better. They haven't been quite so diligent with their consumer chips, which is why people who primarily use products from that market segment aren't quite so ebullient as yourself on the topic.
 
Peolle who are saying that CPU progression has stagnated are usually referring to the consumer processors and for good reason. I feel so good for you that Intel is diligently meeting your needs and aggressively innovating to make the products that you use (in the datacenter) better. They haven't been quite so diligent with their consumer chips, which is why people who primarily use products from that market segment aren't quite so ebullient as yourself on the topic.

Perhaps because people in the consumer market expect others to pay for their workstation/server features (like ECC)...Intel is a company, their are in business to make money...and if you base your "CPU progression has stalled" on a single metric....it says it all ;)
 
Perhaps because people in the consumer market expect others to pay for their workstation/server features (like ECC)...Intel is a company, their are in business to make money...and if you base your "CPU progression has stalled" on a single metric....it says it all ;)

I'm not talking about ECC and it wasn't even on my mind. But since you brought up the topic, answer the following question.

How much should an HEDT CPU cost before it's "worthy" of ECC support??

Now since we are now past the unfortunate and unplanned diversion that you for whatever reason insisted on taking, I was referring to the endless stream of quad cores for the last decade and the glacial increases in HEDT core count as a stark indication of the lack of progression that the consumer market has had to deal with. Of course, you as a datacenter user (who is spending your employer's money by the way) don't see a problem with this...since your core counts increase by leaps and bounds with each new release.
 
  • Like
Reactions: isp
like this
Peolle who are saying that CPU progression has stagnated are usually referring to the consumer processors and for good reason. I feel so good for you that Intel is diligently meeting your needs and aggressively innovating to make the products that you use (in the datacenter) better. They haven't been quite so diligent with their consumer chips, which is why people who primarily use products from that market segment aren't quite so ebullient as yourself on the topic.

By consumer you mean desktop and not mobile? Mobile progress have been amazing!
 
By consumer you mean desktop and not mobile? Mobile progress have been amazing!

I agree. Progress on the mobile front has been much better than that experienced by desktops due to a different set of metrics. Laptops value battery life and simplification, with as strong of an IGP as possible and each of these metrics has seen considerable improvement. The only segment that could use more improvement is the desktop replacement/ultra high performance laptops, as their metrics more closely match those of regular desktops. A six core version of the mobile chips with no IGP would have been ideal to see a few years ago for these >10 lb machines (or one with an IGP that shuts off two cores when the IGP is in use).
 
Peolle who are saying that CPU progression has stagnated are usually referring to the consumer processors and for good reason. I feel so good for you that Intel is diligently meeting your needs and aggressively innovating to make the products that you use (in the datacenter) better. They haven't been quite so diligent with their consumer chips, which is why people who primarily use products from that market segment aren't quite so ebullient as yourself on the topic.

This. I've said the same thing myself, disregarding the fact that I work with Intel Xeon systems with increasing capabilities and core counts in my other job all the time. Intel's 22c/44t CPU's aren't even on the radar of someone who's considering a Core i7 7700K or an AMD Ryzen 1700X in their next build. On the consumer side, we don't have much to show for CPU progression. In the realm of server hardware, things are a lot different as they always have been.
 
One has to wonder how their 16C server parts max out at 2.4-2.9 GHz @155-170W and TR is supposed to clock at 3.4-3.8 GHz at the same TDP.

Inverse binning? The best dies are being selected for desktop, whereas the worse dies are for servers. :p

Now seriously. Most desktop users will not care by AMD selling them a '155W' chip whose real TDP is 220W, and the usual reviews will ignore the discrepancy between marketing TDP and real TDP (only some few reviews of RyZen did care for instance). On the other hand, server users cannot be cheated so easy.
 
Inverse binning? The best dies are being selected for desktop, whereas the worse dies are for servers. :p

Now seriously. Most desktop users will not care by AMD selling them a '155W' chip whose real TDP is 220W, and the usual reviews will ignore the discrepancy between marketing TDP and real TDP (only some few reviews of RyZen did care for instance). On the other hand, server users cannot be cheated so easy.

In a datacenter, we have no use for servers running out of spec...it will mess up the power load and cooling bigtime.
Any vendor trying to sell a 220W TDP servers as a 155W TDP server will not get any return customers.
Only consumers fall for that.
 
I am still debating the whole skylake-sp vs skylake-x thing, for 'intermediate' users like myself.
---
I know all I'm waiting on is final specs for the 12+ core skylake-x stuff, and for motherboards to start coming out for real.
I still think I want 80+ pcie lanes.

/me slaps Supermicro.
 
Processor Arithmetic
2x AMD EPYC 7601 32-Core Processor (4N 32C 64T 3.2GHz, 1.33GHz IMC, 32x 512kB L2, 8x 8MB L3) = 706.18GOPS
2x Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz (28C 56T 3.8GHz, 2.4GHz IMC, 28x 1MB L2, 38.5MB L3) = 1425.82GOPS


Processor Multi-Media
2x Intel(R) Xeon(R) Platinum 8180 CPU @ 2.50GHz (28C 56T 3.8GHz, 2.4GHz IMC, 28x 1MB L2, 38.5MB L3) 5989.90Mpix/s
2x AMD EPYC 7601 32-Core Processor (4N 32C 64T 2.7GHz, 1.33GHz IMC, 32x 512kB L2, 8x 8MB L3) 974.33Mpix/s

http://ranker.sisoftware.net/show_d...f5c8f8deb68bbb9de5d8e8ceabcef3c3e596ab93&l=en
 
Heise said:
Cinebench R15 has changed to Windows Server 2016 [-] is also running. The Cinebech value first swayed between quite a decent 5400 to 6000, but AMD technicians were still working on the configuration and ultimately values came to 6879, about 1300 more than the Xeon E5-2699Av4.

https://www.heise.de/newsticker/meldung/AMD-Epyc-legt-los-3748615.html

So best case scenario in CB where Zen is known to match/beat Skylake MT IPC (and doesn't touch AVX to expose its weakness), top bin EPYC still below Xeon Platinum 8180 - which scores 8200-8300 pts.
 
Last edited:
Status
Not open for further replies.
Back
Top