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

Status
Not open for further replies.
So confirmed results.

In CB15 MT 2P.

EPYC 7601 32C - 6879, but its questionable since the result first was between 5600 and 6000.
Xeon 8168 24C - 7212.
Xeon 8180 28C - 8301.
 
I was wondering where the tea party was at, I guess I stumbled upon it. only in a Intel thread could the three stooges possibly turn this into a masters of the universe feel good time.

But please, don't let me disturb your discussion on how a vastly more expensive Intel part beats an AMD part in SpecINT.

This is epyc, the amount of fanboism is off the charts
 
Don't know about the tea party, but I know someone registered in another forum to publicly bash Juan and praise one of the biggest AMD shills, informal (aka inf64). You know something is (mentally) wrong when you let the emotions take over and offend other people because of PC hardware. Funny that both of these stooges decided to buy a Haswell system at the end of the day.
 
-Sweeper_ said:
Don't know about the tea party, but I know someone registered in another forum to publicly bash Juan and praise a well-known AMD shill, informal (aka inf64). You know something is (mentally) wrong when you let me emotions take over and offend other people because of PC hardware. Funny that both of these stooges decided to buy a Haswell system at the end of the day.
Click to expand...

Haha, spot on :D

I guess he cant argue for his case since he goes after people instead. Seems to be a daily thing now.
 
Only 6174 CPUs. Over a Tflop sustained in Linpack per CPU :D

And those 24 cores is about what 64 EPYC cores does (2P 7601).
 
Shintai said:
Only 6174 CPUs. Over a Tflop sustained in Linpack per CPU :D

And those 24 cores is about what 64 EPYC cores does (2P 7601).
Click to expand...

Yes, compare with the only-CPU Broadwell system (#11), I get each Skylake core is providing about 44% more performance than Broadwell core. There is also a discrepancy between the peak and max gaps. I guess it is due to the Skylake cluster not being still fully optimized.
 
I mean that the Broadwell-based system is able to achieve 87% of peak performance, whereas the Skylake-based system only gets 63%. I guess this points to inferior optimization level on this cluster due to the platform being new. Assuming same level of optimization as the Broadwell-system the Skylake system would hits #8.

Edit: the Skylake cluster is unfinished. A 11PFLOPS update is in the works. The updated cluster would hit #8 without problems.
 
Last edited:
Slide deck is out people. Launch tomorrow!

Xeon-Scaleable-Processor-5.jpg


Xeon-Scaleable-Processor-4.jpg


https://videocardz.com/70874/intels-epyc-response-xeon-scalable-processor-skylake-sp

As usual, BitsandChips wrong... These guys have no clue about Intel/NVIDIA products.
 
After shipping to the big boys for ~10 months its now released for everyone :)

Cascade Lake on 14nm++ will be fun. And same situation with that again, already out there ;)
 
AT doesn't even try to be objective anymore. I would expect a so called journalist/writer for one of the most important tech websites to be at least somewhat less biased (looking at you Johan De Gellas).

AnandTech said:
For the EPYC launch, AMD sent us their best SKU: the EPYC 7601. Meanwhile Intel gave us a choice between the top bin Xeon 8180 and the Xeon 8176. Considering that the latter had 165-173W TDP, similar to AMD's best EPYC, we felt that the Xeon 8176 was the best choice.
Click to expand...

Sure... :LOL:
 
-Sweeper_ said:
AT doesn't even try to be objective anymore. I would expect a so called journalist/writer for one of the most important tech websites to be at least somewhat less biased (looking at you Johan De Gellas).



Sure... :LOL:
Click to expand...
I dunno, after half a decade out of the running AMD definitely needs some media support to get back going again, and I'm of the opinion the community should help support AMD where it's fair to breathe competition back into the high end.
 
bwang said:
I dunno, after half a decade out of the running AMD definitely needs some media support to get back going again, and I'm of the opinion the community should help support AMD where it's fair to breathe competition back into the high end.
Click to expand...

Therefore if you and another person produce grade A and grade C products for sale --your being better because you did your work right-- you would ask us to ignore your product, because we would support the competition. Is that?
 
-Sweeper_ said:
AT doesn't even try to be objective anymore. I would expect a so called journalist/writer for one of the most important tech websites to be at least somewhat less biased (looking at you Johan De Gellas).



Sure... :LOL:
Click to expand...

Leaving apart the performance and power consumption numbers, the review seems fine. :rolleyes:

I am still awaiting for the RyZen IPC measurements that they promised me via twitter.

 
Am I missing something?

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9QLzAvNjkxNjY4L29yaWdpbmFsL2ltYWdlMDExLnBuZw==


87250v212.png


Why are AT results so much worse than other websites?

They even show a generational regression in terms of performance in POV-Ray while we know Skylake-X beats Broadwell-E here even with less cores...

87248.png


87099.png
 
juanrga said:
Therefore if you and another person produce grade A and grade C products for sale --your being better because you did your work right-- you would ask us to ignore your product, because we would support the competition. Is that?
Click to expand...
I think we can all agree that Epyc is hardly grade-C, with strong wins in general-purpose FP code and compilation, a slight lead in enterprise Java, good price/performance in data mining, and an abysmal loss in transactional databases. Skylake-SP is a much better workstation processor thanks to its aggressive turbo modes and better single-threaded performance, but for scale-out applications Epyc is a solid contender.
 
Ladies and Gentlemen ,
While I appreciate the compare and contrast with AMD's latest offering, this thread is about skylake xeons.Best of the best of the best ! SIR !
I don't give an eff about EPYC. I'm interested in top-of-the-line !!!
[H] remember ?
:p

bwang said:
Skylake-SP is a much better workstation processor thanks to its aggressive turbo modes and better single-threaded performance.
Click to expand...

Yessir, I agree.

This unfortunately is bad for my wallet.I'm back to a 10,000$ mobo/cpus/ram again.

Dammit !
:D
 
-Sweeper_ said:
Am I missing something?

aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS9QLzAvNjkxNjY4L29yaWdpbmFsL2ltYWdlMDExLnBuZw==


87250v212.png


Why are AT results so much worse than other websites?

They even show a generational regression in terms of performance in POV-Ray while we know Skylake-X beats Broadwell-E here even with less cores...

87248.png


87099.png
Click to expand...

Ye that's what happens when you try and cripple a review to please the payers. Even being defeated by their own SKL-X vs Broadwell-E numbers :D
 
bwang said:
I think we can all agree that Epyc is hardly grade-C, with strong wins in general-purpose FP code and compilation, a slight lead in enterprise Java, good price/performance in data mining, and an abysmal loss in transactional databases. Skylake-SP is a much better workstation processor thanks to its aggressive turbo modes and better single-threaded performance, but for scale-out applications Epyc is a solid contender.
Click to expand...

It was only an illustrative example. Said that, AT review numbers are misleading.
 
juanrga said:
It was only an illustrative example. Said that, AT review numbers are misleading.
Click to expand...
I wouldn't say misleading, they posted the numbers they got using their benchmarking methodology. A different version of NAMD, a different testcase, and a different kernel might swing results in Skylake's favor as the match is quite close; similarly, we could see a substantial increase in AMD's transactional database scores with more hand-tuning and a more favorable testcase (see, for example, this).
In the end it boils down to how much tweaking you think is acceptable (or mandatory) for reviews. Desktop users are unlikely to recompile their programs or switch Windows versions just to get that last 10% performance, but workstation users might, and you bet large enterprise users will do everything they can to squeeze every last drop of performance they can get out of their systems.
 
amazing how shintai is still around after another meeh intel launch.

EPYC was nothing to write home about, but the Intel launch was even worst.

intel is basically charging 30-100% more money for 5-10% more raw performance and ~5% better power efficiency over their own last generation product. AMD has a cheaper product with quite decent performance and power profiles.
no matter how Shintai questions any given benchmark, i stand 200% behind AT review conclusion:

AMD IS PUTTING A LOT OF PRESSURE ON INTEL PRICES!.

The next conclusion: EPYC is a better enterprise CPU than Ryzen is a desktop CPU. And Ryzen captured 20% market share. Long live EPYC!
 
geok1ng said:
amazing how shintai is still around after another meeh intel launch.

EPYC was nothing to write home about, but the Intel launch was even worst.

intel is basically charging 30-100% more money for 5-10% more raw performance and ~5% better power efficiency over their own last generation product. AMD has a cheaper product with quite decent performance and power profiles.
no matter how Shintai questions any given benchmark, i stand 200% behind AT review conclusion:

AMD IS PUTTING A LOT OF PRESSURE ON INTEL PRICES!.

The next conclusion: EPYC is a better enterprise CPU than Ryzen is a desktop CPU. And Ryzen captured 20% market share. Long live EPYC!
Click to expand...

They better...because the Xeon Platinum og Gold CPU's are what we and our customers are looking at to purchase.

I think one (1!) of our server vendors has a single AMD server on their roadmap.

To give you a hint of the current uphill battle AMD has:
We have +5000 physical servers...less than 20 are AMD platforms...and they are all on their way out and I have no knowledge about any AMD servers on the way into our datacenters.

This is not a unique situation, this is the reality in most data-centers.
 
Factum said:
They better...because the Xeon Platinum og Gold CPU's are what we and our customers are looking at to purchase.

I have no knowledge about any AMD servers on the way into our datacenters.

.
Click to expand...

Is that due to some specific performance issues for the type of software you use?
 
Basically the kind of writing you would expect from BitsandChips, SemiAccurate or one of AnandTech's Red Team Plus forum shills... but it's actually an article in TPU's front page. Sure, every company publishes PR to shed the best possible light on the products. AMD does the exact same thing (Poor Volta anyone?) and I never saw an article with this tone.

TechPowerUp said:
Ah, the "Glued-together" dies. Let's forget how AMD's Zen cores actually look like they were architected from the get-go for modularity and scaling, which has allowed the company to keep die-sizes to a minimum and yields to a maximum.
Click to expand...

Right, let's completely ignore the fact that one Zeppelin die is 195 mm², while Intel packs 6 cores that will be faster in most CPU tasks and an integrated GPU in a space of 149 mm² with Coffee Lake (124 mm² if you want to compare to current 4C Kaby Lake). And no source for their ''superior yields'' fallacy either, but let me guess, BitsandChips tweets?

TechPowerUp said:
So essentially, AMD has 8 more cores, 16 more threads, delivers 16% more performance than Intel's e5-2699 system and 32% more performance than Intel's "non glued-together" Xeon 8176. AMD's chip does all that while consuming 23% less power than the Xeon e5-2699, and 28% less than the Xeon 8176. Not too shabby. I'll take my CPUs with this kind of glue any day.
Click to expand...

They had to pick the crippled POV-Ray results where AT somehow managed to make a 28C Skylake-SP slower than 22C Broadwell-EP, while other websites show the new Xeon clearly ahead of its predecessor.

https://www.techpowerup.com/235092/...cessors-glued-together-in-official-slide-deck
 
Last edited:
-Sweeper_ said:
They had to pick the obviously crippled POV-Ray results where AT somehow managed to make a 28C Skylake-SP slower than 22C Broadwell-EP, while we know from their own (and other) Skylake-X reviews that a 7820X blows away a much more expensive 6950X with more cores in the same test.

https://www.techpowerup.com/235092/...cessors-glued-together-in-official-slide-deck
Click to expand...

I have no idea how or why AT would "obviously cripple" a POV-ray test result; the POV-ray benchmark is as easy as it gets to run, so short of outright lying the result is what it is.
Results from the LCC parts don't necessarily translate into identical relative performance on the XCC parts; if scaling worked that consistently we wouldn't even need server benchmarks, as we know the Zen core is a 10% or so slower clock per clock than the Skylake core, but Epyc offers 14% more cores than SKL-SP. It is kind of odd that the cache structure of Skylake-X performs very well in raytracing tasks, but on a 2P XCC system the situation seems to be reversed. Worth more investigation, but being in denial about the results is no better than the AMD shills who are whining right now about how the Vega FE benchmarks are unfair.

As geok1ng points out, AMD is a huge value proposition right now; regardless of benchmark tuning is is clear that the 32c parts are competitive with the 28c Skylake parts - we're talking a 10-20% discrepancy in results that currently favor Epyc, not 50-100% in favor of Intel like it was in previous generations. Pricewise, the cheapest 32c Epyc competes against a 20c Skylake. Furthermore, this generation AMD doesn't lose horribly on either platform features or power consumption.
 
Fortunately we don't need to compare to LCC parts.

Hardware.info (POV-Ray 3.7)

Xeon 8180: 153,9 sec (lower is better)
Xeon E5-2699 v4: 221,4 sec

https://be.hardware.info/reviews/74...e-voor-servers-benchmarks-windows-server-2016

AnandTech (POV-Ray 3.7)

87248.png


Massive performance per core regression? These are the results that TPU used in their 'article' and will certainly be quoted by the usual click bait websites later on.
 
Last edited:
geok1ng said:
amazing how shintai is still around after another meeh intel launch.

EPYC was nothing to write home about, but the Intel launch was even worst.

intel is basically charging 30-100% more money for 5-10% more raw performance and ~5% better power efficiency over their own last generation product. AMD has a cheaper product with quite decent performance and power profiles.
no matter how Shintai questions any given benchmark, i stand 200% behind AT review conclusion:

AMD IS PUTTING A LOT OF PRESSURE ON INTEL PRICES!.

The next conclusion: EPYC is a better enterprise CPU than Ryzen is a desktop CPU. And Ryzen captured 20% market share. Long live EPYC!
Click to expand...
I agree. Launch was a big downer.
Lack of competition and a shrinking market have made them completely myopic.
AMD will now steal market share or they will have to lower prices to defend their turf. Let's see which poison they choose. So far it seems like they are willing to concede market share in a shrinking market while trying to diversify into other markets now competing with Xilinx.
 
-Sweeper_ said:
Fortunately we don't need to compare to LCC parts.

Hardware.info (POV-Ray 3.7)

Xeon 8180: 153,9 sec (lower is better)
Xeon E5-2699 v4: 221,4 sec

https://be.hardware.info/reviews/74...e-voor-servers-benchmarks-windows-server-2016

AnandTech (POV-Ray 3.7)

87248.png


Massive performance per core regression? These are the results that TPU used in their 'article' and will certainly be quoted by the usual click bait websites later on.
Click to expand...


Thanks for the links! Intel disappointed according to those slides above and I checked out the others same story in most. I would have expected them to crush Epyc given their reputation for manufacturing advantage.
 
geok1ng said:
amazing how shintai is still around after another meeh intel launch.

EPYC was nothing to write home about, but the Intel launch was even worst.

intel is basically charging 30-100% more money for 5-10% more raw performance and ~5% better power efficiency over their own last generation product. AMD has a cheaper product with quite decent performance and power profiles.
no matter how Shintai questions any given benchmark, i stand 200% behind AT review conclusion:

AMD IS PUTTING A LOT OF PRESSURE ON INTEL PRICES!.

The next conclusion: EPYC is a better enterprise CPU than Ryzen is a desktop CPU. And Ryzen captured 20% market share. Long live EPYC!
Click to expand...

I think exactly the contrary. Zen will sell better in desktop, thanks to fanboys and people that can be easily fooled by the usual biased reviews, whereas enterprise people are more difficult to fool because many test things by themselves.

And the claim that Ryzen captured 20% market share is wrong. RyZen impact is about 5%.
 
-Sweeper_ said:
Right, let's completely ignore the fact that one Zeppelin die is 195 mm²,
Click to expand...

I can warrant you that the die size is bigger than that. Googling a bit you can find a value very close to the real die size.

-Sweeper_ said:
They had to pick the crippled POV-Ray results where AT somehow managed to make a 28C Skylake-SP slower than 22C Broadwell-EP, while other websites show the new Xeon clearly ahead of its predecessor.

https://www.techpowerup.com/235092/...cessors-glued-together-in-official-slide-deck
Click to expand...

AT was able to get both the Skylake Xeons and Broadwell Xeons to run slower than Haswell Xeons. LOL
 
bwang said:
I have no idea how or why AT would "obviously cripple" a POV-ray test result; the POV-ray benchmark is as easy as it gets to run, so short of outright lying the result is what it is.
Click to expand...

Like when they cheated compiler flags in one of their recent reviews and people at RWT pointing how biased was their review.
 
juanrga said:
I think exactly the contrary. Zen will sell better in desktop, thanks to fanboys and people that can be easily fooled by the usual biased reviews, whereas enterprise people are more difficult to fool because many test things by themselves.

And the claim that Ryzen captured 20% market share is wrong. RyZen impact is about 5%.
Click to expand...
Dot you think that 5% is an unattainable market share in the server market?
 
Upon further investigation I think the AT POV-ray results are reported wrong; the review text quotes Epyc as being 16% faster than Skylake-SP, and 249 is definitely more than 16% higher than 188 - maybe the Broadwell and Skylake results were swapped?

juanrga I think Epyc will sell better than Ryzen. Ryzen can't touch Core i9's single-threaded performance, not with a 10% IPC deficit and a 20% clock deficit. Workstations need consistent performance across a variety of applications, and the 4.5GHz overclocked i9's will walk all over a 3.8GHz Ryzen in lightly-threaded applications. Servers, on the other hand, rarely operate at the limits of what a manufacturing process can clock at, instead opting to run at a more efficient operating point. Furthermore, datacenters are built to do one thing and do it well, so versatility is less of an issue - your rendering cluster is not going to be running databases, and your database server certainly won't be running POV-Ray.
 
Status
Not open for further replies.
Back
Top