Anandtech：Intel's Skylake-SP Xeon VS AMD's EPYC 7000

[csbin]

With the exception of database software and vectorizable HPC code, AMD's EPYC 7601 ($4200) offers slightly less or slightly better performance than Intel's Xeon 8176 ($8000+).

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade

----------------------------------------------------------

 

[Edrick]

Those prices though!

Planned on buying a Gold 6000 series, until today. Now I may have to settle for a Silver 4110.

 

[kawi6rr]

Good article good read, glad AMD is finally back in the game.

 

[IEC]

Oh my... that memory bandwidth...

 

[insertcarehere]

Excuse me while I recover from the headache caused by Intel's SKU naming scheme...

 

[Rifter]

Yeah intel is going to have some price cuts coming thats for sure.

 

[LTC8K6]

Is this comment relevant?

http://www.anandtech.com/comments/1...amd-epyc-7000-cpu-battle-of-the-decade/567356

 

[LTC8K6]

Why was the 8180 not used?

EDIT:

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.

 

[dnavas]

Seems like the only big miss for AMD is the high latency, distributed L3, which looks like it's hurting transactional databases. Pretty impressive show, I hope they sell plenty.

 

[HurleyBird]

Keep in mind that the EPYC system is running at DDR4-2400 vs. 2666 on the Skylake-EP platform.

Why was the 8180 not used?

The reason given was that they wanted to match TDPs, which makes sense given the market.

 

[moinmoin]

What an incredible bad timing for Intel to come up with such an inane naming scheme and market segmentation that even causes additional confusion (Gold being split between 6100 and 5100, the latter being way too close to Silver, with FMA being cut in half etc.).

 

[raghu78]

With a USD 4200 price EPYC 7601 is going to take significant amount of market share. The key here is with a small die of around 200 sq mm and yields of 8 core enabled dies above 80% AMD will sell a lot of 32C/64T EPYC 7601 CPUs. Performance in real world workloads like e-commerce web servers is impressive as revealed by SPECjbb.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/19

Performance in Big data is competitive.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/20

FP performance is well mighty impressive .NAMD with AVX optimizations ran 20% faster on EPYC 7601 than Xeon 8176.

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/21

 

[LTC8K6]

The reason given was that they wanted to match TDPs, which makes sense given the market.

Yes, I posted that before you did.

 

[TheGiant]

The NAMD molecular dynamics showed pretty much no improvement from 22C Broadwell to 28C Skylake with faster RAM. Why is that?

Looks to me like a software issue. I expected it to be a little faster than the EPYC system....

 

[insertcarehere]

Given the pricing of the Skylake-SPs, it appears that they have given on max-perf 1P systems....AMD is gonna kill it in that segment...

 

[Tuna-Fish]

I really want to see a web serving benchmark of some kind of a lamp stack running on virtual machines on these. Based on the SPECjbb results, I'd expect Naples to be very competitive there, but there have been surprises caused by subpar virtualization implementations before.

Because if these CPUs do that task as well as the Java results hint they might, AMD might just sell the entire output of GloFo as servers...

 

[Burpo]

Tom's review of the Xeon 8176
http://www.tomshardware.com/reviews/intel-xeon-platinum-8176-scalable-cpu,5120.html

 

[Markfw]

Tom's review of the Xeon 8176
http://www.tomshardware.com/reviews/intel-xeon-platinum-8176-scalable-cpu,5120.html

That review is BS IMO, its only benchmarked against older Xeon processors, how lame is that !

 

[Yakk]

AMD delivered here, impressive to see AMD going from non-existent in the high-end server space to top tier in 1 generation.

 

[flash-gordon]

Intel marketing slides and all this "scalable" talk says it all, they gave up the 1S market for now.

On a side note, how tuff is for the reviewers to criticize the Xeons...

 

[jpiniero]

That review is BS IMO, its only benchmarked against older Xeon processors, how lame is that !

They'll probably do an epyc article later. Double the articles, double the clicks!

Intel marketing slides and all this "scalable" talk says it all, they gave up the 1S market for now.

1S is pretty much workstations. Servers are more or less 2S. You do have things like Xeon D though; but that's much lower power and footprint than either Epyc or the typical Xeons.

 

[rchunter]

Yeah for home use I prefer a high core count single socket server. Sure I could probably only run one cpu in a new purley board but what will most likely happen is I will end up just buying amd epyc. Just depends on what the boards turn out to be..as long as I can buy a decent epyc supermicro board with ipmi and 10gb ethernet I will go that route.

 

[raghu78]

1S is pretty much workstations. Servers are more or less 2S. You do have things like Xeon D though; but that's much lower power and footprint than either Epyc or the typical Xeons.

The problem with 1S and 2S in the past was artificial segmentation. Intel reserved high core counts for the higher priced SKUs which were only sold in 2S servers. AMD is exploiting this artificial segmentation and providing customers the option to buy the highest core counts in 1S servers.

Look at similarly priced server CPUs like EPYC 7551P and 2 x Intel's Xeon Silver 4116

http://ark.intel.com/products/120481/Intel-Xeon-Silver-4116-Processor-16_5M-Cache-2_10-GHz

Price - USD 1000 x 2 , TDP - 85W x 2 = 190w

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/3

7551P Price USD - 2100 TDP - 180w

AMD EPYC 1S 7551P provides 33% more cores than 2 x Intel Xeon Silver 4116 at the same CPU cost. The EPYC 1S will cost lesser due to a smaller server chassis due to a much smaller and less complex mainboard and lower overall server power requirements. Intel will concede a lot of market share with their current pricing. Even in 2S servers EPYC 32C/64T 7601 is competitive with Intel 8176 as seen from anandtech 2S benchmarks. AMD has the 1S/2S market very well covered.

 

[krumme]

Interesting review. Intels new mesh looks very very strong and future proof but their wide fp is a disaster because of efficiency not only impacting the major part of fp loads but also in mixed workloads.
Look at power consumption for the work done. Crap. Its a burden. Imo they need to alter their strategy and eg have different cores here.
The irony is ryzen looks like its better core and Intel have better uncore.
Add production cost. Good for Intel they hopefully are masters of gigantic dies.
What Intel have going for them is the more predictable performance and thereby also ability to use in mix of situations.
I get safety and flexibility as a buyer. I guess that matters more for smaller players.

 

[krumme]

Now how is amd going to improve on that performance of the distributed l3?
Its a disadvantage for them and shows the backside of the design.
A way to look at it is simply. They cant. But so what. Where that matters its a lost market.
Imo the best strategy forward is to entrench and build on your strenght. Not try to make the ocean boil.
7nm high perf with just tuned core and uncore and they are in a very good position. It looks balanced as it is and can do. No need to invent a new bd.