Intel Presents Its AMD EPYC Server Test Results

[Guest]

Intel has released its internal benchmarking results with AMD's EPYC server platforms.

Intel Presents Its AMD EPYC Server Test Results : Read more

 

[timmy_area51]

that's intel's reward for aMD's threadripper

 

[DerekA_C]

Could you guys imagine what AMD could make with Intel money and resources I mean really they are so small compared to Intel in both finance and fab capabilities, researches engineers and developers are a fraction of Intel and yet AMD still can pack a punch for what they are.

 

[cmvrgr]

We should thank that AMD gave price/performance in all markets. Intel still very expensive for what they are giving.

 

[berezini]

both companies started out with the similar budget? its about what you do with your investment that counts. Thus the difference between them two!

 

[soulg1969]

I've read the AMD employees created Threadripper in their spare time! Think and write about that.

 

[samopa]

[quotemsg=20423019,0,205814]We should thank that AMD gave price/performance in all markets. Intel still very expensive for what they are giving.[/quotemsg]

For that, I thank AMD very much.
PS: I'm an Intel fanboy, all my rigs are Intel's, all my servers are also Intel's

 

[popoooper]

YouHave to read between the lines, the fact that Intel is so desperate to FIND a benchmark where they outperform amd show's how scared they are.

Plus they are saying a 22 core xeon beats a 32 core epyc in "technical computing"? Bull$hit, I'm not sure what benchmark they are using but they must have dug deep to find one that shows that. also that 22 core is 7 thousand dollars.

screw you intel, amd is winning.

 

[Adam_153]

i only have 8c 190x thread ripper , but im not getting no high latencies from die2die and my chip has two and only 4 cores each, 2 dies, and to sepaerate CCXs for both dies, and max latecy i see is like 107ns but thats on like a very large memory caching. 4gb+ not acuall but i forgot the resuts, but intels "INTERNAL" (FAKED) Tests///theyve don this before? and i now it?

 

[Adam_153]

any i bet 100M $$$ that intel isnt even running the Ram @ anything besides 2133Mhz, i bet my life on it

 

[Adam_153]

so the infinity fabric Link would only be operating @ 1067Mhz, mine is @ 1667* RAM is @ DDR4 Quad channel 3333Mhz, beat that intel?? try those RAMS settings assholoes

 

[ZRace]

@Adam: While I have similar suspicions, I'd like to note that (most likely, not that knowlegdable in this field) in a server environment, you can't go overclocking you RAM as you want, because it could cause instabilities, and that's what you want to avoid most in such an environment.

The most I can imagine being reasonable in a server environment is DDR4-2666, with DDR4-2400 being the standard choice.

 

[ZolaIII]

Intel cheated as usual. Intels properly compiler isn't something that most clients will use anyway & compared to open source ones (GGC, VLLM) which they will it gives some 20% performance boost. This actually EPYC right back into competition.
Demonstration:
https://3s81si1s5ygj3mzby34dq6qf-wpengine.netdna-ssl.com/wp-content/uploads/2017/11/cavium-vulcan-thunderx2-skylake-xeon-spec.jpg
All of this ain't all that important as ARM server's with NUMA support are coming to finally brick X86.
https://www.nextplatform.com/2017/11/27/cavium-truly-contender-one-two-arm-server-punch/amp/
& with a little help of the good old vet's this time around Intel won't have even a compiler advantage to drag about.
http://www.cavium.com/newsevents-Cray-Catapults-Arm-Based-Processors-Into-Supercomputing.html
Fun part that makes this even nostalgic for me is after long time Intel monopoly we are back where & how it all started.

 

[merikafyeah]

[quotemsg=20423245,0,2545609]both companies started out with the similar budget? its about what you do with your investment that counts. Thus the difference between them two![/quotemsg]
The ignorance is astounding. Intel fears competition so much that they paid (bribed) all the major consumer PC manufacturers with BILLIONS of dollars in "rebates" so that they would not use AMD processors. It was so bad that AMD offered to give HP a million FREE processors but HP turned it down saying that they [HP] were too reliant on Intel's rebates and 'could not afford" to take all those FREE processors. This is only a mere fraction of all the completely illegal things Intel has done to maintain its monopoly in the X86 market. This trend of anti-competitiveness dates back to almost the very beginning, not long after the company was founded. Had they succeeded and killed AMD, Nvidia would've been next.

Anyone who considers themselves a technology enthusiast or a fan of innovation in general needs to watch this: https://www.youtube.com/watch?v=osSMJRyxG0k
It is guaranteed to make you reconsider ever buying an Intel CPU again.

 

[shknawe]

Intel is really getting desperate. Showing a chart with only three or four meaningless results, and hanging your hat on those, indicates to me Intel has been caught with their proverbial panties down. All AMD has to show is a price to performance chart and all the mumbo jumbo goes away.
I will give Intel the nod for top performance, but their cost is not worth the extra 10 percent performance gap they still enjoy over AMD.
Someone at corporate in Intel needs to wake up and become price competitive and they would have an avalanche of sales. If you had two different model cars that had same performance, and one was cheaper than the other which car do you think will sell the most units? Silly INTEL.

 

[Wisecracker]

♫ ♪ Intel Presents Its AMD EPYC Server Test Results ♪ ♫

If you believe this massive song and dance 'marketing wall of propaganda' from Chipzilla, I'm thinking you also believe that "Tax cuts pay for themselves" or

"The Moon landing was fake" or

"The check is in the mail"

:lol:

Otherwise, Anandtech did some benchies back in July ...
Sizing Up Servers: Intel's Skylake-SP Xeon versus AMD's EPYC 7000

It paints a different picture, and kinda set the table for an Intel 'bombardment' response of some sort. Can't say it was not unexpected.

In more breaking news, New York Yankee baseball fans think the Boston Red Sox suck!

 

[sinangular]

Look i've always had bad luck with intel. How you supposed to meet girls with only gay friends at 41 versus 19... #smokefit.us

 

[AeroWB]

Intel has been really busy finding benchmarks that show the weaknesses of AMD's EPYC product compared to their own.

Intel is just trying to hide the fact that with the launch of their new Xeon Scalable Processor there is (almost) no improvement in performance per dollar compared to their previous generation product. I really hope AMD will take a significant market share so Intel cannot afford repeat this again.

Also both AMD and Intel should make good performing 8-core parts for 2 socket or 16-core parts for 1 socket systems for a good price due to the core tax that Microsoft introduced with the Server 2016 pricing. Both AMD and Intel's new lineups only feature 2.1GHz base parts in 8-core with a reasonable price which is very disappointing.

 

[jimmysmitty]

[quotemsg=20423469,0,2331410]i only have 8c 190x thread ripper , but im not getting no high latencies from die2die and my chip has two and only 4 cores each, 2 dies, and to sepaerate CCXs for both dies, and max latecy i see is like 107ns but thats on like a very large memory caching. 4gb+ not acuall but i forgot the resuts, but intels "INTERNAL" (FAKED) Tests///theyve don this before? and i now it?[/quotemsg]

So you are saying that you get different latency results on a desktop based system than a server based system?

Seems plausible.

[quotemsg=20423476,0,2331410]any i bet 100M $$$ that intel isnt even running the Ram @ anything besides 2133Mhz, i bet my life on it[/quotemsg]

States in the slides that they used DDR4 2667. Reading does help a lot.

In fact, they also used 1 DIMM per channel to get the most memory bandwidth per system. Look at slide page 18 and 19.

[quotemsg=20424004,0,1425755]Intel cheated as usual. Intels properly compiler isn't something that most clients will use anyway & compared to open source ones (GGC, VLLM) which they will it gives some 20% performance boost. This actually EPYC right back into competition.
Demonstration:
https://3s81si1s5ygj3mzby34dq6qf-wpengine.netdna-ssl.com/wp-content/uploads/2017/11/cavium-vulcan-thunderx2-skylake-xeon-spec.jpg
All of this ain't all that important as ARM server's with NUMA support are coming to finally brick X86.
https://www.nextplatform.com/2017/11/27/cavium-truly-contender-one-two-arm-server-punch/amp/
& with a little help of the good old vet's this time around Intel won't have even a compiler advantage to drag about.
http://www.cavium.com/newsevents-Cray-Catapults-Arm-Based-Processors-Into-Supercomputing.html
Fun part that makes this even nostalgic for me is after long time Intel monopoly we are back where & how it all started.[/quotemsg]

You realize the majority of these benchmarks are third party programs and industry standard benchmarks meaning they probably didn't use Intels compiler.

[quotemsg=20425130,0,117113]
♫ ♪ Intel Presents Its AMD EPYC Server Test Results ♪ ♫

If you believe this massive song and dance 'marketing wall of propaganda' from Chipzilla, I'm thinking you also believe that "Tax cuts pay for themselves" or

"The Moon landing was fake" or

"The check is in the mail"

:lol:

Otherwise, Anandtech did some benchies back in July ...
Sizing Up Servers: Intel's Skylake-SP Xeon versus AMD's EPYC 7000

It paints a different picture, and kinda set the table for an Intel 'bombardment' response of some sort. Can't say it was not unexpected.

In more breaking news, New York Yankee baseball fans think the Boston Red Sox suck!


[/quotemsg]

Those are very different benchmarks. These are for Datacenter servers, not database servers. Very different roles and performance numbers.

[quotemsg=20425955,0,165447]Intel has been really busy finding benchmarks that show the weaknesses of AMD's EPYC product compared to their own.

Intel is just trying to hide the fact that with the launch of their new Xeon Scalable Processor there is (almost) no improvement in performance per dollar compared to their previous generation product. I really hope AMD will take a significant market share so Intel cannot afford repeat this again.

Also both AMD and Intel should make good performing 8-core parts for 2 socket or 16-core parts for 1 socket systems for a good price due to the core tax that Microsoft introduced with the Server 2016 pricing. Both AMD and Intel's new lineups only feature 2.1GHz base parts in 8-core with a reasonable price which is very disappointing.[/quotemsg]

You mean like any company does when comparing? Do you really think Intel would show their product failing vs AMD? DO you think AMD would? What was the last AMD marketing slide that showed off their CPU/GPU losing to the competition? None?

Companies market their products. Their goal is to sell that product.

Imagine going to buy a car and the car salesman tells you "The other car is better but you should buy ours. Look at how the other car performs better, gets better gas mileage and has better features. Don't you want to buy my car now?". That's not how marketing works.

 

[agello24]

intel is expensive for no reason at all. they just know idiots will buy their stuff for the name sake.