Discussion [WCCF] Intel Claims Xeon Cascade Lake-AP 56 Core CPU Up To 84% Faster Than AMD’s 64 Core EPYC Rome 7742 in Real-World HPC Benchmarks

[Hitman928]

Intel Claims Xeon Cascade Lake-AP 56 Core CPU Up To 84% Faster Than AMD's 64 Core EPYC Rome 7742 in Real-World HPC Benchmarks [Updated]

Intel has posted new benchmarks claiming that their Xeon class Cascade Lake-AP CPUs run much faster than AMD's 2nd Generation EPYC Rome CPUs.

wccftech.com

This comes on the heels of another article discussing how consumer work loads don't scale well with cores. Looks like intel is hitting on all fronts with their pr.

 

[gdansk]

Intel compiler, 400W, and a processor so obscure it doesn't have an MSRP. Great sell, Intel.

 

[amrnuke]

Of course, as expected, they used an old version of GROMACS to artificially make their chip look better.

 

[jpiniero]

It's largely moot since you can't actually buy Cascade Lake-AP, if you wanted to. The power draw is too much anyway.

 

[Markfw]

Oh, and these are Intel benchmarks. You think I believe that ? Not until some other sites do some benchmarks.

Intel PR is in high gear, and the idiots in upper management in IT will probably believe it.

 

[NostaSeronx]

Intel #1, AMD #nobody.


Who would have thought that AMD would get beaten by Intel. Like oh my gosh~ I still remember my Chedder hill rekking these Krypton-8 bois.

 

[TheGiant]

well Intel is promoting what they can
looks definitely as the least effective machine for cinepeenbenching but HPC is all about mem bandwitch
12 Mem channels, avx512, looks entirely possible
about that TDP- today no value from either manufacturer

 

[tamz_msc]

Serve The Home has an article up detailing how the GROMACS benchmark was done in a misleading way.

Intel Performance Strategy Team Publishing Intentionally Misleading Benchmarks

We checked the Intel performance strategy team's latest benchmarks and found that they used an intentionally misleading test to back their claims

www.servethehome.com

Besides they are using an old version of the Linux kernel.

 

[ClockHound]

Serve The Home has an article up detailing how the GROMACS benchmark was done in a misleading way.

Intel Performance Strategy Team Publishing Intentionally Misleading Benchmarks

We checked the Intel performance strategy team's latest benchmarks and found that they used an intentionally misleading test to back their claims

www.servethehome.com

Besides they are using an old version of the Linux kernel.

...for Unmitigated Benchmark Success. Cheating Inside.

 

[moinmoin]

Intel's definition of "real world" continues to disappoint.

 

[Markfw]

What I want to know, is how do these people think the rest of the world is stupid enough to not find their deception ? Again, this reeks of desperation.

Nothing to see here, move along....

 

[TheGiant]

Serve The Home has an article up detailing how the GROMACS benchmark was done in a misleading way.

Intel Performance Strategy Team Publishing Intentionally Misleading Benchmarks

We checked the Intel performance strategy team's latest benchmarks and found that they used an intentionally misleading test to back their claims

www.servethehome.com

Besides they are using an old version of the Linux kernel.

interesting, so either they lie or are incompetent
both can happen in such a company of this size

 

[JasonLD]

What I want to know, is how do these people think the rest of the world is stupid enough to not find their deception ? Again, this reeks of desperation.

Nothing to see here, move along....

AMD EPYC Rome NAMD and the Intel Xeon Response at Computex 2019

We discuss the lively exchange between AMD EPYC Rome 64 core benchmarks and the Intel Xeon response at Computex 2019 and help you decipher what it means

www.servethehome.com

Well, AMD does the same thing. We all know these companies will do anything in their benchmarks to make their product look good against their competition.

 

[Markfw]

AMD EPYC Rome NAMD and the Intel Xeon Response at Computex 2019

We discuss the lively exchange between AMD EPYC Rome 64 core benchmarks and the Intel Xeon response at Computex 2019 and help you decipher what it means

www.servethehome.com

Well, AMD does the same thing. We all know these companies will do anything in their benchmarks to make their product look good against their competition.

OK, but with Rome we have seen independent benchmarks, and they don;t show this.

 

[JasonLD]

OK, but with Rome we have seen independent benchmarks, and they don;t show this.

Well, I also want to see independent benchmarks of Xeon Platinum 9200 series, but will they be able to find any lol. I don't think independent reviewers will be able to get 56-core parts until Cooper Lake.

 

[Markfw]

Well, I also want to see independent benchmarks of Xeon Platinum 9200 series, but will they be able to find any lol. I don't think independent reviewers will be able to get 56-core parts until Cooper Lake.

So Intel is publishing benchmarks on a CPU nobody can get their hands on, but Rome is widely available. So again, hos dishonest is that ?

 

[extide]

So Intel is publishing benchmarks on a CPU nobody can get their hands on, but Rome is widely available. So again, hos dishonest is that ?

You know that if Intel really had a part that could easily win benchmarks and stuff they would gladly hand out samples to press. Anandtech, Servethehome, Phoronix and the like at the least...

 

[Nothingness]

What I want to know, is how do these people think the rest of the world is stupid enough to not find their deception ? Again, this reeks of desperation.

They did similar things 10 years ago when they released the first Atom chips, making biased comparisons against ARM chips. Many people accepted that as the truth even educated ones.

Nothing new here alas. Marketing is made by liars, no matter the company.

 

[JasonLD]

So Intel is publishing benchmarks on a CPU nobody can get their hands on, but Rome is widely available. So again, hos dishonest is that ?

What does that have to do with Intel and AMD fabricating benchmarks on their favor on their own PR pieces. I was pointing out that both companies are doing it, regardless of what independent reviews are showing otherwise.

 

[Kocicak]

One really interesting thing is the number of workloads of each benchmark. Why does the number vary from benchmark to benchmark? What was the key to determine how many workloads to run?

BTW what is "manufactoring"? Does Intel have a typo in the headline of the second slide???

 

[TheELF]

Of course, as expected, they used an old version of GROMACS to artificially make their chip look better.

GROMACS is only 20% faster...it's HPL ,LAMMPS and FSI which are 58-84% faster that are pushing the geomean up so high.

One really interesting thing is the number of workloads of each benchmark. Why does the number vary from benchmark to benchmark? What was the key to determine how many workloads to run?

To fill up all of the cores,if it's a two chip setup you will have to run at least one per CPU if the workload can't use even that many cores you will have to run multiple instances even on each CPU, and then you would also want enough to to make the CPU run long enough to get into sustained boost clocks,

 

[Topweasel]

AMD EPYC Rome NAMD and the Intel Xeon Response at Computex 2019

We discuss the lively exchange between AMD EPYC Rome 64 core benchmarks and the Intel Xeon response at Computex 2019 and help you decipher what it means

www.servethehome.com

Well, AMD does the same thing. We all know these companies will do anything in their benchmarks to make their product look good against their competition.

Did you actually read that? They used a highly optimized for Intel benchmark with heavy use of the AVX-512. They didn't quite run it as optimized as Intel ended up running it (maybe better ICC). So to some degree there is a misrepresentation of the number but this isn't anywhere near comparable to using an Intel benchmark that runs like pooh on any non-intel CPU and advertising as a real world example and therefore tons better.

 

[Kocicak]

To fill up all of the cores,if it's a two chip setup you will have to run at least one per CPU if the workload can't use even that many cores you will have to run multiple instances even on each CPU, and then you would also want enough to to make the CPU run long enough to get into sustained boost clocks,

When running two different systems with different CPUs with different core count, one could guess that the number of workloads determined this way would be different for both systems. Which of these two numbers did they choose?

 

[GaiaHunter]

What I want to know, is how do these people think the rest of the world is stupid enough to not find their deception ? Again, this reeks of desperation.

Nothing to see here, move along....

I guess it is about creating an headline someone will see on a news feed or google search and won't read anything else.

That was how I first came into contact with this story.

 

[uzzi38]

Well, I also want to see independent benchmarks of Xeon Platinum 9200 series, but will they be able to find any lol. I don't think independent reviewers will be able to get 56-core parts until Cooper Lake.

Afaik Ian might be the only one that currently has access to one. I don't think STH has gotten one anyway.