Dissecting Intel's EPYC Benchmarks: Performance Through the Lens of Competitive Analysis
by Johan De Gelas & Ian Cutress on November 28, 2017 9:00 AM EST- Posted in
- CPUs
- AMD
- Intel
- Xeon
- Skylake-SP
- Xeon Platinum
- EPYC
- EPYC 7601
HPC Benchmarks
Discussing HPC benchmarks feels always like opening a can of worms to me. Each benchmark requires a thorough understanding of the software and performance can be tuned massively by using the right compiler settings. And to make matters worse: in many cases, these workloads can be run much faster on a GPU or MIC, making CPU benchmarking in some situations irrelevant.
NAMD (NAnoscale Molecular Dynamics) is a molecular dynamics application designed for high-performance simulation of large biomolecular systems. It is rather memory bandwidth limited, as even with the advantage of an AVX-512 binary, the Xeon 8160 does not defeat the AVX2-equipped AMD EPYC 7601.
LAMMPS is classical molecular dynamics code, and an acronym for Large-scale Atomic/Molecular Massively Parallel Simulator. GROMACS (for GROningen MAchine for Chemical Simulations) primarily does simulations for biochemical molecules (bonded interactions). Intel compiled the AMD version with the Intel compiler and AVX2. The Intel machines were running AVX-512 binaries.
For these three tests, the CPU benchmarks results do not really matter. NAMD runs about 8 times faster on an NVIDIA P100. LAMMPS and GROMACS run about 3 times faster on a GPU, and also scale out with multiple GPUs.
Monte Carlo is a numerical method that uses statistical sampling techniques to approximate solutions to quantitative problems. In finance, Monte Carlo algorithms are used to evaluate complex instruments, portfolios, and investments. This is a compute bound, double precision workload that does not run faster on a GPU than on Intel's AVX-512 capable Xeons. In fact, as far as we know the best dual socket Xeons are quite a bit faster than the P100 based Tesla. Some of these tests are also FP latency sensitive.
Black-Scholes is another popular mathematical model used in finance. As this benchmark is also double precision, the dual socket Xeons should be quite competitive compared to GPUs.
So only the Monte Carlo and Black Scholes are really relevant, showing that AVX-512 binaries give the Intel Xeons the edge in a limited number of HPC applications. In most HPC cases, it is probably better to buy a much more affordable CPU and to add a GPU or even a MIC.
The Caveats
Intel drops three big caveats when reporting these numbers, as shown in the bullet points at the bottom of the slide.
Firstly is that these are single node measurements: One 32-core EPYC vs 20/24-core Intel processors. Both of these CPUs, the Gold 6148 and the Platinum 8160, are in the ball-park pricing of the EPYC. This is different to the 8160/8180 numbers that Intel has provided throughout the rest of the benchmarking numbers.
The second is the compiler situation: in each benchmark, Intel used the Intel compiler for Intel CPUs, but compiled the AMD code on GCC, LLVM and the Intel compiler, choosing the best result. Because Intel is going for peak hardware performance, there is no obvious need for Intel to ensure compiler parity here. Compiler choice, as always, can have a substantial effect on a real-world HPC can of worms.
The third caveat is that Intel even admits that in some of these tests, they have different products oriented to these workloads because they offer faster memory. But as we point out on most tests, GPUs also work well here.
Facebook
Twitter
RSS
105 Comments
View All Comments
piesquared - Tuesday, November 28, 2017 - link
And the hilarity continues. So AMD posts in house benchmarks and the crowd goes: Derp, these are AMD supplied benchmarks, best wait for third party benchmarks.Intel posts in house benchmarks and the crowd goes: Wow awesome dude, that's the shitsors! Who needs third party benchmarks, AMD should post more in house benchmarks. derp derp
tamalero - Tuesday, November 28, 2017 - link
Guerrilla marketing at its finest? The hilarity is that when Intel was dominating.. they never mentioned intel nor they needed.Now that AMD has a compelling product. They suddenly started doing "comparisons" left and right and claiming how bad "glue" is in AMD cpus (while ignoring the drama bout using cheap TIM instead of solder)
bmf614 - Tuesday, November 28, 2017 - link
Epyc really hasnt even launched yet. Try buying a Dell or HP with Epyc. Nope.supdawgwtfd - Wednesday, November 29, 2017 - link
It's launched. Demand has outstripped supply. They are now starting to get on top of it.Maybe stop being an Intel biased dickhead and go look at what is actually happening?
Topweasel - Wednesday, November 29, 2017 - link
Yeah, I don't get it. I mean even Ryzen mobile launched well before we saw it. Eypc announcement early was important to build up demand with OEM's. Something that wasn't as important with a consumer product that needed announcement with availability. EPYC's announcement wasn't for the end purchaser. Both these need long testing periods and seed supply. Epyc then has ODM builds for cloud services that they have supply. Ryzen mobile launched when OEM's had products to ship. EPYC launched when they products to ship to manufacturers. When those Manufacturers offered EPYC depends completely on their development cycle.Johan Steyn - Monday, December 18, 2017 - link
Haha so trueNinhalem - Tuesday, November 28, 2017 - link
Can we get ANSYS Structural or Comsol benchmarks for the HPC sections? Building machines using Xeons for these applications is beyond expensive for engineering design on fixed price contracts.anactoraaron - Tuesday, November 28, 2017 - link
No, because AT didn't test anything here. They are just 'publishing' Intel's benchmarks and calling it an 'analysis'.Doesn't this qualify for the #ad in the title?
Ryan Smith - Wednesday, November 29, 2017 - link
To throw some context in here, the purpose of this article isn't to publish Intel's benchmarks. Rather, it's commentary on what has been a very unusual situation.Up until now, neither AMD nor Intel have engaged in any serious Skylake Xeon vs. Zen EPYC technical marketing.
"AMD's technical marketing of the new CPU has been surprisingly absent, as the company not published any real server benchmarks. The only benchmarks published were SPEC CPU and Stream, with AMD preferring for its partners and third parties to promote performance"
This despite the fact that AMD and Intel's server products haven't been competitive like this in nearly a decade. Normally you'd expect there to be case studies flying out left and right, which has not been the case. And it's especially surprising since, as the underdog, AMD needs to claw back lost ground.
Consequently, Intel's own efforts are, to date, the first efforts by a server vendor to do a comprehensive set of benchmarks over a range of use cases. And let's be clear here: this is Intel doing this for Intel's own benefit. Which is why we've already previously reviewed the two CPUs, as have other 3rd party groups.
Still, I think it's very interesting to look at what Intel has chosen to represent, and what their numbers show. Intel has more resources than pretty much everyone else when it comes to competitive analysis, after all. So their choices and where they show themselves falling behind AMD says a lot about the current situation.
(And no, this doesn't quality for #ad as Intel hasn't paid us. That's not how this works; that's not how any of this works)
deltaFx2 - Wednesday, November 29, 2017 - link
@Ryan Smith: "Up until now, neither AMD nor Intel have engaged in any serious Skylake Xeon vs. Zen EPYC technical marketing." I think that's largely because the market is different from a decade ago. Hyperscalers do their own testing and aren't swayed by Intel's or AMD's whitepapers. They do their own thing. There are still many companies that buy and maintain their own servers, but my understanding is that this market is shrinking or at least not growing. Cloud is where the money is, and they know what they want. I don't think AMD is trying to go after enterprise this time around (I'm sure they'll take their business but the main target seems to be hyperscalers. The CCX, MCM, large memory footprint etc all point to them saying we'll target scale-out as opposed to scale-up. AMD does quite well in scale-out, while taking a hit in scale-up.).Also, AMD might still be in the process of doing minor firmware tweaks as evidenced by tier-1 server availability (HP/Dell) coming online only end of Q4.