Intel Bites Back At AMD For Misleading Rome EPYC Zen 2 Benchmarks Vs Xeon 8280

AMD made some waves at Computex where it delivered its first keynote for the event. To kick things off, the company talked about its second-generation EPYC servers processors, codenamed Rome, and promptly demonstrated a pair of 64-core EPYC chips thrashing two 28-core Intel Xeon Platinum 8280 (Cascade Lake) CPUs in a benchmark that showed AMD's hardware performing 2X better. That demonstration did not sit well with Intel, which has fired back with a more thorough comparison.

How things actually shake out between AMD and Intel, both on consumer desktops and in the server space, will be made more clear once reviewers have had a chance to properly test and benchmark AMD's new stuff. Make no mistake, though, both companies see the importance of holding the performance crown. The server industry in particular is huge—Intel pulls around half of its revenue from datacenter sales.

Therein lies the importance of a public demonstration, and why Intel cares about the benchmark and settings AMD used to pit its EPYC hardware against a Xeon chip. Recent data shows AMD chipping away at Intel's server CPU market share. AMD is still in a distance second, but the gap is narrowing. How things play out with AMD's second-generation EPYC processors will have an affect on market share, one way or the other.

As for the benchmarks, AMD pitted a 2P (dual-socket) EPYC system against a 2P Intel Xeon Platinum 8280 system, the former with 128 total cores and 256 threads, and the latter with 56 total cores and 112 threads. Obviously AMD has a big advantage here in both cores and threads, and the benchmark showed the EPYC system as being more than twice as fast (9.68ns versus 19.60ns per day), as shown above.

While not a fair fight in terms of cores and threads, that is also the point—AMD has the hardware available to muscle through workloads even faster than Intel's Xeon chips, or so it seems.

Not all cores are created equal, and more cores doesn't always equate to better overall performance. With an ideal mix of CPU cores, I/O and memory channels, Intel Xeon processors deliver unmatched workload performance. pic.twitter.com/7LFG9gEgg9

— Intel Business @ #Computex 2019 (@IntelBusiness) May 27, 2019

Intel, however, has been been posting all over social media that there is more to performance than just core counts, that the underlying architecture and optimizations matter too. According to Intel, AMD's demonstration left out NAMD optimizations on its hardware, and therefore unfairly tipped the scales in AMD's favor.

Click to Enlarge (Source: Intel)

What happens when NAMD optimizations are applied? Intel is glad you asked. Using NAMD optimizations, Intel's own testing shows its Xeon Platinum 8280 setup scoring 30 percent higher, at 12.65ns. AMD still beats Intel in this scenario, but the difference is far less dramatic, especially when factoring in the core and thread disadvantage that Intel is working with in this comparison.

Intel didn't stop there, however. It also showed internal performance data for a for its Xeon Platinum 9000 series (Cascade Lake-AP), which come with up to 56 cores and 112 threads, along with a dozen memory channels in a single package. In this scenario, Intel takes the lead over EPYC—two of the Xeon parts (112 total cores and 224 threads) outpaces AMD's setup by around 23 percent.

Click to Enlarge (Source: Intel)

Intel provided details for its test conditions, which you can see above. One thing to note is that the Linux setup is updated to deal with the latest security vulnerabilities, so on the surface, it appears to be a fair representation.

As for AMD's test conditions, it provided a statement to TomsHardware that lays things out.

"Based on AMD internal testing of the NAMD Apo1 v2.12 benchmark. AMD tests conducted on AMD reference platform configured with 2 x preproduction 2nd Generation EPYC 7nm (“Rome”) 64 core SoC, 16 x 32GB DDR4 2933MHz DIMMs, and Ubuntu 19.04, 5.0 kernel and using the AOCC 2.0 beta compiler with OpenMPI 4.0, FFTW 3.3.8 and Charms 6.7.1, achieved an average of 19.60 ns/day; versus a Dell 740 server configured with 2 x Intel Xeon Platinum 8280 28 core CPUs, 12 x 32GB DDR4 2933MHz DIMMs and Ubuntu 19.04 , kernel 5.0 using the ICC 19.0.3 compiler with FFTW 3.3.8 and Charms 6.7.1, achieved an average of 9.68 ns/day. Performance may vary with production silicon," AMD said.

From the information that Intel countered with, it certainly appears as though AMD left out some important optimizations, and by extension was able to show its EPYC hardware in a more favorable light as compared to Intel's Xeon hardware.

We are a long ways from having a full picture, though. It's always important to take vendor-provided performance data with a grain of salt. Reviews by outside parties will provide a complete evaluation of the situation. As it relates to that, outside reviews will also use multiple benchmarks, rather than focus on just a single test. Then there are things like power consumption, heat output, pricing, and support to consider. They are all factors.

That is our way of saying to hang tight—we'll know more about how these two platforms compare with one another in the near future.