Intel Xeon 6766E/6780E Sierra Forest vs. Ampere Altra Performance & Power Efficiency

Written by Michael Larabel in Processors on 5 June 2024 at 11:22 AM EDT. Page 3 of 6. 10 Comments.
Speedb benchmark with settings of Test: Read While Writing. EPYC 9754 was the fastest.
Speedb benchmark with settings of Test: Read While Writing. EPYC 9754 was the fastest.
Speedb benchmark with settings of Test: Read While Writing. EPYC 9754 was the fastest.

The RocksDB-compatible Speedb key value store also exhibited similar performance and power efficiency to RocksDB between these server processors.

Timed Node.js Compilation benchmark with settings of Time To Compile. EPYC 9684X was the fastest.
Timed LLVM Compilation benchmark with settings of Build System: Ninja. EPYC 9684X was the fastest.

The code compilation workloads carry the caveat that there can be code path differences between AArch64 and x86_64 builds. But in any event the Ampere Altra M128-30 was much slower than the modern AMD EPYC and Intel Xeon processors when it comes to compiling some prominent open-source projects and using a job for each core.

Timed LLVM Compilation benchmark with settings of Build System: Ninja. EPYC 9684X was the fastest.

The Ampere Altra Max M138-30 was on average pulling 123 Watts while compiling the massive LLVM codebase and hitting 185 Watts, right above its TDP rating. The only CPU consuming less power while compiling LLVM was the AMD EPYC 8534PN "Siena" processor with a 117 Watt average and 147 Watt peak. And that AMD EPYC 8534PN 64-core processor compiled LLVM in 73% the time of the 128-core Ampere Altra Max processor. Intel Xeon 6700E processors do look very good for CI/CD purposes and other build farms if frequently compiling a lot of code either for massive open-source codebases or if you'll be building a lot of different codebases (or revisions of the same code) concurrently to exploit the 144 core or 288 core (dual socket or upcoming Xeon 6900E) potential.

OpenSSL benchmark with settings of Algorithm: SHA256. EPYC 9754 was the fastest.

When looking at the SHA256 performance with OpenSSL, the Ampere Altra Max M128-30 outperformed prior generation Intel Xeon CPUs with their lower core counts but fell well short of the Xeon 6766E/6780E in raw performance as well as the 96 to 128 core EPYC processors.

OpenSSL benchmark with settings of Algorithm: SHA256. EPYC 9754 was the fastest.
OpenSSL benchmark with settings of Algorithm: SHA256. EPYC 9754 was the fastest.

But for the SHA256 performance-per-Watt, the M128-30 did come out ahead of the AMD EPYC processors and was comparable to the Sierra Forest CPUs in power efficiency.

John The Ripper benchmark with settings of Test: Blowfish. EPYC 9754 was the fastest.
John The Ripper benchmark with settings of Test: Blowfish. EPYC 9754 was the fastest.
John The Ripper benchmark with settings of Test: Blowfish. EPYC 9754 was the fastest.

With John The Ripper's crypto software, the Blowfish performance saw the M128-30 performing similar to the Xeon Platinum 8592+ but well short of the new Sierra Forest and AMD EPYC 9004 processors. On a performance-per-Watt basis the M128-30 came out ahead of the EPYC 9654/9684X but behind the Intel Sierra Forest and AMD Bergamo/Siena processors.

John The Ripper benchmark with settings of Test: bcrypt. EPYC 9754 was the fastest.
John The Ripper benchmark with settings of Test: bcrypt. EPYC 9754 was the fastest.

It was a similar story as well with the bcrypt benchmark.


Related Articles