Benchmarking Amazon's Graviton2 Performance With 64 Neoverse N1 Cores Against Intel Xeon, AMD EPYC

John The Ripper performance put the Graviton2 with its sixty-four N1 cores comparable to an EPYC 7502P with 32 cores / 64 threads. The Blowfish workload performance was also similar to the Xeon Cascade Lake Refresh 6258R with 28 cores / 56 threads and higher clock frequencies.

For AV1 video decoding, the Graviton2 CPU performance was better than Xeon Scalable but behind most of the EPYC Rome processors tested.

Very surprising was the Graviton2 picking up a win when it came to dav1d's performance for 10-bit content. Graviton2 was much faster than the x86_64 CPUs, possible due to some optimization difference between x86_64 and AArch64 for this open-source decoder.

The Graviton2 performance for building out the Linux kernel was comparable only to the EPYC 7262.

But for the more demanding timed LLVM compilation, the Graviton2 build speed was comparable to the EPYC 7F52 and Xeon Platinum 8280.