Benchmarking Amazon's ARM Graviton CPU With EC2's A1 Instances

Written by Michael Larabel in Computers on 26 November 2018. Page 3 of 5. 33 Comments

With the threaded Zstd compression test, it was another case of the ARMv8 A1 instances pretty much needing twice the vCPU count to compete with the Xeon/EPYC instances.

With the Perl benchmarks, which is obviously single-threaded, took twice as much time to run as the new AMD EPYC M5A instances.

The build performance for the Go programming language didn't come out any better for Graviton.

Golang's JSON parsing and GC tests were the first cases where the A1 instances weren't significantly worse off than AMD/Intel, but still was a very measurable difference in performance.

The only benchmark carried out where Graviton managed to do better than AMD EPYC instances was the Rustlang Prime benchmark, which is more of an outlier we've seen on AMD hardware with its performance coming in abnormally short of expectations on AMD Zen.

The EPYC loss with the prime benchmark fortunately isn't widespread to Rustlang as with the language's Mandelbrot test, EPYC is back to the game with Xeon in running ahead of the Amazon ARM processor.

Related Articles