Hands On & Initial Benchmarks With An Ampere eMAG 32-Core ARM Server
I also ran some benchmarks against another ARM server I recently tested (remotely) of a Cavium ThunderX1 with 96 ARMv8 cores... Each ARM server was running its stock software stack.
Understandably with it being 32-cores versus 96 cores, in the highly threaded tests like code compilation, the ThunderX did come out ahead. Interestingly though the 32-core eMAG server wasn't really that far behind.
The multi-threaded performance was also very close in the C-Ray multi-threaded ray-tracer. While the Cavium ThunderX1 was tested in a 96 core configuration, its cores can only clock up to 2.5GHz compared to the 3.3GHz turbo with Ampere.
For single-threaded workloads, the Ampere eMAG is easily much faster than the Cavium server benchmarked.
Overall, it's interesting to see how well the 32-core Ampere eMAG server was competing with a 96-core ThunderX1 server. (Unfortunately I have no ThunderX2 hardware or access at this time.)