3 ms·
Does anyone have some more real-world benchmarks of the Ampere part? The article implies that it performs similarly to the high-end Intel and AMD parts, but th
by PragmaticPulp 5y ago
Does anyone have some more real-world benchmarks of the Ampere part?
The article implies that it performs similarly to the high-end Intel and AMD parts, but the linked benchmarks are mostly synthetic tests like Stress-NG. There are a few Blender benchmarks ( https://www.phoronix.com/scan.php?page=article&item=ampere-altramax-benchmarks&num=4 https://www.phoronix.com/scan.php?page=article&item=ampere-a... ) but the rest of the tests aren’t really relevant to real workloads.
- kvathupo 5y agoMany benchmarks fail to capture that if one were programming for speed, intel CPUs seem to be the default option by virtue of the intel compiler, unless you use a HDL [1]. It's quite interesting how hardware manufacturers have been able to capture market share due to their software offerings, see NVIDIA with CUDA for almost any deep learning library. [1] - https://www.agner.org/optimize/blog/read.php?i=49 https://www.agner.org/optimize/blog/read.php?i=49
- ksec 5y agoYou could test it on AWS Graviton 2, the benchmarks will be similar given they are both based on N1. The only differences are cache, PCI-E and Memory.
- TheCondor 5y agoFWIW, dollar for dollar, Graviton2 is impressively quick.