4 ms·
I’d love to see benchmarks of LLM inference, and an apples to apples comparison between Genoa and Saphire Rapids for that use case.
by densh 3y ago
I’d love to see benchmarks of LLM inference, and an apples to apples comparison between Genoa and Saphire Rapids for that use case.
- rbanffy 3y agoI volunteer to write about my experience with it as a daily driver ;-) But then they'd need to make a desktop with it. Would be ~90% heatsink in volume.
- formerly_proven 3y agoAre there any common benchmarks for this, preferably pre-packaged as a container? Edit: Point being that if you can point someone towards a container or simple to build benchmark like that it becomes infinitely more likely that people will run it on some hardware compared to "Yeah I guess if someone could develop a benchmark and then also happen to have access to a diverse set of hardware to run it on". The easier it is to run a benchmark the more results are going to be floating around (regardless of if it's a good benchmark, Cinebench and Crystaldiskmark would be prime examples of this).
- rbanffy 3y agoPhoronix did some, but not really anything centered on that - may be possible to infer from their data though. https://www.phoronix.com/review/epyc-9684x-3d-vcache https://www.phoronix.com/review/epyc-9684x-3d-vcache