Open Hou5e opened 1 year ago
We can only estimate FLOPS but this could be added. The machine itself does not have anyway of telling you how many floating-point instructions a particular process has executed. Also, GPU FLOPS are calculated differently from CPU FLOPS. It's not an simple task. I'd welcome a PR.
We can only estimate FLOPS but this could be added. The machine itself does not have anyway of telling you how many floating-point instructions a particular process has executed. Also, GPU FLOPS are calculated differently from CPU FLOPS. It's not an simple task. I'd welcome a PR.
Understood but even an estimate of the FLOPS either on the GPU or CPU would be useful and appreciated,
Add reporting FLOPS (both native and x86 equivalent) for more accurate supercomputing reporting, and for benchmarking purposes. (This enhancement request was from someone on Discord)
FLOPS are only estimated here: https://stats.foldingathome.org/os and described here: https://foldingathome.org/support/faq/flops
Obviously, if it's going to require more FLOPS to compute FLOPS than the WU uses, then it's probably not worth it. But, there should be a way to periodically take a sample of FLOPS during computing each WU, to get a more accurate estimate than what is currently shown on the website.
It might be interesting to see a utilization comparison for making coding improvements for: Average FLOPS the hardware was capable of, versus average FLOPS actually used for the FAH WU.
Many websites use FLOPS (not FAH points) for a more apples-to-apples comparison for computing performance: https://en.wikipedia.org/wiki/Computer_performance_by_orders_of_magnitude https://en.wikipedia.org/wiki/List_of_volunteer_computing_projects (Sort the performance column on the right)