Flops to points

[WMCheerman]

I was wondering if there is a any correlation between FLOPS and the points that Folding@home give out? If not is there any way to calculate my accurate FLOPS output here at Folding@home?

[foldy]

A gtx 980 ti is specified around 6 TFLOPs and gets 500,000 PPD. So 6 TFLOP per second and 500K points per day.

Some math: 6 TFLOPs * 60s * 60m *24h = 518,400 TFLOP per day

So looks like PPD is similar to TFLOP per day (on current nvidia GPUs).

The other way round: If you get e.g. 100,000 PPD then your TFLOPs is
TFLOPs = 100,000 PPD / 24h / 60m / 60s = 1.16 TFLOPs in folding@home

[bruce]

I was wondering if there is a any correlation between FLOPS and the points that Folding@home give out? If not is there any way to calculate my accurate FLOPS output here at Folding@home?

Part of the reason why they won't correlate is the time-dependent bonus. If 10 machines each process a WU and another machine with 10x the FLOPs processes 10 WUs from the same project, they'll take about the same length of time and will represent the same FLOPs.

The Points will be radically different.

If people kept track of baseline points rather than [baseline+bonus], one might be able to find a correlation, but nobody I know pays any attention to isolating the baseline points.

[7im]

The FLOPs are directly measured (sampled) from the fahcore performance (in the returned work units). In addition to the bonus points not correlating to FLOPs, the size and complexity of work units also skews direct (apples to apples) comparisons for the many various hardware types. The FLOPs FAQ explains other variations as well.

See also: https://folding.stanford.edu/home/faq/faq-flops/