I benchmarked one of my rigs running W7 64 bit with a Core i7 980X running stock speed 3.3 Ghz with hyperthreading.

I tested with the Dhrystone and Whetstone Benchmarks adapted to this CPU. I measure:

Dhrystone 147.32 GIPS
Whetstone 106.7 GFLOPS

If I check with the values of the CPU benchmark in the Boinc application I get:

36.2 GFLOPS and 114.9 GIPS which is very different.

If we look at the definition we use, 700'000 WCG points are equal to 1 TFLOPS or 1000 GFLOPS. In short when my solar system generates 900'000 points in a day it is equivalent to 1.285 TFLOPS.

We have two possibilities to check this value:

a) If I consider the 980X benchmark with 106.7 GFLOPS per CPU, and I multiply by 18 x 980X and 2 x 950 (OC'ed) which crunch fully at the moment. Many 980X run at 4.0 or 4.2 GHz so they would have even a higher score. But for simplicity sake, suppose we have 20 units at 106.7 GFLOPS then this aggregates to a total of 2.134 TFLOPS.

b) If I consider the Boinc benchmark with 36.2 GFLOPS then with 20 machines I am at 0.724 TFLOPS. This result is coherent with the Boinc approach that considers the FLOPS rate as half what WCG considers. But it is a logical result because the Boinc credits are allocated on the basis of the benchmark that the Boinc application does on each machine at start.


We have then three values: 1.285, 2.134, and 0.724 TFLOPS

Were lies the truth?

If we average these three values we get 1.381 TFLOPS
And if we look at the WCG rate that is 1.285 TFLOPS we see that this value is not far from the average.

Coclusion let's keep the way WCG estimates TFLOPS.