How to accurately measure the performance of the Bitcoin network in FLOPs?

Question

Mathematics

Hello!
The question arose - how to estimate the power of the Bitcoin network in FLOPS with the least errors? The problem is that FLOPS are operations on floating point numbers, and sha2 counts are operations on integers. Therefore, it is incorrect to compare these indicators.
There is a formula by which bitcoinwatch.com calculates the power of the Bitcoin network:
1 INTOP = 2 FLOP
1 hash = 6.35K INTOP
1 hash = 12.7K FLOP
At the moment, according to bitcoinwatch, the power of the network is 271 PFLOPS, which exceeds the figure of the most powerful supercomputer Titan 15 times. The figure is beautiful, but it turns out a little from the ceiling.
Here is a table for different video cards
ATI HD 7970 3.8 TFLOPS, 650 Mhash/s, 1 hash = 5.84k FLOP
ATI HD 5870 2.7 TFLOPS, 400 Mhash/s, 1 hash = 6.75k FLOP
nVidia GTX 580 1.5 TFLOPS, 140 HASH/s, 1 hash = 11.29k
FLOP , but since most of them are mining on ATI video cards, these values ​​are not correct.
There is also a discussion on the wiki en.wikipedia.org/wiki/Talk%3AFLOPS#Distributed_com... but there is no answer there either.
It is clear that the comparison is not entirely correct, but I want to understand how to evaluate the computing power of the network with the least distortion.

Answer 1

I think we need to compile a table on TFLOPS / Ghash indicators for different video cards, evaluate the limit values ​​\u200b\u200bfrom it (minimum and maximum) and some average - from assumptions like "X% of miners use video cards A, Y% - video cards B".
Maybe the TFLOPS / Ghash scatter plot will tell you some features.
And the estimate 1 INTOP = 2 FLOP is most likely taken by the CPU.