Sek and Jean Pierre, I agree that I polluted a little your threads.
Sorry for that . So here is a separate one.

For those who just read for the first time please check the threads:
Technology News and Building the 400 TFLOP Machine.
I understand that there was something already defined from the beginning at WCG, on ratios between Points, Credits and GFLOP, TFLOP etc. But this is not carved in stone. This is not the absolute truth. Things may change, may evolve like the computer industry has evolved over the last 20 years etc.
The ratio I mentioned of 1 TFLOP = 8 WCG points was rounded (8.1 in fact)and calculated on one day the 1st June. On the 2nd June yesterday it was 8.04.

Saying that we prefer to stay with 1TFLOP = 700'000 WCG points is an arbitrary choice as WCG Points could be anything. I do not discuss the 1 to 7 ratio BOINC to WCG. But then if we keep the same logic do not call those TFLOP, but WCGFLOP or anything else, and then I will have nothing to say. But I will also have no idea what my CPU really does and how he performs.

If we want members to rate their machine more or less to some real crunching performance of their CPU in terms of operations done, then we should have a ratio calculated back from real crunching data, and not from theoretical formulas. What I tried to do, (and I am far from being perfect) is to relate some real CPU benchmark, embedded in Boinc and the overall global result as measured by WCG.
Each CPU has different crunching performances and each project is also a different mix in type of operations FLOP (floating point) or IOP (integer) and IP (instructions).
If the aim is to have a common base of comparison then the global ratio is the best one because it is the average result over all projects and all type of machines.
By making the test on a specific machine like the 980X and if we suppose the projects crunch with half FLOP and half IOP then we get to a ratio of:
1 TFLOP = 9.3 WCG points.
This is not so far to the real WCG average of about 8. This also shows that the connection between the physical benchmark carried out and WCG global data there is consistency.
If we look at theoretical and commercially published maximums by Intel we are well under that max performance for sure but again it is inside the envelope of what can be expected, as projects are not optimized for one specific CPU but are written to run across a very large quantity of machines and operating systems.

As a conclusion everybody is free to do what he wants. As an example my solar systems emits HEGP (High Energy Graviton Particles) and not WCG points

But if we want to talk performance in terms of real CPU work the nearest way to approach the truth is : 1 TFLOP = 8 WCG points.
Period.