Is the "Current GigaFLOPS" stat in the Server Status page accurate?

I mean, it did go up A LOT during the last few weeks, so it does seem to correlate tightly with how much is being proceeded, but is this number comparable to the number given by Folding@Home regarding their total computing power? I mean 1 GFLOP reported here = 1 GFLOP reported by Folding@Home.

---

I'm not sure how up to the minute it is, and how it's flops compare to the FaH flops, but I have been keeping an eye on it and it's been consistently going up for about 2 weeks now.

FaH is still over an order of magnitude more powerful than Rosetta it seems though. The crunching capability over there is mind-boggling.

Is the "Current GigaFLOPS" stat in the Server Status page accurate?

No.
Such values are usually derived from the RAC/Credit of the attached hosts, or the Tasks themselves. And due to the broken nature of the BOINC Credit mechanism (just look at the huge variability of Credit granted here for different Tasks and different Applications) any such claims of compute performance need to be taken with an ocean of salt. And then you have other BOINC projects that grant Credit in their own way, making their claims of compute performance to be complete and utter fantasy with no basis in reality.
The variability of Rosetta Credit makes me suspect they are using the BOINC Credit New mechanism, which tends to underpay significantly, and as Credit is used to determine FLOPs...

I mean 1 GFLOP reported here = 1 GFLOP reported by Folding@Home.

I've no idea how Folding@home come up with their FLOPs numbers- if they use FLOP counting then they should be pretty close to the true value. If they use some other mechanism then all bets are off.
But the FLOPs between the projects are nowhere near comparable due to the different methods of coming up with the value; you can't even compare FLOPs between BOINC projects, let alone projects outside of the BOINC framework.

---

Grant
Darwin NT

Is the "Current GigaFLOPS" stat in the Server Status page accurate?

No.
Such values are usually derived from the RAC/Credit of the attached hosts, or the Tasks themselves. And due to the broken nature of the BOINC Credit mechanism (just look at the huge variability of Credit granted here for different Tasks and different Applications) any such claims of compute performance need to be taken with an ocean of salt. And then you have other BOINC projects that grant Credit in their own way, making their claims of compute performance to be complete and utter fantasy with no basis in reality.
The variability of Rosetta Credit makes me suspect they are using the BOINC Credit New mechanism, which tends to underpay significantly, and as Credit is used to determine FLOPs...

I mean 1 GFLOP reported here = 1 GFLOP reported by Folding@Home.

I've no idea how Folding@home come up with their FLOPs numbers- if they use FLOP counting then they should be pretty close to the true value. If they use some other mechanism then all bets are off.
But the FLOPs between the projects are nowhere near comparable due to the different methods of coming up with the value; you can't even compare FLOPs between BOINC projects, let alone projects outside of the BOINC framework.

I kinda thought this was the case. I just want to know how "far behind" R@H is compared to Folding compute-power wise. I know the difference if probably huge, considering they can use GPUs. But it'd be good to know exactly how much and keep track of it. It'd also be great to see how R@H compares to a TOP500 Supercomputer and stuff like that.

---

R@h credits are based on the reported average credit claims of other hosts for each specific batch of work. This was the R@h solution to the credit spoofing that used to be such an issue. Credit claims are based on the BOINC CPU benchmarks for the machine. With this averaging of reported and completed work, anyone spoofing their benchmarks etc. would only inflate the credit granted for tasks reported AFTER theirs. Thus, the incentive to spoof R@h is not there.

---

Rosetta Moderator: Mod.Sense