But why on earth would they need double precision (i.e. floating point hardware) to do integer maths?
The video cards use FP for 3D rendering. OpenGL is full of matrix transforms and transcendental functions. While the results are integer very little else is.
I had the same feeling. But since I don't know PrimeGrid enough to reason about it I just stuck to their Web page. Certainly Eratosthenes' sieve does not need floating point math. But perhaps the use the Riemann's conjecture.
Tullio
I had the same feeling. But since I don't know PrimeGrid enough to reason about it I just stuck to their Web page. Certainly Eratosthenes' sieve does not need floating point math. But perhaps the use the Riemann's conjecture.
Tullio
Crunched the Proth Prime Search (Sieve)CUDA WU's on 2 Fermi-cards but cannot remember having false results.
Thank you for letting us know that you have seen these problems.
Well, not exactly. Our tests didn't exhibit these accuracy issues and they do occur only on a certain fraction of systems. The volunteer computing landscape is vastly heterogeneous so things like this can escape even a thorough testing phase.
Quote:
Good luck on finding a fix, or at the very least finding out what is causing this.
We know which things have an impact on this and we'll be working on a fix very soon. However, we might depend on your feedback in order to verify that those fixes indeed improved the situation since we probably can't test/reproduce the observed behavior using our own test systems.
When I worked in the Honeywell-Bull UNIX marketing, the computers that had survived all inhouse testing were immediately brought down by users who did the wrong things that test engineers had not even imagined.
Tullio
RE: RE: RE: AFAIK
)
Correct: I have a GTX460 in one of m machines and this does double precision. All NVIDIA cards with FERMI architecture do
But why on earth would they
)
But why on earth would they need double precision (i.e. floating point hardware) to do integer maths?
RE: But why on earth would
)
The video cards use FP for 3D rendering. OpenGL is full of matrix transforms and transcendental functions. While the results are integer very little else is.
I had the same feeling. But
)
I had the same feeling. But since I don't know PrimeGrid enough to reason about it I just stuck to their Web page. Certainly Eratosthenes' sieve does not need floating point math. But perhaps the use the Riemann's conjecture.
Tullio
RE: I had the same feeling.
)
Crunched the Proth Prime Search (Sieve)CUDA WU's on 2 Fermi-cards but cannot remember having false results.
RE: Crunched the Proth
)
Which is the quorum? On QMC quorum is 1 and so in AQUA and CPDN, since they all use the MonteCarlo algorithm.
Tullio
RE: RE: Crunched the
)
For PPSS it was 2.
RE: RE: The cross
)
Thank you for letting us know that you have seen these problems.
Good luck on finding a fix, or at the very least finding out what is causing this.
RE: Thank you for letting
)
Well, not exactly. Our tests didn't exhibit these accuracy issues and they do occur only on a certain fraction of systems. The volunteer computing landscape is vastly heterogeneous so things like this can escape even a thorough testing phase.
We know which things have an impact on this and we'll be working on a fix very soon. However, we might depend on your feedback in order to verify that those fixes indeed improved the situation since we probably can't test/reproduce the observed behavior using our own test systems.
Thanks,
Oliver
Einstein@Home Project
When I worked in the
)
When I worked in the Honeywell-Bull UNIX marketing, the computers that had survived all inhouse testing were immediately brought down by users who did the wrong things that test engineers had not even imagined.
Tullio