mersenneforum.org - RPS 10th Drive: Back to the Low-Weights

mersenneforum.org (https://www.mersenneforum.org/index.php)

- Riesel Prime Search (https://www.mersenneforum.org/forumdisplay.php?f=59)

- - RPS 10th Drive: Back to the Low-Weights (https://www.mersenneforum.org/showthread.php?t=11858)

738-746 complete, no primes.
Taking 778-786.

736-738 complete, one prime.

Already testing 776-778. I marked it on the top page but forgot to announce here.

746-776 complete (3 primes)

778-786 complete, 2 primes.
Taking 798-822.

The sievers should consider using a GPU to sieve, amphoria? I suppose sieve speed should be quite fast with only 128 k's.

[QUOTE=em99010pepe;242015]The sievers should consider using a GPU to sieve, amphoria? I suppose sieve speed should be quite fast with only 128 k's.[/QUOTE]

I already tried GPU sieving for this drive on a relatively cheap ATI Radeon 5670. It reaches only about 85% of the speed compared to a single cpu core (Q6600) running sr2sieve, and the GPU sieve thereby utilizes a full cpu core too! Not to mention the memory requirement (about 2 GB vs. a few MB using sr2sieve)...

[QUOTE=Thomas11;242138]I already tried GPU sieving for this drive on a relatively cheap ATI Radeon 5670. It reaches only about 85% of the speed compared to a single cpu core (Q6600) running sr2sieve, and the GPU sieve thereby utilizes a full cpu core too! Not to mention the memory requirement (about 2 GB vs. a few MB using sr2sieve)...[/QUOTE]
ppsieve's strong point is more in the way of large continuous blocks of k and n, rather than a handful of individual k's. I had a PM discussion with Ken (ppsieve's author) a while back about this and he explained that this is due to the rather different algorithm ppsieve uses as compared to sr2sieve. This algorithm allows for huge speed boosts on large continuous blocks, but totally stinks on just a few k's.

From what I gather, the sr2sieve algorithm does not lend itself nearly as well as ppsieve's to being ported to run on GPUs. At any rate, Ken said that he wasn't planning to undertake such a project in the forseeable future so for now, sr2sieve on CPUs is still the way to go for smaller sieves like this.

[QUOTE=mdettweiler;242233]This algorithm allows for huge speed boosts on large continuous blocks, but totally stinks on just a few k's.
...
sr2sieve on CPUs is still the way to go for smaller sieves like this.[/QUOTE]

Thanks for that explanation. I will stay with sr2sieve for this drive...