mfaktc: a CUDA program for Mersenne prefactoring

[RichD]

Originally Posted by ET_

Reaching 2.5 PetaFLOPS with an incrastructure holding 7,000 Intel cores and 14,000 nVidia graphic cpus.

The numbers are reversed but who's really counting. :-)

[fivemack]

Also, it's 14336 (14<<10) six-core CPUs (multiplying up to a peak 2.93GHz * 4 FLOP/cycle * 12 cores/box * 7168 boxes of just over one petaflop) and 7168 Fermis - so 7168 boxes each with dual hex-cores and a Fermi. Not a cheap machine: maybe 400 million yuan.

[Lorenzo]

Hello!

Why not windows version of mfaktc 0.13?

[amphoria]

Originally Posted by Lorenzo

Hello!

Why not windows version of mfaktc 0.13?

A version compiled for Windows can be downloaded from here:

http://www.sendspace.com/file/obirpl

[Svenie25]

Originally Posted by amphoria

A version compiled for Windows can be downloaded from here:

http://www.sendspace.com/file/obirpl

Thanks a lot. But there is no readme or any other file how you can use it. Or is there a a description somethere in these thread?

[TheJudger]

Just take the README.txt from the source tarball.

Oliver

[Svenie25]

Originally Posted by TheJudger

Just take the README.txt from the source tarball.

Oliver

Thanks a lot. Found it. ;)

[Lorenzo]

Originally Posted by amphoria

A version compiled for Windows can be downloaded from here:

http://www.sendspace.com/file/obirpl

Thank you!

[nucleon]

Thanks for the windows executable.

I've been running it. I have a nvidia 460GTX running on Win7. I've been grabbing TF work manually. On my machine it seems to average about 2GHz-days work per hour.

Is that what I should expect?
(I think that figure is awesome btw)


-- Craig

[amphoria]

Originally Posted by nucleon

Thanks for the windows executable.
I've been running it. I have a nvidia 460GTX running on Win7. I've been grabbing TF work manually. On my machine it seems to average about 2GHz-days work per hour.

Is that what I should expect?
(I think that figure is awesome btw)

On a GTX465 (the nearest equivalent I have) I am getting the equivalent of 4.6 GHz days per hour on 100M digit exponent. This is on 64-bit Linux but the difference between Linux and Windows is negligible on v0.13. However, the size of the exponent does matter. If you are testing a smaller exponent then it will run slower.

[TheJudger]

Hi Craig,

Originally Posted by nucleon

Thanks for the windows executable.

I've been running it. I have a nvidia 460GTX running on Win7. I've been grabbing TF work manually. On my machine it seems to average about 2GHz-days work per hour.

Is that what I should expect?
(I think that figure is awesome btw)


-- Craig

please provide some more information:
- exponent size
- bit level(s)
- CPU type/clock
- settings in mfaktc.ini (if changed from default)

When providing performance numbers please copy&paste one or two lines of the mfaktc output starting with "class xxxx:". I'm not thinking in GHz-days. A GTX 460 should provide similar speed to a GTX 465, perhaps a little bit slower.

If the CPU ist fast enough you should see something like 150M/sec for a exponent around 100.000.000 at bitlevels about 64.

Oliver