Cuda on GEForce 210?
 

For reasons unrelated to Gimps, (my flat panel at work died, I wanted an upgrade to dual monitors) I purchased a GEForce 210 card with 512 Meg of RAM. Yeah, not a lot compared to the higher end 460s and such. I was wondering
a) what is the best type of GIMPS work to put it to, assuming I don't want the GIMPS part loading down the Pentium IV host,
b) Is it worth asking it to trial factor (or even P-1 factor) my favorite unfactored small mersenne number, 2^1061-1 ?, and if so, how would I do it? (or maybe I should be looking at other exponents under 10K, since NFS@home and I might get into a race condition).

:deadhorse::cmd::devil:

Hi, i´m using Nvidia 210 GT for TF to lower limits. Results are regular. 1 exponent checked 64 to 65 bits/7min, doing the same with a 955 BE overclocked it´s around 10 minutes per exponent.
Best regards, Carlos

[quote]b) Is it worth asking it to trial factor (or even P-1 factor) my favorite unfactored small mersenne number, 2^1061-1 ?[/quote]
AFAICT, 2^1061-1 is nearing the point where it's considered ready for NFS - if it's not there already - after a large amount of ECM work at high B1 values was performed on it.
Even P-1 would be very unlikely to find a factor :smile:

[QUOTE=Christenson;254681]... since NFS@home and I might get into a race condition[/QUOTE]

Although the status page is not yet updated, you may wish to try connecting to the project... :smile:

Can I get some help in the practicalities of setting up the card? The OS is Win XP 32 bit.
Yes, I have been reading the NFS@home people, even here on mersenneforum.org, and they definitely have M1061 in their sights, although the sieving may not *quite* have begun. Thus the need for more suggestions; M1061 will very likely be factored this year, and I might even help the sieving with one of my regular CPU cores.

[QUOTE=Christenson;254706]
Yes, I have been reading the NFS@home people, even here on mersenneforum.org, and they definitely have M1061 in their sights, although the sieving may not *quite* have begun. Thus the need for more suggestions; M1061 will very likely be factored this year, and I might even help the sieving with one of my regular CPU cores.[/QUOTE]
I think you missed the point... :smile:

I did not miss that work on M1061 by my non-sieving CUDA device has a reasonably high probability of only warming up my work area and raising my electric bill. That's why I asked for other targets. :tantrum:

The question remains: Do we have a practical TF or P-1 implementation for CUDA on 32-bit WinXP for my GEForce 210 with 512K of RAM? Where do I download it from? [And if it works, maybe I'll put a 460 on this or my next Linux compute box and find a few factors with it]

[QUOTE=pegaso56;254686]Hi, i´m using Nvidia 210 GT for TF to lower limits. Results are regular. 1 exponent checked 64 to 65 bits/7min, doing the same with a 955 BE overclocked it´s around 10 minutes per exponent.
Best regards, Carlos[/QUOTE]
[b]Christenson,[/b]

On Windows, you can use mfaktc (Trial division with Cuda thread) for cuda trial factoring. It is slower on lower end cards, the 210 being here just 50% faster than a core of an overclocked 955BE.
I use a 220, which is about 3 times as fast as a 2.4GHz Core-2 quad (Q6600) core. (But I really want to upgrade it perhaps to a 430 or 450.)
The 460 is many, many times faster than a 210 for tf and can _also_ do Cuda LL using CudaLucas (MacLucasFFTW thread). (You may need to upgrade your power supply.)

Finally got mfaktc running on the GEForce 210 tonight. Roughly 3.7+Ghz days/day at TF, (remembering that the host is just a Pentium 4) with Maxstreams=2, and screen response for rest of system is unreasonably slow, not usable for proper work but I don't care when it's running overnight.
I will try CUDALucas tomorrow, with luck.