Best type of work to get for a GeForce GTX 970?
 

I've been doing only CPU testing so far and would like to try to start doing GPU testing as well.

I've got an old Gigabyte Geforce GTX970. Are there any advantages/disadvantages of doing TF using mfakt vs. doing LL using CUDALucas? Or it doesn't really matter?

Thanks.

[QUOTE=ZFR;535578]Or it doesn't really matter?[/QUOTE]

A GTX 970 will get [URL="https://www.mersenne.ca/mfaktc.php?show=714&sort=ghdpd"]about 1175 GHzD/D doing TF work[/URL]. So, [URL="https://www.mersenne.ca/cudalucas.php?model=567"]according to this graph[/URL], it is "optiomal" to TF up to 77 "bits" before doing a LL run ***on that same card***.

Also, a plug... We could [B][I][U]really[/U][/I][/B] use some more [URL="https://www.gpu72.com/"]trial factoring[/URL]... The tsunami soon arrives!!!

P.S. Importantly, do whatever you enjoy; it's your kit/time/electrons. And know that you'll never find an MP doing TF'ing. You will, however, help someone else find one just a little bit sooner...

[QUOTE=chalsall;535579]A GTX 970 will get [URL="https://www.mersenne.ca/mfaktc.php?show=714&sort=ghdpd"]about 1175 GHzD/D doing TF work[/URL]. So, [URL="https://www.mersenne.ca/cudalucas.php?model=567"]according to this graph[/URL], it is "optiomal" to TF up to 77 "bits" before doing a LL run ***on that same card***.

Also, a plug... We could [B][I][U]really[/U][/I][/B] use some more [URL="https://www.gpu72.com/"]trial factoring[/URL]... The tsunami soon arrives!!!

P.S. Importantly, do whatever you enjoy; it's your kit/time/electrons. And know that you'll never find an MP doing TF'ing. You will, however, help someone else find one just a little bit sooner...[/QUOTE]
GTX970 is nominally 388GhzD/day at TF, 23 LL per mersenne.ca tables. About equivalent to a GTX1060, not faster than a GTX1080. Actual numbers depend on exponent size, and bit level in TF, or fft length in LL, unavoidably and naturally. So when you see variation from the nominal, it doesn't necessarily mean anything is wrong.

Best depends on your goal. Maximum contribution to the project from the gpu is probably mfaktc. Get reserved assignments, tune mfaktc for best output rate. Have fun tinkering and contributing. No problem running prime95 and mfaktc etc on the same system.
And you might find something useful by skimming through [URL]https://www.mersenneforum.org/showthread.php?t=24607[/URL]
Use of GPU72 or various client management software versus entirely manual independent operation is a personal choice. I am in favor of people knowing how the app operates by hands-on use initially.

[QUOTE=kriesel;535580]GTX970 is nominally 388GhzD/day at TF, 23 LL per mersenne.ca tables.[/QUOTE]

Oh, shoot. Thanks for the correction; I must have copy-and-pasted erroneously. The "economic cross-over" graph I linked to is still correct though: 77 bits is "optimal".

Thanks. I'll start with TF, and see how it goes from there.

[QUOTE=chalsall;535579]The tsunami soon arrives!!!
[/QUOTE]

What do you mean by this? Is there going to be an influx of factoring jobs soon?

[QUOTE=ZFR;535603]Is there going to be an influx of factoring jobs soon?[/QUOTE]

No... The LL'ing throughput of Primenet has quadrupled in the last few weeks, so the TF'ing effort is having to readjust its targets in order to deal with the soon-to-come large number of LL requests in the higher 9xM ranges.