Tesla cards (specifically C1060)

[MacFactor]

I've seen scarcely any mention of Nvidia Tesla cards in this forum -- is there *anyone* out there who can give me pointers on how to get started with Teslas ? I'm afraid the C1060 may not be up to the job, even though it's CUDA capable -- double-presision is only 78 Gflops, vs 933 for single precision.

Anyone have experience with these cards and CUDALucas or mfactc ? Or whatever the CUDA P-1 program is called ?

thanks, please talk beginner level ...

CE

[kladner]

LaurV is your man for this info. He will probably have something to offer on this subject.

[xilman]

Originally Posted by MacFactor

I've seen scarcely any mention of Nvidia Tesla cards in this forum -- is there *anyone* out there who can give me pointers on how to get started with Teslas ? I'm afraid the C1060 may not be up to the job, even though it's CUDA capable -- double-presision is only 78 Gflops, vs 933 for single precision.

Anyone have experience with these cards and CUDALucas or mfactc ? Or whatever the CUDA P-1 program is called ?

thanks, please talk beginner level ...

CE

I have a C1060.

Feel free to ask questions here. We can take it to PM or email if they get too specialised for a forum audience.

[kladner]

Originally Posted by xilman

I have a C1060.

Feel free to ask questions here. We can take it to PM or email if they get too specialised for a forum audience.

I, for one, would find the discussion interesting.

[chalsall]

Originally Posted by MacFactor

I've seen scarcely any mention of Nvidia Tesla cards in this forum -- is there *anyone* out there who can give me pointers on how to get started with Teslas ? I'm afraid the C1060 may not be up to the job, even though it's CUDA capable -- double-presision is only 78 Gflops, vs 933 for single precision.

I regularily "spin up" cg1.4xlarge instances on Amazon's EC2 for about $0.135 an hour. They each host 2# Tesla M2050s.

I can't speak about anything except TF'ing, but they perform better than the simple number of cores might suggest.

If you'd like, I'd be happy to run some benchmarks for LL'ing and P-1'ing.

[Mark Rose]

I'm curious if mfaktc's GPU sieving can work on the compute 1.3 cards. Last I heard it couldn't.

[TheJudger]

Originally Posted by Mark Rose

I'm curious if mfaktc's GPU sieving can work on the compute 1.3 cards. Last I heard it couldn't.

mfaktc 0.21 will do, expect around 50-60GHz equivalent... not really worth once you have to pay for electricity. There is no future for CC 1.x cards.

• CUDA 7.0 has no support for CC 1.x cards
• mfaktc 0.22 will drop support for CC 1.x cards, too.

Oliver

[Mark Rose]

Originally Posted by TheJudger

mfaktc 0.21 will do, expect around 50-60GHz equivalent... not really worth once you have to pay for electricity. There is no future for CC 1.x cards.

• CUDA 7.0 has no support for CC 1.x cards
• mfaktc 0.22 will drop support for CC 1.x cards, too.

Oliver

Gotcha. I'm still using 4 cards that produce 50 GHz or under, but I only pay for the electricity for one. Much more efficient than the the CC 1.x cards though.

[frmky]

I still have a S1070 chugging along, which contains 4 C1060's, but they really are obsolete now.

[LaurV]

Originally Posted by kladner

LaurV is your man for this info. He will probably have something to offer on this subject.

Me? No, no, no, I only know** the Roadster, few years ago when they came out with it and asked for kickstart (the offer was for about USD 40k-50k per piece), I was almost to deposit the starter, hehe..., luckily mrs LaurV had more sense than myself...

---------------

** Note that the current title of the thread is "Tesla cars", I don't know if it morphed or it had a typo from the beginning, as it is now the first time I see it.

[VictordeHolland]

Ssttt, don't ruin the joke! (I just wanted to use the smilie with my name in it).