Recommended GTX 760 clockspeed?

UBR47K · 2015-12-16, 02:05

I own a factory overclocked GTX 760 (non-reference)
Core specs: (Base:1085MHz / Boost:1150 MHz) / Reference: (Base: 980MHz / Boost: 1033MHz)
Memory: 6008MHz

I am thinking on using this card with CUDALucas but it throws errors at the "-r 1" self-test and any exponent test.
Since I will be using Linux, I will flash a modified BIOS to it containing the new clockrates and fan speeds.

Should this card be used for CUDALucas or TF? Any suggestions on what should I set the clockspeeds?

kladner · 2015-12-16, 02:18

Try turning up the core voltage in steps. I think this makes the DP functions more reliable, while the DP throttling keeps the card from overheating. It depends on how many times you want to run -r 1.

The other thing to try is turning down the memory clock. 100 MHz steps have worked for me. Of all the (consumer) CUDA cards I have run, only one has been able to do CuLu with the memory clock at the card's stock speed.

Are you familiar with MSI Afterburner? You can use it to control and monitor your GPU's condition. It is what I use to play with frequency and voltage.

Also, there is memtest for CUDALucas. Someone else will have to explain that. I get by, but would not want to advise anyone else. Memory stability is one of the key elements of success with CuLu.

LaurV · 2015-12-16, 04:41

Additional to what kladner said - he didn't say about TF - this card is average-to-good at TF, so if you can't make it reliable at LL testing, then dld mfaktc 0.21, dld misfit, make a gpu72 account, and give gas to it. It can produce over 200GHzDays/Day of TF. The TF does not use (much) memory, so even if the card won't prove to be reliable at LL (this depends on each particular card, and not about the type nor the manufacturer), it still can be very reliable at TF.

Mark Rose · 2015-12-16, 06:31

My GTX 760 that will do about 260 GHzd/d with the factory overclock. I could push it higher, but I'm lazy.

TheJudger · 2015-12-17, 18:30

Please keep in mind that mfaktc is very bad in detecting hardware issues. This is caused by design (of mfaktc) and how TF is done in general.

Oliver

chalsall · 2015-12-17, 18:57

Quote:

Originally Posted by TheJudger

Please keep in mind that mfaktc is very bad in detecting hardware issues.

An important thing to be reminded of. Thanks.

UBR47K · 2015-12-20, 17:22

Lowering the memory clock by 105MHz made the card LL reliable, 1 match so far and 3 -r 1 tests passed.
Thanks for the input.

2015-12-16, 02:05	#1
UBR47K Aug 2015 2²·17 Posts	Recommended GTX 760 clockspeed? I own a factory overclocked GTX 760 (non-reference) Core specs: (Base:1085MHz / Boost:1150 MHz) / Reference: (Base: 980MHz / Boost: 1033MHz) Memory: 6008MHz I am thinking on using this card with CUDALucas but it throws errors at the "-r 1" self-test and any exponent test. Since I will be using Linux, I will flash a modified BIOS to it containing the new clockrates and fan speeds. Should this card be used for CUDALucas or TF? Any suggestions on what should I set the clockspeeds?

2015-12-16, 04:41	#3
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 10100000101001₂ Posts	Additional to what kladner said - he didn't say about TF - this card is average-to-good at TF, so if you can't make it reliable at LL testing, then dld mfaktc 0.21, dld misfit, make a gpu72 account, and give gas to it. It can produce over 200GHzDays/Day of TF. The TF does not use (much) memory, so even if the card won't prove to be reliable at LL (this depends on each particular card, and not about the type nor the manufacturer), it still can be very reliable at TF. Last fiddled with by LaurV on 2015-12-16 at 04:45

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GTX 1080 ... which is recommended...	petrw1	GPU Computing	27	2017-12-14 14:54
Current recommended TF bit depth?	endless mike	GPU Computing	3	2015-08-07 23:00
Some recommended sequences	schickel	Aliquot Sequences	7	2009-07-25 10:00
Recommended TF bit levels for M(>10^8)	NBtarheel_33	Math	19	2008-11-03 17:19
Recommended versions Prime95	markhl	Software	4	2008-08-04 13:46

2015-12-16, 02:18	#2
kladner "Kieren" Jul 2011 In My Own Galaxy! 2·3·1,693 Posts	Try turning up the core voltage in steps. I think this makes the DP functions more reliable, while the DP throttling keeps the card from overheating. It depends on how many times you want to run -r 1. The other thing to try is turning down the memory clock. 100 MHz steps have worked for me. Of all the (consumer) CUDA cards I have run, only one has been able to do CuLu with the memory clock at the card's stock speed. Are you familiar with MSI Afterburner? You can use it to control and monitor your GPU's condition. It is what I use to play with frequency and voltage. Also, there is memtest for CUDALucas. Someone else will have to explain that. I get by, but would not want to advise anyone else. Memory stability is one of the key elements of success with CuLu.

2015-12-16, 06:31	#4
Mark Rose "/X\(‘-‘)/X\" Jan 2013 2·1,549 Posts	My GTX 760 that will do about 260 GHzd/d with the factory overclock. I could push it higher, but I'm lazy.

2015-12-17, 18:30	#5
TheJudger "Oliver" Mar 2005 Germany 2×557 Posts	Please keep in mind that mfaktc is very bad in detecting hardware issues. This is caused by design (of mfaktc) and how TF is done in general. Oliver

2015-12-20, 17:22	#7
UBR47K Aug 2015 2²×17 Posts	Lowering the memory clock by 105MHz made the card LL reliable, 1 match so far and 3 -r 1 tests passed. Thanks for the input.