Thank you very much, I only needed your observations and I realize that should do everything on my own risk:) But some universal temp monitoring utility would be a good thing to have for any setup. Should we start bugging NVIDIA guys about this ?:)
This is really cool!
Have you tried to run Linpack on one of these? (Like the Massimiliano Fatica paper) My guess is that it might run as high as 400 Gflops.
–Mark
Not yet. You know how it goes in a software development shop… lots of tasks and no time for romance. :-) I’d like to get some of these assigned to do folding when they’re not in use for coding. The folding community solved all the key integration questions to run four GPU cards, so everyone here wants to return the favor by doing some folding. Talk about a great effort and a great community - the folding guys rock!
From experience I can tell you that running three (3) or four (4) GTX-295’s, at full-bore, is not possible for very long even on a 1500-watt PS. Two GTX-295’s is max even on a 1500-watt PS. Check out my post… MY POST
This doesn’t make sense. If you can run two cards OK on a 1500W PSU you probably have enough 12V on the plugs, but if you add just one more to get to three you are not coming near the combined total power of the 12V system on the Silverstone ST1500 PS. My guess is that whatever problem you are having it is not really about a fundamental limit on the number GTX-295 cards being limited to only two with that PSU, let alone with any power supply.
It sounds like your post (that you jump to) reports a problem with just one particular power supply, maybe a limit of that vendor’s power supply or with your office’s power supply. The Enermax Galaxy 1250W power supplies we use seem to run fine with four GTX-295’s day in, day out.
Enermax says it provides 30 amps “rated output” to each 12V plug. The Silverstone ST1500 PS you cite provides only 25 amps continuous output to each 12V plug, or peak power of 30 amps for 12 seconds. It could be the Enermax provides just enough more 12V or you got a dud Silverstone or those Silverstone PSUs in general don’t quite power up a GTX-295 or low office power did your PSU in. Without testing more than just one sample it is really difficult to say, but even so it doesn’t make sense to me that you can run only two cards but not three.
I’d advise: 1) Make sure you have ample power in the office. This is not easy if you have a lot of machines like this. 2) Try a different PSU, like the Enermax 1250W.
I have also run four GTX 295 cards for days on the Enermax 1250W PSU like Dimitri and not noticed any problems. Although the rated max power consumption of the GTX 295 is 289W, I measured a total power consumption from the system of about 1150W, suggesting these cards were only drawing ~230W while performing my computations. Aside from the GPUs, the only other significant power draw in the case was the Core i7-920/motherboard/12 GB DDR3. In my case, line voltage was provided by a sturdy 20A circuit in a laboratory.
My own 3x GTX295 system wattage per GPU is nearly identical to seibert’s… for me, about 220 watts per GTX295 at full load. That’s wall socket power, so it’s probably more like 180 watts in the GPU itself.
I suspect that for GRAPHICS the boards pull more wattage than for most CUDA apps… perhaps even 50% more. I’m sure that you can make the GTX295 burn 275+watts using something like FurMark.
It could also be that different CUDA apps might use a different wattage (perhaps if they were more memory intensive, or more FP, or some balance.) For my Monte Carlo work, 220 watts per board is pretty stable measurement even with different problems running.
I’ve had no problem running a i7 930 and 3 GTX295s with a 1200 Watt PSU (BFG Tech), using about 900 wall socket watts.
Some UPSs can show realtime wall socket wattage. I also have a $20 “Kill-o-Watt” meter which works well.
I run 3 gtx 295 on a thermaltake toughpower 1200W w/o problems.
I think 4 fermis will draw 1000W and the rest (cpu+ram+hd+dvd+fans) up to 250W at peak, so 1250 total.
Thanks for all of the input. My rig was on two dedicated 20-amp duplex outlets – nothing else on those circuits. And with 4 GTX 295’s running for months on end, I finally melted pin #10 (12volt) on the 24-pin motherboard connection from the power supply. And it also killed the power supply. The Silverstone website, which I checked after the burnout occurred, had a GPU SUPPORT LIST that listed the max GTX295’s at 2… on the ST1500 PS.
Yep, I have a KILL-A-WATT meter, and it showed the entire unit was drawing only about 900 to 950 watts total about two days before the burnout.
Thanks for all of the input. My rig was on two dedicated 20-amp duplex outlets – nothing else on those circuits. And with 4 GTX 295’s running for months on end, I finally melted pin #10 (12volt) on the 24-pin motherboard connection from the power supply. And it also killed the power supply. The Silverstone website, which I checked after the burnout occurred, had a GPU SUPPORT LIST that listed the max GTX295’s at 2… on the ST1500 PS.
Yep, I have a KILL-A-WATT meter, and it showed the entire unit was drawing only about 900 to 950 watts total about two days before the burnout.
Interesting.
Pulling a total of 900 to 950 watts and a 1500 watt power supply dies? Even allowing for imprecision of ratings and meter reports that’s a big gap. Together with the 12V pin melting that leads me to guess the problem is indeed topping out on the 12V rating of that particular power supply and/or what the motherboard is pulling through that particular pin, but not the overall current consumed. Try with a power supply that has a higher 12V capacity and maybe a different motherboard (Don’t know if the Deluxe you are using is the same as the ASRock “Supercomputer” configuration I am familiar with).
Now, of course, it’s time to move to the next frontier: four GTX 470’s, 480’s or Fermi Teslas.
Interesting.
Pulling a total of 900 to 950 watts and a 1500 watt power supply dies? Even allowing for imprecision of ratings and meter reports that’s a big gap. Together with the 12V pin melting that leads me to guess the problem is indeed topping out on the 12V rating of that particular power supply and/or what the motherboard is pulling through that particular pin, but not the overall current consumed. Try with a power supply that has a higher 12V capacity and maybe a different motherboard (Don’t know if the Deluxe you are using is the same as the ASRock “Supercomputer” configuration I am familiar with).
Now, of course, it’s time to move to the next frontier: four GTX 470’s, 480’s or Fermi Teslas.