(I orginally had this posted in the Number Crunching message board and thought this board would benefit from it as well...)

FYI...

For those who want to squeeze out some more GFlops out of your GTX 590's (but are as scared as I am with increasing core voltage), it seems the cards are happy with the 700/1825 setting at stock voltage, increasing the GFlops from 1290 to 1434...

WirelessDude

FYI...

For those who want to squeeze out some more GFlops out of your GTX 590's (but are as scared as I am with increasing core voltage), it seems the cards are happy with the 700/1825 setting at stock voltage, increasing the GFlops from 1290 to 1434...

WirelessDude

What are your GTX590's stock voltages?
I'm using at 700MHz my GTX590 since the beginning, on 913 and 925mV. (memory at stock)
Yesterday - inspired by your post - I've raised the GPU's clock to 725MHz, and I'm planning to go further... I've installed a 12cm 2500rpm cooler fan on top of my GTX590, blowing cool air directly to the card's cooler through some air duct made of cardboard paper by my gf. My GPU's temperatures as low as 71-74C.
Since the second release of GTX590's driver, it's not allowed to rise significantly the voltages of the GPU, so I'm hoping it won't hurt the card.

I took a deep breath, and I've raised both GPUs' clock on my GTX590 to 750MHz.
It soon turned out, that my GPU0 (the one that runs on 913mV) can't handle 750MHz, so I took it back to 730MHz. The other (on 925mV) however, is running fine on 750MHz for more than an hour now.
I don't understand, why are the maximum voltages different on the same card?

The other (on 925mV) however, is running fine on 750MHz for more than an hour now.

Unfortunately it's run to an error after 3 hours, so I took it back to 740MHz.

I don't understand, why are the maximum voltages different on the same card?

2 Different GPU's, each with different stock voltages, set in factory.

Hmmm... I'll have to give those extra MHz a try on my liquid-cooled and air-cooled GTX 590 setups and see if they hold.

Thanks for the inspiration! ;o)

WirelessDude

I don't understand, why are the maximum voltages different on the same card?

2 Different GPU's, each with different stock voltages, set in factory.

The factory settings are 900mV for both GPUs.

913mV and 925mV are close, but not the same. Anyway they are different chips, and all chips are different. Both are well below my 1.025V AMP ed GTX470 (FOC), and my 2nd ed ASUS (sub 1V), so I would suggest you have a bit of scope to play with, if doable and if power usage and cooling are not obstacles. Try reducing the GDDR first though.

When future apps come out, you will have to test again as they may well use more juice, and don't forget to clean the GPU, periodically.

What are your PCB temps?

... Try reducing the GDDR first though.

I will do it later. I returned to 725MHz, and now I'm trying to make my i7-970 run steadily at 4.5GHz. I realized that I can gain some extra percent of GPU usage from overclocking not only the Core2s but the Core i7 too. Actually, My i7-970 were running at 4GHz before, but there is so much combination of the four multipliers (CPU, RAM, QPI, and UnCore) and their corresponding voltages, that I didn't have time to play with them enough. Until now...
It's interesting: my PC can crunch for hours, then I start browsing on some adobe flash enriched sites, and I start receiving error messages from different applications, and some blue screens too :) as a result, a couple of my GPUGrid tasks have failed.

When future apps come out, you will have to test again as they may well use more juice, and don't forget to clean the GPU, periodically.

I'm doing these things regularly.

What are your PCB temps?

My PCB is 51°C (could be that is there only one sensor on my MB?)
CPU cores are 60-68°C (at 4.5GHz, core voltage: 1.475V)
This PC has no case.

The GPU PCB temps look OK; my FOC GTX470 PCB temp is 49deg C. Yeah, would only be one temp - one board.

I would be slightly worried about heat radiation from the CPU heatsink if overclocking too high; the more you overclock the higher the temps and the more heat has to be expelled. Even with a high end cooler and the case off, it could be an issue, and not just to the GPU; to the RAM and MB too.

I'm sure you know that sometimes just moving the case slightly can reduce CPU temps by several degrees (for example if the case sits in a corner/confined space) - so for others.

If you OC and find reliable clocks, just bear in mind that when you are crunching and doing other activities, the extra burden can cause failures.
Don't know of a flash test for OC'ed CPU's, but it's generally a good idea to find a stable clock and then drop it a few notches.

I'm aware these things, and I agree with you. I'm (leastways I was) a quite experienced overclocker. The weather turned cold here, so it's a good time to experiment with overclocking. I can boot up even at 4.6GHz, so that's why I thought I can make my PC run steadily at 4.5GHz. There is so much misinformation on the internet, and also in the BIOS about overclocking. There is a huge difference in overclocking Socket 1156 and Socket 1366 CPUs, and I have both...

The GPU PCB temps look OK; my FOC GTX470 PCB temp is 49deg C. Yeah, would only be one temp - one board.

I misunderstand you, because my GTX 590 and 580 tricked me. There is no PCB temperature reading on them. So I told you my motherboard's temperature.
The GTX 590's metal back plate and the metal heat spreader is over 55°C (because I can't touch it for over 5 sec...) Both GPUs are 69°C right now on the 590.

55 deg.C might be a bit on the hot side. GPUZ tells me what the PCB of the GPU is (usually 46 to 49), but it might not report it on GTX500 cards, or just the 590?
I once placed an exhaust fan one slot up from the back of the GPU, and found that it significantly reduced the temp of the backplate (warm to the touch rather than hot), but it depends on case space.
On another system, I still have a fan at the rear of the case (outside) pulling air out from a GTX470. Again makes a difference.

I, too, was concerned with hot (unbearably hot!) GTX 590 fan-cooled GPU's (almost 100C full blast!!!) But, after a couple months of 24/7 at mid-90C temps, they seem to keep on performing. I also noticed that the fans are at 50% in the 80C's and don't hit 100% until the very high 90C's. You cannot cool them down any further when running them full blast. I have the case air-cooled with over 350 CFM air flow which keeps everything else in the system very cool, even with the hot air coming from the internal exhausts of the GTX 590's (I have two of them in one system at this time.)
---
Not that I will be keeping them in this mode for long. I am in the midst of converting all air-cooled systems to liquid-cooled with just these two GPU's to go and any further GTX 590's immediately being converted to liquid-cooled (I really like Alphacool's water block/backplate solution -- comes with everything you need for the conversion and has a very pretty black and copper look to it...)
---
In my largest number cruncher, I am running 4x GTX590's and 2x Intel X5650 CPU's, liquid-cooled with an external Black Ice Extreme 360 w/3x BGears b-Blaster 103 CFM fans (running 100% - very quiet!) before the fluid is fed into my Koolance EXOS 2.5 (3x 120mm aluminum radiator/fan system - also a quiet system running 100%!) The 8x GPU's run anywhere from 47-61C at max...
---
WirelessDude

I, too, was concerned with hot (unbearably hot!) GTX 590 fan-cooled GPU's (almost 100C full blast!!!) But, after a couple months of 24/7 at mid-90C temps, they seem to keep on performing.

Impressive systems!

Regarding reliability: as a rule of thumb the mean time between failures is cut in half for every 10 K more. So running hot doesn't mean you're chips are suddenly going to die (not before some really temperatures), it's just that they're wearing out faster. What this means for your specific chip is hard to tell without taking it apart and examining it for years in an electron microscope. And if this matters to you or not is an entirely different question ;)

MrS

---

Scanning for our furry friends since Jan 2002

Thanks for that info! Thank goodness for lifetime warranties!!! ;o)
---
WirelessDude

I had to set my GTX 590's clock down to 715MHz, because since the last driver update (285.58) I got a couple of BSOD @725MHz. This very highly GPU utilizing (99%) NATHAN tasks turned the heat up to 80°C in my GTX 590 (@725MHz), so it is a good idea to take back a little from overclocking anyway.

I had reverted back to the WHQL 275 drivers due to some rare unrecoverable "error while computing" anomalies on my 8x GPU system using either WHQLs 280 and 285. I'm not having such issues on my 4x GPU systems using the latest WHQL 285 drivers, even on those 90%+ GPU load WUs (fan- or liquid-cooled...)
---
WirelessDude

I've got the following NATHAN Long running times (with swan_sync=0 applied) on my main cruncher host:
GTX 580 running at 850MHz: 10800s (3 hours) 99% GPU usage
GTX 590 running at 715MHz: 18300s (5 hours and 5 minutes) 99% GPU usage. another one
If I take the running time of the GTX 580 as the base, it should be 12800s (3 hours and 34 minutes) considering the ratio of the two card's clock speeds.
But in reality it's 43% slower than the ideal.
I also checked the performance of WirelessDude's hosts.
Host 109655 has 4 GTX 590's at 630MHz: 15400s (4 hours and 17 minutes)
Host 107252 has 2 GTX 590's at 630MHz: 15000s to 16600s (4 hours and 37 minutes)
Host 109318 has 2 GTX 590's at 630MHz: 14600s to 15300s
His hosts' running times are not as bad as mine's, but it's still up to 14% slower than ideal.
It means that the GTX 590 is throttling down its clock while these highly GPU utilizing NATHAN long wus running are on them.
Especially mine, which is overclocked.
I have to reduce the overclocking of my GTX 590 as the tasks' GPU utilization grows. Possibly a slight underclock could cause performance gain on (multi-) GTX 590 systems, when running highly GPU utilizing tasks simultaneously on a single card.

43% is far too great for it to be just PCIE contention/bandwidth (and I doubt that is an issue anyway).

As well as possible downclocking, most likely, it could be recoverable memory losses due to memory rates being slightly too high.

WirelessDude is using W7, which would account for at least 11% of the 14% drop. CPU difference might account for the other 3%.

I think your 17% OC is too much. Try 10% and stock GDDR rates.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

WirelessDude is using W7, which would account for at least 11% of the 14% drop. CPU difference might account for the other 3%.

This is true with other type of wus, but the running times of a lot of NATHAN Long wus on WirelessDude's W7 hosts is in the 0-3% range of the running times (multiplied by the ratio of the clock rates) on my W-XP host.

I think your 17% OC is too much. Try 10% and stock GDDR rates.

Since then the I've tried 700MHz (15.3%) and 660MHz (8.7%). The throttling is still kicked in at these frequencies. (while I didn't lower the clock of the other GPU on my GTX 590). GDDR is at stock.