Originally Posted by Timmaigh!;13619017
Thank you for your response, i will take care of my sig later.
As i said, i would rather not touch the voltage and BIOS, not only because of warranty, but generally i am not that experienced with the overclocking and too many things could go wrong in this case...so if it means, i wont be able to go beyond 700 MHz, so be it. I can live with that, but i would kill myself if i killed the card.
Still this does not mean, i do not want to OC it at all, i want to do it, but eliminate the risks as much as i can. It seems that keeping voltage at stock value is the basic step :-P BTW you said 0,925, is that supposed to be stock? GPU-z says 0,912 in my case, is this the difference due to the Vdrop/vdroop like with CPUs? (i am bit more experienced with OCing those). Or am i missing something?
Sorry I am so selfish, I often forget their are other brands out there
, Gainward is my favorite brand but tough to get here State-side. I remember the best of the best Geforce 2s/4s were Gains and Chaintechs.
The stock voltages for EVGA are 0.925 with a core clock of 630 MHz on Core. The memory I believe is all stock with exception of what PoV or was Palit as they have an agressively clocked card.
Yesterday for a few minutes i tried the 700 MHz clock with Octane and everything looked normal, no crash or explosion
so i guess it might work at least at this frequency OK and there is a performance increase. Octane states in the main window, how fast is it rendering in Megasamples, on my benchmark scene i get 5,34 on 607, 5,82 on 670 and 6,03 on 700... so it scales well so far...
i am really tempted to try up to 775 though, the guy on the Refractive Software forums did it and the Wizzard of TPU did it as well (before killing the card while overvoltaging)...
I am bit hopeful this might be my case:http://forums.nvidia.com/index.php?showtopic=197878
I have seen almost every dead card related to over voltage. The Power regulation system has been called into question numerous times but the truth of the matter it seems to be running stock fine, and have plenty of over-head for a bit of overclocking with stability. I achieved 830 MHz @ 1.038~1.05v stable enough for 4, maybe 5 runs of the Heaven benchmark. I am also on water so take that with a grain of salt, as every dead card I have seen has not been on water.
Less powerdraw = higher clocks, right? OFC under condition, the cores are cherry picked enough to be stable on default volage there...
Less power draw = less power draw. A higher achieved clock ratio will produce more heat and require more power to draw upon. This being said there are many variables to consider especially the Voltage Regulation, and it may seem like you are not drawing any more power because the numbers change, it is just reaching that voltage cap of whatever it is said to. It seems a bit confusing but for example, you set it to 0.913v, it reads 0.913v, but the VRM is actually at 0.900v and will provide enough voltage to the run the GPU up to 0.913v.
You set it to 0.925v and it reads 0.925v but the VRMs might still read 0.900v because you haven't increased the draw of power through the GPU. VRMs are allowed to allow enough juice to go through up to 0.925v, however it gates the current, but it doesn't have to be 0.925v. On these newer Geforce cards though, I have noticed, if you set the value to 0.925v, it immediately gives it that current and the VRMs compensate and fluctuate to above and below the value. In essence, the gap has greatly decreased. Think of it as Load-Line Calibration in a sense, however its sometimes hard to read without a voltmeter. Our third party sensors aren't as accurate as we like to believe and many of the values we see are as off as +/- 15% on a lot of them, but for the most part they are good enough to get the job done.
That is my current understanding of how these more complicated VRMs work, and I maybe off slightly, but I think I got the general idea of how it works. Sometimes the system can provide extra performance with no added juice. Sometimes you need a nudge of extra voltage to increase the speed just a fraction, but by doing so, you are able to give yourself even more headroom to go even faster.
Indeed, it can always crash after 2 hours, but i believe it would be stable. Maybe if i tried it with Furmark/3dMark11, it would crash on these settings within first minute, but i am not going to do it, as the performance within Octane is what is important to me.
3DMark11 is your best bet, Furmark is good for seeing max temperatures really. 3DMark11 also has another constraint on it called the PDL. The Power Draw Limit is something that appears to be a non-hardware type of throttling (slowing down) of the GPU. We worked it out that as the Power Draw increases, a limiting performance bouncing effect occurs of jumping between the core clocks you have set, and the PDL "safe" underclock which I think is 553 MHz.
It easy to think of this as an oscillating function that will diverge to a specific benchmark score, for those that want a visual, I guess a converging alternating series function would suffice.
Core @ Time: 0.00s = 775 MHz
Core @ Time: 0.15s = 553 MHz
Core @ Time: 0.30s = 775 MHz
I believe the only way it can happen "without being seen" is that it is happening so fast the the GPU monitoring program isn't "polling" the change, so we are talking really fast up and down frequency changing.
We found to avoid this invisible limitation, to aim for the 725 MHz with a Max Voltage of 0.980v if at all possible, even lower to 0.968v if you have what you called a "cherry GPU."
The question now is, should i try even more MHz, given the 90C temps... it was 88C max on 670, so assuming it would work at 770, i suppose it could be 92C there...Maybe if i tried it with Furmark/3dMark11, it would crash on these settings as well, but i am not going to do it, as the performance within OCtane is what is important to me.
Another question, if its stable within Octane while both cores are working at 95-99percent on these 723Mhz clocks, is it granted to be stable while gaming with ONE core only? Or not at all?
To be honest, the GTX 580 TJMax is I believe 105c. I wouldn't recommend keeping a GTX 580 GPU any higher than 90c to expect a decent life span 2-3 years+
For a GTX 590, I would guess that to be safe, keep it below 85c as its a more delicate flower.
Only way you know if it is gaming stable is if you test it Edited by RagingCain - 5/24/11 at 6:55pm