PCIE slot power (Very Successful Experiment Overclocking Maxwell Quadro)

[oobymach]

sorry man but maxwell left kepler in the dust ocing. the 780 and 780ti i had got 300-350mhz whereas both 980tis got 400-450mhz. its no contest. not to mention kepler was a pita to unlock the voltage, because if you didn't unlock voltage control - you would get nowhere ocing.

Both my 750ti and 760 I had no luck but this was way back before I knew about such things, the 750ti was maxwell but the 760 was kepler. I know the ram took little to oc but never got anywhere on the gpu but as you mention it might have been that I didn't give it enough juice. The 750ti was pci slot powered.

 

[looniam]

Both my 750ti and 760 I had no luck but this was way back before I knew about such things, the 750ti was maxwell but the 760 was kepler. I know the ram took little to oc but never got anywhere on the gpu but as you mention it might have been that I didn't give it enough juice. The 750ti was pci slot powered.

voltage is a thing but honestly you're probably more relevant talking your smaller chip than i am throwing my big chips around.

 

[damric]

Had one of my buddies not linked me to the version of nvflash that bypasses certifications, this would not have been possible. I may go back and try getting more clocks out of a GTX 750 next week. I was never successful at modding that card's firmware with the regular flash.

 

[eidairaman1]

Had one of my buddies not linked me to the version of nvflash that bypasses certifications, this would not have been possible. I may go back and try getting more clocks out of a GTX 750 next week. I was never successful at modding that card's firmware with the regular flash.

Well youve reached the boards max power on the pcie slot.

 

[damric]

Well youve reached the boards max power on the pcie slot.

I did not include the Furmark data, but it was more interesting. I think the Nvidia driver does something to hold that back from melting cards. I actually had the highest PCIE draw from graphics test #2 in Fire Strike, with peaks above 120W. That's when I decided I would not be trying to wring out any extra MHz by increasing voltage. The voltage I used, which I copied from the ASUS GTX 950 firmware is actually slightly lower than the stock Quadro voltage. I hope you all found this little experiment as fun as I did.

 

[eidairaman1]

I did not include the Furmark data, but it was more interesting. I think the Nvidia driver does something to hold that back from melting cards. I actually had the highest PCIE draw from graphics test #2 in Fire Strike, with peaks above 120W. That's when I decided I would not be trying to wring out any extra MHz by increasing voltage. The voltage I used, which I copied from the ASUS GTX 950 firmware is actually slightly lower than the stock Quadro voltage. I hope you all found this little experiment as fun as I did.

Cards since the Launch of the ATi Radeon HD 5000 series and nvidia fermi 2.0 (nv was late to the game-gtx 580) have hardware protection to throttle furmark/kombustor to prevent damage to the gpus.

ATI Cypress (Radeon HD 5870) Cards Have Hardware Protection Against Power Virus Like FurMark and OCCT | Geeks3D

The brand new Radeon HD 5870 also called Cypress has a hardware protection against overcurrent generated by FurMark or OCCT GPU. The power supply part or VRM (Voltage Regulation Module) of Radeon HD 4850 / 4870 is under-sized and thus are not able to support the current needed by the GPU when...

www.geeks3d.com

NVIDIA GeForce GTX 580: The Anti-FurMark DX11 Card | Geeks3D

NVIDIA GeForce GTX 580: The Anti-FurMark DX11 Card

www.geeks3d.com

ATI Cypress (Radeon HD 5870) Cards Have Hardware Protection Against Power Virus Like FurMark and OCCT | Geeks3D

The brand new Radeon HD 5870 also called Cypress has a hardware protection against overcurrent generated by FurMark or OCCT GPU. The power supply part or VRM (Voltage Regulation Module) of Radeon HD 4850 / 4870 is under-sized and thus are not able to support the current needed by the GPU when...

www.geeks3d.com

 

[damric]

This is how the project looks now that it's under water. AMD Wraith Stealth fan for a little air flow on front, and a P120 blowing on the back. Lots of 100-120W power spikes according to GPU-Z but the slot is cool and so is the PCB so I don't think they will weld together or catch fire anytime soon. I may boost my R5 1600AF clock a bit since it seems to be bottlenecking the card at 95% utilization due to single thread, It's at stock but will do 4.2GHz without much effort.

 

[eidairaman1]

Good deal, the steel legend you have effortlessly supports a Ryzen 7 5800

 

[damric]

Good deal, the steel legend you have effortlessly supports a Ryzen 7 5800

That test bench is like 3rd in line for hand-me-downs

The totem pole is like
1) Wife's 4k gaming pc: has 5800x and waiting for Zen3D.
2) My main bench: has 2700x and waiting for wife's 5800x.
3) Folding bench #1: has 1600AF waiting for 2700x.
4) Folding bench #2: has 3200G waiting for 1600AF.

If Zen3D is a wash or unobtainable, then I'm considering 5700G for my main bench since it's all PCIE 3.0 anyway.

Edit:

I only had to raise CPU core frequency to 4GHz and OC the RAM/IF to get 100% utilization on the M2000 while folding. 5% more usage means a lot when folding 24/7.

 

[damric]

I did it again. This time with a Quadro M4000. This card was even more interesting than the M2000. The M4000 is a similar GPU as the GTX 970, but with 8GB VRAM and some seriously low clocks, 773MHz with not even a boost feature. The PCB is very sparse, but better built than the M2000. The cooler is also a little bit better, with a couple small heat pipes. The card also has a 6-pin connector. After testing it at stock with 3D and computational workloads, I don't even see why they built a better PCB, cooler, and added a power connector because the thing never made it over 70W and 50C. The stock voltage was very low, set at like 0.95v.

I went about cracking the overclocking locks the same as I did on the M2000, with the Maxwell II editor and NVFLASH with Bypass.

1500MHz core was easy, but I did have to tweak the voltages a bit in the editor. There's probably more left in the tank, as I didn't even try for faster.

My main concern was if I was going to run this card faster than GTX 970, wouldn't I exceed the power capacity of the slot and 6-pin. But nope, the highest I saw was 120ish watts during some 3Dmarks, and about 110w steady while folding. Ok so I figured why not see if I could boost the VRAM from 1500/MHz. First try 1750MHz, no problem. Doing that added about 10-15 more watts, so I called it a day at 1500/1775. I stress tested the overclocks with the stock cooler at 100% fan speed and it peaked about 60C. Today I put it under water because it's going to now be folding 24/7.

Tomorrow I'll post some pictures and benches I ran. Pretty fun time and only cost me about $140 on ebay.

 

[eidairaman1]

I did it again. This time with a Quadro M4000. This card was even more interesting than the M2000. The M4000 is a similar GPU as the GTX 970, but with 8GB VRAM and some seriously low clocks, 773MHz with not even a boost feature. The PCB is very sparse, but better built than the M2000. The cooler is also a little bit better, with a couple small heat pipes. The card also has a 6-pin connector. After testing it at stock with 3D and computational workloads, I don't even see why they built a better PCB, cooler, and added a power connector because the thing never made it over 70W and 50C. The stock voltage was very low, set at like 0.95v.

I went about cracking the overclocking locks the same as I did on the M2000, with the Maxwell II editor and NVFLASH with Bypass.

1500MHz core was easy, but I did have to tweak the voltages a bit in the editor. There's probably more left in the tank, as I didn't even try for faster.

My main concern was if I was going to run this card faster than GTX 970, wouldn't I exceed the power capacity of the slot and 6-pin. But nope, the highest I saw was 120ish watts during some 3Dmarks, and about 110w steady while folding. Ok so I figured why not see if I could boost the VRAM from 1500/MHz. First try 1750MHz, no problem. Doing that added about 10-15 more watts, so I called it a day at 1500/1775. I stress tested the overclocks with the stock cooler at 100% fan speed and it peaked about 60C. Today I put it under water because it's going to now be folding 24/7.

Tomorrow I'll post some pictures and benches I ran. Pretty fun time and only cost me about $140 on ebay.

Lmao that board will outlive the og 970 with 3.5G ram.

 

[damric]

Lmao that board will outlive the og 970 with 3.5G ram.

The question then, is would the 8GB Quadro variant have 7GB of normal VRAM speed and 1GB of the slower VRAM?

 

[qubit]

If you want the holy grail of PCIe specs, including power draw tolerances, then look no further than the official website where these things are defined:

Welcome to PCI-SIG | PCI-SIG

pcisig.com

 

[eidairaman1]

The question then, is would the 8GB Quadro variant have 7GB of normal VRAM speed and 1GB of the slower VRAM?

That's a good point. I'd look at each ram chip to see if they are exactly the same.

 

[damric]

Thanks for the link qubit. It's kind of a moot point now for exceeding the 75W PCI Slot power limit. The M2000 been folding that way 24/7 for 6 months now. I actually bumped the voltage, clock speed, and VRAM a bit higher so now that card is folding at 1557/1753.

It's too late to look at the ICs on the M4000 as I already put some heat sinks on there but they were all Hynix.

Here's a few pics I took of the M4000 before and after. Don't you just love how the factory threw on that 6-pin connector? It works though. Easily picked up the gold hardware cups. M4000 is now added to the 24/7 folding farm.

 

[qubit]

Thanks for the link qubit. It's kind of a moot point now for exceeding the 75W PCI Slot power limit.

Sure, no problem. My intention is just to give people the official reference for when trying to understand the baseline specs of power delivery on the PCIe slot. It's amazing that you can pull so much power from it 24/7 without a problem. Cheaper mobos would likely burn up.