RTX 2060 Folding at GTX 1070 Ti Speeds

[Mstenholm]

... I assume that you run with a faster CPU than I do since 1M PPD @2025 core MHz is that best I have seen so far on the 2060 and 790k on the 1070ti @ 1966 MHz core. More data in the sheet. The 2060 is clearly a better folding card running colder and using less energy.

I don’t know about having a faster CPU for the 2060. I’m currently driving it on a GigaByte Aorus z370 Gaming 5 at PCIe3 x8 with a Pentium Gold 5500 and a single stick of DDR4-2400 along with a 2070 in another x8 slot. I’ve run the 2070 with similar results in a 10 year old Acer mini ITX board with 3GB of DDR3 and a Athlon II x2 220 at 2.8GHz in a PCIe2 x16 so I think the case thermals are more of a limiting factor than a CPU as long as a single core/thread is reserved for feeding the GPU.

There’s another thread in this sub form where the consensus seems to be that a CPU won’t limit a mid to high-tier card too much as long as it’s at least 2GHz or so.

The 1070Ti is in another rig with a GigaByte z370 SLI with a Pentium Gold 5400 and the other stick of DDR4-2400 along with a 1060 6GB.

I’ve now dropped the power limit on the cards and am measuring their efficiency. The results so far over a 3 day average are:

2060 1.04MPPD 140W 1.93GHz 69C
2070 1.29MPPD 160W 1.94GHz 63C
1070Ti 885kPPD 145W 1.96GHz 64C
1060 6G 447kPPD 110W 1.96GHz 73C

And these are all Windows numbers?

[MeeLee]

... I assume that you run with a faster CPU than I do since 1M PPD @2025 core MHz is that best I have seen so far on the 2060 and 790k on the 1070ti @ 1966 MHz core. More data in the sheet. The 2060 is clearly a better folding card running colder and using less energy.

I don’t know about having a faster CPU for the 2060. I’m currently driving it on a GigaByte Aorus z370 Gaming 5 at PCIe3 x8 with a Pentium Gold 5500 and a single stick of DDR4-2400 along with a 2070 in another x8 slot. I’ve run the 2070 with similar results in a 10 year old Acer mini ITX board with 3GB of DDR3 and a Athlon II x2 220 at 2.8GHz in a PCIe2 x16 so I think the case thermals are more of a limiting factor than a CPU as long as a single core/thread is reserved for feeding the GPU.

There’s another thread in this sub form where the consensus seems to be that a CPU won’t limit a mid to high-tier card too much as long as it’s at least 2GHz or so.

The 1070Ti is in another rig with a GigaByte z370 SLI with a Pentium Gold 5400 and the other stick of DDR4-2400 along with a 1060 6GB.

I’ve now dropped the power limit on the cards and am measuring their efficiency. The results so far over a 3 day average are:

2060 1.04MPPD 140W 1.93GHz 69C
2070 1.29MPPD 160W 1.94GHz 63C
1070Ti 885kPPD 145W 1.96GHz 64C
1060 6G 447kPPD 110W 1.96GHz 73C

And these are all Windows numbers?

My 2060 runs at 1.040M PPD as well, fully stock, factory OC on Linux, and can reach as high as 1.35 M PPD with additional overclocking.
Only mine uses 170W in the process, 185W when overclocking.
If I limit it to 140W, I'll be closer to 1,004M PPD.
At 130-140W my 2060 hits 1M PPD.

I bought the higher end 2060 ($375), vs the cheaper $350 one; which came with a dual fan design, and higher factory overclock.
Windows numbers are usually about 5-10% slower.
The cheaper cards usually come with slightly lower factory overclocks; but you should hit about 1M PPD on a 2060 on windows.

[gordonbb]

...
2060 1.04MPPD 140W 1.93GHz 69C
2070 1.29MPPD 160W 1.94GHz 63C
1070Ti 885kPPD 145W 1.96GHz 64C
1060 6G 447kPPD 110W 1.96GHz 73C

And these are all Windows numbers?

Sorry didn’t list the OS. Ubuntu 18.04.1 LTS Server with the ppa 415.27 Driver.

[gordonbb]

My 2060 runs at 1.040M PPD as well, fully stock, factory OC on Linux, and can reach as high as 1.35 M PPD with additional overclocking.
Only mine uses 170W in the process, 185W when overclocking.
If I limit it to 140W, I'll be closer to 1,004M PPD.
At 130-140W my 2060 hits 1M PPD.

I bought the higher end 2060 ($375), vs the cheaper $350 one; which came with a dual fan design, and higher factory overclock.
Windows numbers are usually about 5-10% slower.
The cheaper cards usually come with slightly lower factory overclocks; but you should hit about 1M PPD on a 2060 on windows.

I’m running an EVGA RTX 2060 XC Ultra Gaming which is a 2-slot dual fan design with a higher max TDP (217W) though I’ve not been able to push much past 160W but haven’t been overly aggressive with the GPU overclock yet as even the current 415 driver does not permit changing the overclock from nvidia-settings via the command line so the GUI must be used.

I see 418 was promoted from Beta to Mainline on Friday with the release of the 1660 so I might give that a try in a week or so.

[Theodore]

My 2060 runs at 1.040M PPD as well, fully stock, factory OC on Linux, and can reach as high as 1.35 M PPD with additional overclocking.
Only mine uses 170W in the process, 185W when overclocking.
If I limit it to 140W, I'll be closer to 1,004M PPD.
At 130-140W my 2060 hits 1M PPD.

I bought the higher end 2060 ($375), vs the cheaper $350 one; which came with a dual fan design, and higher factory overclock.
Windows numbers are usually about 5-10% slower.
The cheaper cards usually come with slightly lower factory overclocks; but you should hit about 1M PPD on a 2060 on windows.

I’m running an EVGA RTX 2060 XC Ultra Gaming which is a 2-slot dual fan design with a higher max TDP (217W) though I’ve not been able to push much past 160W but haven’t been overly aggressive with the GPU overclock yet as even the current 415 driver does not permit changing the overclock from nvidia-settings via the command line so the GUI must be used.

I see 418 was promoted from Beta to Mainline on Friday with the release of the 1660 so I might give that a try in a week or so.

I'm running the same card as you!
I've noticed the same on mine. I'm running 415.27, and I couldn't overclock on mine either.
I presumed it had to do with my limiting the power settings via NVidia-smi.
I am running it on a 16x PCIE slot, so that might make the difference!
From what I've noted, there's about a 5% performance difference between the 16x slot running at 8x, and the 4x slot.
There's a possibility you could run higher PPD and draw higher watts if you replaced the card to an actual 16x slot (no riser)?

Also, my card is installed in a rack, where it runs at 76C, 70% fan. I believe heat also can contribute to slower performance on these cards.

Let us know how the 4.18 driver works out!
I don't want to mess with my current system too much, now that it's working fine!

[gordonbb]

I'm running the same card as you!
I've noticed the same on mine. I'm running 415.27, and I couldn't overclock on mine either.
I presumed it had to do with my limiting the power settings via NVidia-smi.
I am running it on a 16x PCIE slot, so that might make the difference!
From what I've noted, there's about a 5% performance difference between the 16x slot running at 8x, and the 4x slot.
There's a possibility you could run higher PPD and draw higher watts if you replaced the card to an actual 16x slot (no riser)?

Also, my card is installed in a rack, where it runs at 76C, 70% fan. I believe heat also can contribute to slower performance on these cards.

Let us know how the 4.18 driver works out!
I don't want to mess with my current system too much, now that it's working fine!

The inability to overclock from the command line appears to be due to driver issues. I had the same issue with my 2070 on the 410 driver and also issues with Fan Control. Fan Control is now working on the 2070 with the 415.27 driver from the command line but Overclocking is still only possible from X Windows using the nvidia-settings GUI with coolbits 4 or 12 for both models.

I’m running my cards in PCIe3 x16 slots running at x8 on z370 motherboards with PCIe switches (PLX) to break the 16 PCIe3 lanes from the CPU into 8 for each slot when the second slot is populated (x16/0 -> x8/x8). If you’re trying a x4 slot it might be from the PCH on Intel or Southbridge on AMD in which case the performance degradation is likely due to contention for the limited (DMI on Intel, PCIe3 x4 on AMD) interconnect between the PCI/Southbridge and CPU.

Boost 3 will start to limit at 60C and more aggressively at higher GPU temps. 76C is a little warm but you’d likely only get one or two more bins (13MHz atomic steps on Pascal, 15MHz on Turing) dropping to the low 60s. 70% fan is a little high. I prefer to keep them under 60 as anecdotal evidence from miners is that this leads to longer life. EVGA, however, does make solid cards and they do have a three year warranty which is likely close to the useful lifetime of the card (4 years, or two generations?). Sill, having to RMA a card for a fan replacement would be a pain.

[Theodore]

Gordon, does NVidia server allow you to adjust each fan individually on the 2060?
It's a first for me!