RTX cores for folding@home

[foldy]

I'm surprised! Maybe these optimized Turing architecture is good for compute tasks.
First test on a RTX 2080 shows 1.4M PPD viewtopic.php?f=83&t=31051

[kiore]

These initial results seem well beyond expectations, am interested to see further results with more details.

[foldy]

Would be interesting if on Windows there are stronger PCIe speed limits. How much PPD would be lost on PCIe gen3 x8 or x4 on Windows?

[fevaoctwh]

Would be interesting if on Windows there are stronger PCIe speed limits. How much PPD would be lost on PCIe gen3 x8 or x4 on Windows?

For your reference, gaming test on PCIE scaling

https://www.techpowerup.com/reviews/NVI ... ing/6.html

[foldy]

This is for games only. FAH is more demanding on PCIe on Windows so this could be easily 50% performance loss on pcie x4.

[bruce]

This is for games only. FAH is more demanding on PCIe on Windows so this could be easily 50% performance loss on pcie x4.

It's important to note that there they do present three different charts ... for 1920x1080, for 2560x1440, and for 3840x2160. Let's not forget that FAH DOES NOT produce video video output for any of those resolutions. We'd have to ask what would a chart show for 0000x0000 and that data cannot be derived from any information presented there.

There is no doubt that video performance is important if you are a gamer but FAH does not use any of the features that are being evaluated there. FAH will perform exactly the same if no monitor is connected to the GPU.

[rwh202]

Finally got around to installing the RTX 2080 FE but forgot to put the power meter in the loop.
Stock clocks and boosts to 1905 MHz at 72 degC and 50% fan. Running Linux mint 17.3 on PCIe gen 2 x16:

P11728 47 second TPF = 1,596,000 PPD

Edited to add: 260 W from the wall on a pentium itx mb with gold psu.

[fevaoctwh]

This is for games only. FAH is more demanding on PCIe on Windows so this could be easily 50% performance loss on pcie x4.

It's important to note that there they do present three different charts ... for 1920x1080, for 2560x1440, and for 3840x2160. Let's not forget that FAH DOES NOT produce video video output for any of those resolutions. We'd have to ask what would a chart show for 0000x0000 and that data cannot be derived from any information presented there.

There is no doubt that video performance is important if you are a gamer but FAH does not use any of the features that are being evaluated there. FAH will perform exactly the same if no monitor is connected to the GPU.

Open GPUZ and we will have the answer... on the sensor page, there's an item called bus interface load. Should be the PCI-E bus workload. And my case with 2080 is running on PCI-E 3.0 16X, and avg. load is 41%. Meaning PCI-E 3.0 8X or PCI-E 2.0 16X should be sufficient for F@H. plus/minus 2% I will just call it error...

[fevaoctwh]

Finally got around to installing the RTX 2080 FE but forgot to put the power meter in the loop.
Stock clocks and boosts to 1905 MHz at 72 degC and 50% fan. Running Linux mint 17.3 on PCIe gen 2 x16:

P11728 47 second TPF = 1,596,000 PPD

Terrific!

[foldy]

So we have RTX 2080 with 1.4m PPD on Windows and 1.6m PPD on Linux.

[Biffa]

Just switched out a GTX 1080Ti with a RTX 2080Ti on Windows 10

Performance has gone from circa 1.1mil PPD to 1.87mil PPD (32s TPF)
Power use (for whole machine) has gone from 185W to 251W

Update:
Project 11728
1080ti Frame Time 00:56, PPD 1,236,447
2080ti Frame Time 00:39, PPD 2,127,456

Project 11726
1080ti Frame Time 01:13 PPD 1,228,291
2080ti Frame Time 00:50 PPD 2,166,859

Project 11719
1080ti Frame Time 01:19 PPD 1,225,067
2080ti Frame Time 00:53 PPD 2,229,394

[toTOW]

This means nothing if you don't tell us on which project you got those numbers ...

[Biffa]

This means nothing if you don't tell us on which project you got those numbers ...

Updated

[scott@bjorn3d]

Just switched out a GTX 1080Ti with a RTX 2080Ti on Windows 10

Performance has gone from circa 1.1mil PPD to 1.87mil PPD (32s TPF)
Power use (for whole machine) has gone from 185W to 251W

Update:
Project 11728
1080ti Frame Time 00:56, PPD 1,236,447
2080ti Frame Time 00:39, PPD 2,127,456

Project 11726
1080ti Frame Time 01:13 PPD 1,228,291
2080ti Frame Time 00:50 PPD 2,166,859

Project 11719
1080ti Frame Time 01:19 PPD 1,225,067
2080ti Frame Time 00:53 PPD 2,229,394

That is impressive. Can't wait for 2080TI FTW3 to show up.

[bruce]

The RTX 2080Ti reportedly gets 11750 (13448) GFLOPS with 4352 SP shader processors @ 1350 (1545) GHz.
The GTX 1080Ti reportedly gets 10609 (11340) GFLOPS with 3584 SP shader processors @ 1480 (1582) GHz.
so I'd expect the RTX to get 110% (118%) of the performance of the GTX based JUST on either the SP or DP Shader performance -- totally ignoring the Tensor cores, the speedup of Memory, the speed of the bus, etc.

Because of the way bonus points work, the change to PPD is always greater than the inverse of the change of the TPF so that's not unreasonable. (PPD always gives you an incentive to run faster hardware.)

You're saying that the power use went from 185W to 251W or to 136%. I guess that's also reasonable since all the new features that FAH isn't using do require some power..