Folding Forum

Skajaquada wrote:So i just installed my RX6800xt and it´s a real beast of a gpu.
Is there any timetable for getting it into F@H?

viewtopic.php?f=83&t=36457

Take it away remember, if there is no report of TPF and PPD, it did not happen good luck

Haha im just testing a bit, but it works as expected. Almost 3 million PPD and TPF 1 min 08 at stock, a bit better then two RX5700xt.
And the cooling is really great. Silent and powerful with a big reserve for the summer
Tuning is a bit different and i dont want to kill the bios on the first day...
Now i let it crunch some WU and report the stats tomorrow.

Skajaquada wrote:Haha im just testing a bit, but it works as expected. Almost 3 million PPD and TPF 1 min 08 at stock, a bit better then two RX5700xt.
And the cooling is really great. Silent and powerful with a big reserve for the summer
Tuning is a bit different and i dont want to kill the bios on the first day...
Now i let it crunch some WU and report the stats tomorrow.

Is that p13428? If yes, then it is very impressive

Oh sorry i forgot the type of the WU, yes it is p13428.
It comes close to a 2080Ti without CUDA "bonus" at a similar power consumption.
I got a really good chip which runs at almost 2500Mhz with 180W GPU Chip Power.
Undervolting is less effective then overclocking. So it is sad that AMD cut the powerlimit to +15%.
I think a custom model can possibly become much faster.

Awesome stuff, thanks

Skajaquada wrote:Oh sorry i forgot the type of the WU, yes it is p13428.
It comes close to a 2080Ti without CUDA "bonus" at a similar power consumption.
I got a really good chip which runs at almost 2500Mhz with 180W GPU Chip Power.
Undervolting is less effective then overclocking. So it is sad that AMD cut the powerlimit to +15%.
I think a custom model can possibly become much faster.

I've seen 2080Tis without Cuda run +4 M PPD.

I've seen 2080Tis without Cuda run +4 M PPD.

I run three watercooled 2080Ti and they reach 4.3 M PPD on 2070 Mhz and 85% Powertarget, but this is with CUDA.
Maybe you can get 4 Mio without CUDA but Overclocking and much more power consumption.
That is also the point for the RX6800xt.
It could reach the speed of a 2080Ti, with a Powerlimit of 450W. And that is not really efficient in comparison to Nvidia GPUs.

Keep in mind we have 3m ppd on small atom count with relatively undervalued base credit for AMD cards. They should show more once we receive some bigger atom count projects.
Overall, we know CUDA is game changer. Until AMD brings something similar to the table we can only take what it gives and there is no need to compare them to nVidia cards

I have an interesting issue on p17311 with the RX6800xt. It start normal at full speed but after some time it went down to 535Mhz. The GPU hold this for some minutes and then returns to full speed.
It does not stop with a Bad Work Unit, just slow it down to about 2.9 M PPD overall.
This does not happen on p13428, but i dont think this is a problem of p17311. Maybe a Driver Bug if running the GPU over 95% for a longer time.
I need to test this a bit longer and on other projects...

Possible thermal throttling? 17311 has quite a lot of atoms, so it might be pushing GPU to the limit. What CPU is in the system? Maybe that one is slowing GPU down?

Skajaquada wrote:

I've seen 2080Tis without Cuda run +4 M PPD.

I run three watercooled 2080Ti and they reach 4.3 M PPD on 2070 Mhz and 85% Powertarget, but this is with CUDA.
Maybe you can get 4 Mio without CUDA but Overclocking and much more power consumption.
That is also the point for the RX6800xt.
It could reach the speed of a 2080Ti, with a Powerlimit of 450W. And that is not really efficient in comparison to Nvidia GPUs.

No, a year ago, when core 22 just came out, they had a good project for the 2080 Ti.
One of my 2080 Tis would run 4.4M ppd. It was overclocked to 2050-2150Mhz (maximum) and ran at 220W (80% of it's rated 275W).
The GPUs thermal throttled quickly (+83C), even in an open bench, unless more cooling was provided.
Since then I set the power levels between 180W and 200W, and got closer to 65-75C, 3.2M at 180W, 3.6M PPD @ 190-205W, avg frequency between 1785 and 1875W.

You can also try to run 2 work units in parallel on the same GPU to help with small work units not able to use full AMD GPU.

foldy wrote:You can also try to run 2 work units in parallel on the same GPU to help with small work units not able to use full AMD GPU.

That is not supported and should not work
When proposed to nVidia, we got an impression that in order to have 2 or more WUs fold in parallel on same GPU is a lot of experimental programming
We are even afraid to propose that to AMD

muziqaz wrote:Possible thermal throttling? 17311 has quite a lot of atoms, so it might be pushing GPU to the limit. What CPU is in the system? Maybe that one is slowing GPU down?

CPU is a watercooled Ryzen 7 3700X and the RX6800xt runs at 70°C average and 86°C hotspot. So i dont think it will be a temperature problem.
It is very funny that the client is folding on and it does not fail, so i dont think the driver crashes for any reason.
I will try to test it, but the last three WU were all p13428.

Is anything running on CPU folding wise?
Navi tends to act weirdly with certain projects of there is something happening with the CPU. Though during testing 17311 did not show any issues on Navi :/