F@H core22 0.0.13 in BETA testing : CUDA support at last!

[biodoc]

You'll need to log into the folding forum to read the thread.

The the current core 22 GPU app is openCL. The new core 22 version 0.0.13 which is in beta testing is a Cuda app and is significantly faster especially on Linux.

Some snippets from the thread at the folding forum:

p13426 at 10% in Windows 10 on GTX 1080 Ti running OpenCL: TPF of 1:57
p13426 at 10% in Windows 10 on GTX 1080 Ti running CUDA: TPF of 1:16
p13426 at 10% in Debian on GTX 1080 Ti running OpenCL: TPF of 1:34
p13426 at 10% in Debian on GTX 1080 Ti running CUDA: TPF of 0:59

So if Windows-OpenCL is our baseline, then on this project, the speed improvements are:
Windows-CUDA: +53%
Debian-OpenCL: +24%
Debian-CUDA: +98%
CUDA average: +75%

The core will attempt to use CUDA when available. This translates into a very significant gain in PPD, as Adam Beberg found for another project across different Nvidia hardware in Linux:

TPF 73s - GTX 1080Ti running OpenCL/ 1.554 M PPD
TPF 57s - GTX 1080Ti running CUDA / 2.253 M PPD
TPF 49s - RTX 2080Ti running OpenCL/ 2.826 M PPD
TPF 39s - RTX 2080Ti running CUDA / 3.981 M PPD
TPF 36s - RTX 3080 running OpenCL / 4.489 M PPD
TPF 31s - RTX 3080 running CUDA / 5.618 M PPD

 

[Pokey]

You'll need to log into the folding forum to read the thread.

The the current core 22 GPU app is openCL. The new core 22 version 0.0.13 which is in beta testing is a Cuda app and is significantly faster especially on Linux.

Thanks @biodoc. I will give it a try and see what happens.

 

[Markfw]

Will it auto-use the new version , since 22 was the old core version number also ?

 

[biodoc]

Will it auto-use the new version , since 22 was the old core version number also ?

I would assume so. Core 22 has been released in different versions. The current workhorse (OpenCL) version is 0.0.12. The beta version that supports cuda is version 0.0.13.

 

[Markfw]

I would assume so. Core 22 has been released in different versions. The current workhorse (OpenCL) version is 0.0.12. The beta version that supports cuda is version 0.0.13.

But it looks like you do have to have "client-type" "beta"

 

[Pokey]

Yep. client-type beta

I have core22 ver .13

"21:27:21:WU02:FS00:0x22: Using CUDA and gpu 0"

The workunits are 134XX series. Hard to get a sense of speed just yet because they are different.


EDIT: a couple hours later my gross PPD via FAHControl jumped from 23k to 33k. Will check again tomorrow. cuda is definitely in play.

EDIT 2: On my mix of rigs it looks like a 40-45% increase cuda over open cl, all on linux.

 

[Bubbleawsome]

Seems like it's reached wide-release now

 

[Nephross]

Seems like it's reached wide-release now

It would seem so: https://foldingathome.org/2020/09/28/foldingathome-gets-cuda-support/

I have caught project 13426 on both my GPU's with CUDA enabled. It doubled PPD compared to "normal" OpenCL operations.

One thing I ran into though. I had to update the nvidia driver from 435 to 450 on my linux box, as the core would crash after a checkpoint. It's probably just me being slow to do the upgrade, but figured I would leave it here if anyone else ran into something similar.

 

[Endgame124]

It would seem so: https://foldingathome.org/2020/09/28/foldingathome-gets-cuda-support/

I have caught project 13426 on both my GPU's with CUDA enabled. It doubled PPD compared to "normal" OpenCL operations.

One thing I ran into though. I had to update the nvidia driver from 435 to 450 on my linux box, as the core would crash after a checkpoint. It's probably just me being slow to do the upgrade, but figured I would leave it here if anyone else ran into something similar.

Did you specifically enable CUDA, or did it all happen automatically after the driver update?

 

[JeffCos]

Mine happened automatically. I was wondering why my ppd was so high.

 

[Nephross]

Did you specifically enable CUDA, or did it all happen automatically after the driver update?

It happened automatically on both machines. The machine with older drivers too. In the log, it seems the new core does a check to see if the gpu supports cuda, and then proceeds to use the cuda acceleration if available.

 

[Endgame124]

My output is up around 60% without having to make any changes. The 1660 super increased a lot more than the 1080ti - not sure why, but I’ll take it.

 

[Pokey]

My output is up around 60% without having to make any changes. The 1660 super increased a lot more than the 1080ti - not sure why, but I’ll take it.

Windows or Linux?

 

[Endgame124]

Windows or Linux?

Linux. Fedora 30 if it matters

 

[Pokey]

Just curious. I don't think the Linux distro matters, but Linux over Windows is substantial, or at least it used to be.

 

[Endgame124]

It might just be that I hit the right work Units with the 1660 super, as this page shows some work units being over 50% gain:

Together we are Even More Powerful: GPU folding gets a powerup with NVIDIA CUDA support! – Folding@home

foldingathome.org

I’m really curious what is so different between the 1660 ti and the 1660 super that the ti sees a 200% increase while the super sees substantially less.