Folding Forum

please add AMD GPU MI100
0x1002:0x738c:131:0:0:Advanced Micro Devices, Inc. [AMD/ATI]:

Welcome to Folding@Home!

https://www.techpowerup.com/gpu-specs/r ... i100.c3496

First the good news, the two known Prerequisites for folding are OpenCL 1.2 or higher, and 64 bit (double precision) floating point math. Your card meets both!

Now the ugly news, your card is the first of a new architecture by AMD called CDNA, so it may be some time before AMD's ROCm drivers and F@H's Cores agree on how to fold on this card.
(So new that Wikipedia has no article on CDNA)
https://www.amd.com/en/technologies/cdna

There is no real bad news! I am sure someone will add the ID to the GPUs.txt on the server. Once the server knows about it, you can then delete your GPUs.txt and restart, that will force your PC to have the same one the server has in its database. Then your PC will be the guinea pig to find out if it works!

I am also hopeful that when the developers get the right information from AMD, that folding with their RDNA/RDNA2 technologies will be possible....

But, I'm not holding my breath.

Paul

psaam0001 wrote:I am also hopeful that when the developers get the right information from AMD, that folding with their RDNA/RDNA2 technologies will be possible....

But, I'm not holding my breath.

Paul

RDNA/RDNA2 cards are already supported, what gave you the idea they weren't?

psaam0001 wrote:I am also hopeful that when the developers get the right information from AMD, that folding with their RDNA/RDNA2 technologies will be possible.l

There WAS a rocky start with RDNA, some Cores supported it and others did not. I believe it is fully supported now.

Here is a post in June, if Core_21 ever comes back, It could be a RDNA issue still. But Core_22 should support RDNA with correct drivers.
viewtopic.php?f=16&t=35477&p=336438&hilit=rdna#p336438

I have no idea if CDNA is supported, which would be on topic here.

Added the following entry to GPUs.txt file :
0x738c / Arcturus GL-XL [AMD Instinct MI100]

Let us know how it runs ...

toTOW wrote:Added the following entry to GPUs.txt file :
0x738c / Arcturus GL-XL [AMD Instinct MI100]

Now that the wonderful toTOW has added it to the server's database, you can force you PC to update your copy by deleting the GPUs.txt file and restarting.

C:\Users\JimboPalmer\AppData\Roaming\FAHClient\GPUs.txt where JimboPalmer is really your login name on the PC. (that is just where it is on my PC)

Alternatively, if you're using V7.6.20 or higher, then it would be here:
C:\ProgramData\FAHClient

Alright, waiting for fireworks

Joe_H wrote:

psaam0001 wrote:I am also hopeful that when the developers get the right information from AMD, that folding with their RDNA/RDNA2 technologies will be possible....

But, I'm not holding my breath.

Paul

RDNA/RDNA2 cards are already supported, what gave you the idea they weren't?

I know the cards themselves are supported... I was just curious as to whether or not RDNA/RDNA2 was anything like NVidia's CUDA technology.

Paul

RDNA / RDNA2 are a hardware microarchitecture, not a GPU programming language/API. In the case of nVidia they support the CUDA and OpenCL instruction sets. For AMD their current and recent cards support OpenCL, Vulkan, and the latest cards support another programming API called HIP.

The holdup with RDNA was that some architecture specific changes needed to be added to support using OpenCL on cards based on this new microarchitecture. Will have to see if the same holds for CDNA.

F@h uses OpenCL on both nVidia and AMD, and recently adde CUDA capability to Core_22 GPU processing. OpenCL is also used for GPU computing on Intel iGPUs

TY for that info... I'll just stick to using what I know is going to work.

Paul

Good news the card is working but the performance of the 8 cards is not good (PPD 15096554). Is at the level of a system with 2 Nvidia V100

Frank

fswork wrote:Good news the card is working

Code: Select all

16:18:29:******************************* System ********************************
16:18:29:            CPU: AMD EPYC 7302 16-Core Processor
16:18:29:         CPU ID: AuthenticAMD Family 23 Model 49 Stepping 0
16:18:29:           CPUs: 64
16:18:29:         Memory: 503.59GiB
16:18:29:    Free Memory: 488.58GiB
16:18:29:        Threads: POSIX_THREADS
16:18:29:     OS Version: 3.10
16:18:29:    Has Battery: false
16:18:29:     On Battery: false
16:18:29:     UTC Offset: 1
16:18:29:            PID: 42383
16:18:29:            CWD: /var/lib/fahclient
16:18:29:             OS: Linux 3.10.0-1160.11.1.el7.x86_64 x86_64
16:18:29:        OS Arch: AMD64
16:18:29:           GPUs: 8
16:18:29:          GPU 0: Bus:99 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 1: Bus:67 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 2: Bus:38 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 3: Bus:3 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 4: Bus:227 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 5: Bus:195 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 6: Bus:163 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:          GPU 7: Bus:131 Slot:0 Func:0 AMD:6 Arcturus GL-XL [AMD Instinct MI100]
16:18:29:           CUDA: Not detected: Failed to open dynamic library 'libcuda.so':
16:18:29:                 libcuda.so: cannot open shared object file: No such file or
16:18:29:                 directory
16:18:29:OpenCL Device 0: Platform:0 Device:0 Bus:99 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 1: Platform:0 Device:1 Bus:67 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 2: Platform:0 Device:2 Bus:38 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 3: Platform:0 Device:3 Bus:3 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 4: Platform:0 Device:4 Bus:-29 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 5: Platform:0 Device:5 Bus:-61 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 6: Platform:0 Device:6 Bus:-93 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:OpenCL Device 7: Platform:0 Device:7 Bus:-125 Slot:0 Compute:2.0 Driver:3212.0
16:18:29:***********************************************************************

but the performance of the 8 cards is not good (PPD 15096554). Is at the level of a system with 2 Nvidia V100

Frank

Could you take a screenshot of frame times or a log, or fahclient?

~2m PPD is about right for GCN type of architecture

P.S. 8 AMD cards working on Linux, at the same time, is a miracle on its own