Graphics Cards

[ccbadd]

I have a couple W6800s and they are supposed to have ECC memory but I have no idea how to go about enabling/disabling it.  Any ideas where to look?

Thanks

[terminusaquo]

AFAIK you can't disable ECC on the GPU as this is part of the RAM itself. Is there a reason why you need ECC memory on your GPU? Why would you need to disable it?

[ccbadd]

I just want to view the settings.  I have 2 W6800s and one shows 32GB VRAM and the other shows 30GB VRAM.  With NVidia cards that support it, they don't have true ECC and thus use some of the VRAM for parity so I guessed that might be what AMD has done.  I have tried multiple avenues to get support from AMD to answer this question but never get anything precise back.  Just maybes and passible type answers.  Really get sick of them and no real support.

[fsadough]

ECC memory can be enabled in Radeon Pro Settings under Graphics / ECC Options. To your 2nd inquiry as to why one GPU shows 30GB and the other one 32GB, I would need more detailed info from your system.   

Please provide an AMDZ Report and upload the results via "wetransfer"

AMDZ Report
- Please extract the amdz-v334.zip available from https://go.wetransfer.com/t-XjitlZnwFh
- Run amdz.exe file as an Administrator
- Select “Save All“ and “TXT“ as the output format
- Click on the blue button to save the report
- The .txt file will be saved in the same folder where you extracted the zipped file

[ccbadd]

This is running on Ubuntu Linux not Windows so I don't have access to that setting or the ability to run the reporting tool.  I did find a post elsewhere where they stated that enabling ECC support did indeed reduce the available memory to the amount I have.  I guess I'll have to pull the card and change the setting then replace it.  I do believe this should be an option and assessable via rocm-smi on linux but it's just another AMD frustration.  It's getting pretty hard to support AMD on the GPU side of things.

[fsadough]

ECC should be enabled automatically in Linux, unless there is an error message in the dmesg. See below mine:

[ 3.276427] amdgpu 0000:0a:00.0: vgaarb: deactivate vga console
[ 3.276429] amdgpu 0000:0a:00.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported
[ 3.276456] amdgpu 0000:0a:00.0: amdgpu: MEM ECC is active.

[farshadg]

I know this is an old thread, but I recently was looking into finding a way to disable ECC support on my AMD W7900 as I would like to have that additional VRAM (full 48GBs instead of 45GBs) for loading AI models. On Windows this is really easy, but on Linux this tutorial may help you to patch the Linux kernel to disable/enbable ECC support. Disclaimer I haven't tried it out yet myself: https://lunnova.dev/articles/amdgpu-disabling-ecc/