Radeon VII (2nd gen consumer Vega GPU)

[ixfd64]

I suspect some systems actually provide more power to one GPU than the others. I'm currently running mfakto on a Mac Pro with two AMD FirePro D700 GPUs, and one consistently produces more GHz-days despite also having to power the display.

[hansl]

Originally Posted by Prime95

One mystery I still do not have an answer for. Why does one GPU report more voltage but lower power draw?

If the one drawing more power is also the one at higher temps, then its probably those higher temps leading to increased leakage current on the transistors. GPU power efficiency gets worse with temperature.

[Prime95]

To better control temperatures, I've thinking of replacing the cases that house the Radeons with an open air mining frame. In the current case, one radeon is about 1 centimeter above the power supply and the other is about 1.5 centimeters above that. Even with 6 fans in the case, I suspect airflow is not good. Any comments on if the parts below would (a) work and (b) improve temps? Even better ideas are most welcome.

I'm looking at this:
https://www.amazon.com/Kingwin-Stack...NsaWNrPXRydWU=

or this:
https://www.amazon.com/Veddha-6-GPU-...ustomerReviews

and this:
https://www.amazon.com/dp/B076RSLZWT...xpY2s9dHJ1ZQ==

[paulunderwood]

Originally Posted by Prime95

To better control temperatures, I've thinking of replacing the cases that house the Radeons with an open air mining frame. In the current case, one radeon is about 1 centimeter above the power supply and the other is about 1.5 centimeters above that. Even with 6 fans in the case, I suspect airflow is not good. Any comments on if the parts below would (a) work and (b) improve temps? Even better ideas are most welcome.

I'm looking at this:
https://www.amazon.com/Kingwin-Stack...NsaWNrPXRydWU=

or this:
https://www.amazon.com/Veddha-6-GPU-...ustomerReviews

and this:
https://www.amazon.com/dp/B076RSLZWT...xpY2s9dHJ1ZQ==

Do you need a special motherboard to handle all those GPUs? Ditto PSU?

[Prime95]

I am running 2 GPUs in 2 different Linux boxes. I'm simply looking at moving all the components from a traditional tower case to an open air setup. Since these frames support 6 GPUs, there will be plenty of space between the 2 GPUs I'd be running.

[Mark Rose]

In my experience, air flow matters tremendously. I used to run GPUs in caseless systems. I ended up putting complete systems in mesh front cases and that worked out very well. My experience with non-mesh cases hasn't been as great.

It might be cheaper to just buy new cases if you're not going to start running many GPUs per system. They certainly stack better and are less accident-prone.

[kriesel]

Originally Posted by Prime95

To better control temperatures, I've thinking of replacing the cases that house the Radeons with an open air mining frame. In the current case, one radeon is about 1 centimeter above the power supply and the other is about 1.5 centimeters above that. Even with 6 fans in the case, I suspect airflow is not good. Any comments on if the parts below would (a) work and (b) improve temps? Even better ideas are most welcome.

I'm looking at this:
https://www.amazon.com/Kingwin-Stack...NsaWNrPXRydWU=

or this:
https://www.amazon.com/Veddha-6-GPU-...ustomerReviews

and this:
https://www.amazon.com/dp/B076RSLZWT...xpY2s9dHJ1ZQ==

Something to note is that the 4 frame mounted fans blow air toward the gpus; the gpus shown have fans and shrouding designed to discharge air out the case-back bracket; so for the first link, the fans are blowing at each other, butting heads sort of. Conversely, on the second, they appear to be working together.

One thing neither shows is the rats' nest of cabling for power and signals after it's fully connected.
Also gpu mechanical support can be very limited; in some frame kits just the single screw in the back bracket, not very mechanically sound!

These open frame designs are very far from pet or toddler proof. How secure will these be from any pets or grandchildren or small guests? Fan guards are not included, but probably advisable.

[Prime95]

I ran rocminfo as root today and got this message:

george is not member of "video" group, the default DRM access group. Users must be a member of the "video" group or another DRM access group in order for ROCm applications to run successfully.

Is this a useful clue as to why I can only run gpuowl as root?

BTW, I added "george" to the group "video" and still could not run gpuowl as a normal user. However, I am far from a Linux expert, so a more knowledgeable user might have success.

EDIT: I logged out and back in. That seemed to be necessary. Now I am a member of the video group and can run gpuowl! The "gold standard" instructions in post #76 should be updated.

[retina]

Originally Posted by Prime95

BTW, I added "george" to the group "video" and still could not run gpuowl as a normal user. However, I am far from a Linux expert, so a more knowledgeable user might have success.

Did you logout and login again? I found that groups memberships only update upon login.

[paulunderwood]

Originally Posted by Prime95

I ran rocminfo as root today and got this message:

george is not member of "video" group, the default DRM access group. Users must be a member of the "video" group or another DRM access group in order for ROCm applications to run successfully.

Is this a useful clue as to why I can only run gpuowl as root?

BTW, I added "george" to the group "video" and still could not run gpuowl as a normal user. However, I am far from a Linux expert, so a more knowledgeable user might have success.

EDIT: I logged out and back in. That seemed to be necessary. Now I am a member of the video group and can run gpuowl!

Thanks for sharing this. I will do the same for my system.

Edit: I am a member of the video group, but I have to prefix with sudo still.

[Prime95]

Originally Posted by paulunderwood

Edit: I am a member of the video group, but I have to prefix with sudo still.

Rocm 2.9 here -- if that makes a difference.