Been a lurker for years, finally had something that requires me to make an account and contribute to the conversation.
I have a Dell 7920 tower I use as a home server. Run Proxmox on it, have passed through both Nvidia and AMD GPUs, oem and aftermarket. Currently it’s just a Tesla P4 and a Quattro 2000. So far it has been a wonderful machine.
Picked up an AMD Instinct Mi25 to try and have some fun with LLMs and the such. When I install it on the 7920, it will not boot to an OS, but instead will pull up the BIOS/system settings. Looking at system info in the BIOS, it recognizes that there is a GPU in that PCIe slot, but will not boot to an OS. Power gets to the card, verified by little red light on the card and it gets warm. Plugging it into my AM5 system, and running lspci shows an Mi25. The 7920 has support for 4 dual slot GPUs, so I try it in another slot. I get the same results.
I am now at a loss at to what wall I am throwing myself against and how to tackle it. My next step is to try it in my old Dell 5810 to see if I am dealing with a Dell proprietary thing. Maybe my first gen Xeon scalables can’t handle the load? Maybe it’s a power draw issue with two 8 pin connectors in the same card? Will continue my way forward, but any ideas at all would be greatly appreciated.
1 Like
What CPU’s are those systems running and do they have enough PCIe lanes to run that many GPU’s? Also make sure the above 4G is enabled in the bios before installing that GPU. More info about the system in use will get you more help.
Edit: You can find a lot of info and help on the setup of that video card here.
Currently I only have two silver 4114 in there, should have mentioned that in my main post. The only thing that gives me pause about meeting my pcie lane limit is even with just the Mi25 in the machine, I run into the same issue.
I do have the bios option enabled for 4gb or more on PCIe lanes.
Make sure CSM is disabled in MB bios.
Just double checked to make sure about CSM. It is disabled.
bit of interesting info, when I plugged it into my smaller, older dell running a single e5 v4 xeon, it boots and works fine. it over heats in this case, but I can fix that with some 3d printing.
Hope I’m not too late. I had the same problem with an R7910 (2x Xeon 2687W v4)and solved it by using the onboard graphics chip for video out instead of the PCIE. I think it has to with the bios looking for a VGA adapter and these cards(MI25) are accelerators. Hope its a solve for you as well, cheers!
@sw3333t The precision 7920 doesn’t have any video out unfortunately…
@snekerpmp did you ever figure out the cause of the issue? I’ve been trying to get this thing to work for a few days now on my 7820.
The cards have speakers built in, it will scream at you if it gets too hot, so cut the power asap if that happens
Even if you have a over 4g decoding you still have limited bar space, try removing the Tesla p40 and just use the quadros and MI25
Alternatively you said it works in your other xeon machine
Use that PC and use the guides on the forum to flash the MI25 to either a wx9100 or Vega64 FE which have a standard 256MB bar space instead of the full fat 16GB bar space, you still get all your VRAM it just doesn’t try and squeeze it through the bus at once like the server cards like to do
Drivers for those are much easier anyways and you get working mini dp with the wx9100 bios