love it… iam gonna safe some cash up for the b60 though
it will be my sriov multi gaming emulation dream <3 pcie 5 8x awsooooooome it s perfekt for small families and need for compute
I tried using the i915-sriov-dkms package with proxmox 9, no luck
I don’t even see the sr-iov capability on the b50 in lspci (and i’ve run vgpu with an rtx a6000 on this system in the past, so i know sr-iov does work…)
i might try the rebuilt kernel next, we’ll see.
Gonna grab some beers and start working on this tonight. Finally got moved over to proxmox.
unfortunately no change with the rebuilt kernel either. i guess something about this system is incompatible right now (x10drh-iln4, so quite an old platform) ![]()
i915 only supports running first generation Xe silicon or Alchemist cards, you want the newer Xe kernel drivers for Battlemage Xe2 graphics hence the rebuilt kernel which you did. But that may not be the problem you are running into. See this snippet from Wendell’s guide
You may want some way to apply said firmware update to the card as it seems crucial to getting this working.
Ahhhhh totally missed the firmware update part
thank you thank you.
Can confirm SR-IOV is now working with i915-sriov-dkms (despite the name, the latest release also contains the 6.17 xe driver). It is also working with the patched kernel, but personally I’ll stick with dkms.
ayyy nice. does the dkms sriov work with stock proxmox 9? that has not been working as of a few weeks ago
yes stock proxmox 9. the 6.17 xe driver was only merged into it 3 days ago, so this is very recent.
You mean that it does not work if you do standard PCI Passthrough of the full GPU to a Windows VM and try from there? I did had success flashing what today would be 15 years old Radeon HD 5770 from within a Windows VM, and I don’t recall if I did the same (Or was in another system) with a Polaris Radeon RX 570 to use custom voltage/frequency tables.
I assume the Intel way of updating would be quite more picky and require special parameters, blah blah blah. Or perhaps with an alternate tool? Otherwise you can’t get it working from scratch in-situ.
That’s a good idea. I bet a windows VM getting the whole GPU would actually be able to do a firmware update. That might be living somewhat more dangerously than bare metal however . . .
It does. Did that yesterday. Had my fingers crossed, but no issues. Probably a smarter idea to just do it on bare metal.
Dangerous or not, I have been doing this on my eight A series cards monthly for over a year and it’s been totally fine. Sample size of 1, YMMV of course.
The only possible issue that you may have is that if it ask for a reboot after flashing, you actually want to shut down the VM and power cycle the computer because the GPU could hang in an undefined state if you merely reboot the VM only. Otherwise I don’t see it any worse than bare metal, assuming that there is more people than got it working that way.
What link speed are your B50’s reporting? Mine reports x1 in lspci in a Mint VM (6.14 kernel). In GPU-Z in a Win11 VM it very breefly shows x1 and then changes to PCI-Express. Also have seen change it to PCI. Using a PowerShell script to check this, it still says x1.
Also tried Furmark on Win11 and Mint, and it still shows x1.
I saw this first on a bare metal Linux Mint install with kernel 6.14, but thought it was driver/firmware.
have to check in lspci, probably, not xpu-smi? its normal to be in x1 or gen1 for power management. but lspci should have the real truth.
Any idea if this will also work with an Intel Arc Pro A40? I am not able to find a coherent answer to this question …
Opt-in Linux 6.17 Kernel for Proxmox VE 9 is now available on test & no-subscription, is that all that is needed for SR-IOV to work on b50 now? Besides enablement in BIOS of course and GPU firmware update right? Seems like 6.17 will be the new default when Proxmox VE 9.1 releases later in Q4.
will intel be making more of these? they have been out of stock since launch and a lot of scalpers seem to have bought them up on ebay.
yes, huge batches inbound
Not to get greedy but I’m curious if you think we’ll eventually see over-provisioning.
My dream is being able to have 2 VMs open, run an AI with one and get 100% GPU usage, turn the AI off and run a game in the other and get 100% GPU usage. Without having to reconfigure or shut down and swap around ownership.