Dual Arc Pro B60 Windows 11 Graphics and Fan Control Issues

[phillipcarroll]

265K

Z890 Asus Proart MB

Dual ASRock Arc Pro B60 GPUs each in cpu lanes x8 slots

The metrics and tuning sections in the Intel Pro Graphics Software for each B60 gpu populate data from only the primary B60 display out gpu. 

Ideally I would have my display out B60, then my other B60 sitting idle and used when doing inference, pytorch, gaming and pass any related video output through the primary GPU. A typical dual GPU setup. This btw works fine if I put literally any other GPU as a secondary, including an arc pro b50, or an rx 6800 xt etc...

I have ran into this same exact problem running dual B580's, dual A770's, it seems like only 1 gpu can be controlled at a time in the software. It also means that windows cannot address the second GPU under graphics settings. If I set the second GPU as the primary for performance/graphics use it will just revert back to the GPU that is outputing display. 

I have reinstalled drivers, DDU'd , removed just the intel graphics software, reinstalled, bounced the graphics service, reboots between all these. 

I have tried the gaming drivers, and the pro drivers. I have reinstalled windows. I thought these gpus were mean for up to 8x in a single system. 

In linux everything seems to work fine. But what do I need to do in windows to make this consistent so I can set independent fan profiles, the metrics show data for each gpu and not just the same data which is clearly from the primary gpu outputing display.

How can I get windows to stop reverting back to selecting  just the primary gpu as the performance unit.

[phillipcarroll]

I should add, running 1 gpu at a time everything works 100%

[Porgy]

I am really p**ed off, I bought 2 Intel Arc Pro B60 Creator 24gb GPUs running on a Rog Strix 8x8 PCIe
I have tried every thing in my power to get these to work, I am running them in LM Studio and through  Open WebUI.
In LM Studio I can get  10% satisfaction in performance, in Open WebUI that becomes 0% satisfaction, a little Gemma4 12b QAT model is just a flashing dot, maybe if I am lucky it will break through the prompt and say hello, then the inference speed is a slow lazy dog style maybe 2 tokens a second on a lucky day.

So I thought my daily PC not a purpose built PC like the AI PC with the 2 x Intel B60s machine, in my general PC I have a Nvidia RTX 3060 12 GB Vram GPU   so I thought let me pull a small model from my B60 laden machine across and try it in my 1 Nvidia RTX 3060 machine, Downloaded LM Studio loaded the model Gemma4 B12, INSTANT hello from the model.
So I thought okay lets bring Gemma4 26b and Gemma4 31b across to the Rtx 3060 machine, and there you have it they both loaded they both worked well, had to reduce GPU offload a little on the 31b, but can live with that. - Fact in my case 1 Nvidia RTX 3060 works better on 12 GB Vram than the 2 X B60s at 48 Gb RAM!

I ordered another Rtx 3060 from Amazon, it arrives next week and the 2X Intel B60s are coming out of the AI Computer and the 2 RTX Nvidia 3060 GPUs will now replace them in my AI pc along with my 45 other Models that will all at last run properly.

I am seriously disappointed in Intel, I am in South Africa and Intel Arc Pro B60s are not cheap, at least my doors won't slam shut anymore  in the wind, My 2 month old Intel B60s will made good door stoppers...Seriously Intel, Get your act together!