Yep, but not this model.
They have an unannounced motherboard variant, without USB 4 v2 support.
About the USB 4 V2 of the mini PC, the TB 5 controller connects to a PCIe x4
Yep, but not this model.
They have an unannounced motherboard variant, without USB 4 v2 support.
About the USB 4 V2 of the mini PC, the TB 5 controller connects to a PCIe x4
My friend told me in an APU system, the APU VBIOS will help for initialize dGPU VBIOS, and this part is very buggy.
But the report is rare because there’re too few of Laptops which equiped APU + AMD dGPU
Support got back to me yesterday.
Graphics cards are not officially supported and “-behavior may vary depending on BIOS behavior and eGPU models.”
We’re talking motherboard BIOS? So, Framework’s BIOS?
If it’s Framework motherboard BIOS then surely such capability (adding an eGPU) is a Framework choice? Or am I missing something?
It would be their BIOS but that’s not the only limiting factor.
There’s various different limits going on right now but I’m still working on a way to get a graphics card running.
Can you test connecting your AMD dGPU to the M.2 slot?
My friend asked Minisforum to do more tests, and the result is that the AMD dGPU works well in the M.2 slot.
A possible reason is that the AGESA will also help initialize dGFX on a specialized PCIe slot, while M.2 slots appear to skip the initialization, allowing Windows to handle the task.
However, they forget to gather the power consumption data. I’m afraid they would encounter the same situation as I did, where the power of the dGPU is limited to 120 watts.
Finally, my friend asks Minisforum to disable SmartShift, and they’re considering disabling it on production BIOS. If that works, we can propose the same change to Framework.
I’ve spent the last four hours trying to get a good, repeatable boot with the 7900 xtx on ANY pci-e lane and I just cannot make it work. I’ve got the GPU connected to the rear M.2 slot. The nvidia card seems to work a lot better on the m.2 slots but right now I’m fighting some type of bios issues immediately after boot depending on what display the bios sees as available. These problems don’t manifest when the nvidia card is installed instead.
I’ve switched my setup now to the JHH DOCK-OC4 eGPU oculink board. It provides external power for 12v, 5v, and 3.3v. It also lets me set a delay on the PERST and REFCLK lines. Delaying REFCLK seems to work the best in terms of getting a booting system but can tank your PCI-e speed training. With this setup, I get 4 lanes of PCI-e 4.0 which is all I’m expecting.
Lastly, I’ve recently gone through the schematics that were released. I’m very disappointed in the choices that were made as current desktop units being shipped WILL NOT be able to support graphics cards as there is an load switching IC (UK3) which controls power to the PCIE slot and limits the wattage to under 24 watts. This seems to be controlled by RK35. If support is ever added via a bios update, YOU WILL NEED to have an external source of 12v power for the PCI-e port unless you also modify the hardware. It seems like a decent amount of reworking was done to try and get this working but was not done due to timeframes on release. Framework-Desktop/Mainboard/Mainboard_Interfaces_Schematic_Framework_Desktop_Ryzen_AI_Max.pdf at main · FrameworkComputer/Framework-Desktop · GitHub
It is absolutley brilliant of you to be putting in so much effort @Lukew4lker Thank you very much!
I personally would be OK adding an external card plus power. I would be very disappointed if the firmware excluded this option
It would be helpful to hear from Framework in relation to adding a GPU (d or e) to the Desktop @nrp … pls
It does have two m.2 slots, so you can use an Oculink Adapter.
Requesting comments from @geerlingguy as he’s done a bit of work with PCI-e devices in his GPU on PI series on YouTube.
I really hope the BIOS / Drivers can get updated for eGPU enclosures to work properly. I was considering picking up a DEG1 to throw my 7900XTX into but saw this thread and was disappointed.
Has anyone tried using a USB-C eGPU enclosure? I realize its less bandwidth and not as good, but would be interesting to know if that works better for AMD eGPU’s.
TB gpu docks seem to work fine as they initialize once you get past the BIOS. My setup below is a TB3 board from the HP OMEN Accelerator in a 3d printed case with the Nvidia GPU.
It looks to me, based on a photo in this thread, that @Frix_Rioux is running an AMD eGPU through an Oculink card
I have not even installed an OS yet, I’m not there. I’m gonna see what I can do, may have better chance with an NVidia card but I’m gonna try some stuff with the AMD card too, maybe I will get lucky hacking something.
I’ll make a little update. I installed Debian Trixie and started playing around. I have an Minisforum DEG1 with a Micro SATA Cables reDriver oculink adapter and a high quality oculink cable. I tried LMStudio at first, but it have problem detecting the igpu when paired with a dgpu so I switched to llama.cpp. Pluging a Nvidia card give me no problem. For a amd card, things get weird, I have a weird delay before the bios start. When linux start I get error, timeout and sometime the cpu lock. I tried many thing for hours, in the like of powering the dpgu before or after starting the computer, witch hdmi port to use, some grub options. I did not manage to make much sense of it. It managed to boot 3 times. It feel that it take like 5 min to boot. I cannot help much as I do not understand what is going on.
Okay, I’m managing to boot consistently with my 9070 XT. I’m using x11 and gdm3 and using the amd_iommu=off grub option if that matter. Here is my method:
Yeah, this is weird and do not make sense and it’s hard to be confident there is not something else involved.
Interesting. So it’s only AMD dGPU that gives you this issue? Not the Nvidia one? Strange, I would have thought that given this is an AMD product it would be reversed.
So I wouldn’t be able to plug a 3090 into the X4 slot with a riser cable, unless it had external power?