I've bought a Radeon Pro W5700 and paired with an AKiTiO Node Titan eGPU enclosure to connect to my Lenovo Yoga 9i 14 ITL5 (Yoga Pro 14c 2021) which has a Intel Core i7-1185G7 processor. I directly use the internal screen of my laptop and don't connect any display to the W5700.
I succeed in connecting the W5700, installing any necessary drivers (Pro driver 24.Q2) and passing 3DMark Steel Nomad stress tests. However, after a while the eGPU started to cause serious disconnecting issues, which eventually cause IOMMU faults, DMA errors and ultimately a BSOD.
From Windows Event Viewer, there are many disconnect and reconnect events from nhi (the thunderbolt driver) in the System section as well as PCI SURPRISE_REMOVED events in Windows/PCI section.
I've tried to connect an empty enclosure without the W5700 GPU and it would disconnect/reconnect every 60 seconds. However, the reconnect interval with W5700 is varying from only several seconds to several minutes.
I guess that it would be caused by the thunderbolt controller inside the enclosure is resetting either caused by the GPU not detected or something else. The link between the enclosure and the computer is probably fine as well as the GPU, for the eGPU system would sometimes get normal without any resetting issues even after firstly experiencing the disconnecting issue. However I have no clue to prove that the controller is resetting caused by GPU link issues or any incompatibilities between the devices.
Does anyone have similar issues and resolutions?
