Well, it is probably a RDNA problem then. If it is any consolation, with AMD GPUs the RX
and TX
numbers seem to be an upper bound at best and not an actual estimate of the data transfer rates, judging by the implementation.
Otherwise my expectation is that all recent AMD GPUs work after UEFI hands off the reins to Linux as long as the kernel version is at least 6.15
and the firmware build is the latest release, i.e. 0.3.0-1
(and possibly others such as 9.0.0
); ACPI mode would be necessary, of course. That is certainly my case - I even managed to get Steam and Cyberpunk 2077 running, albeit with an unplayable framerate. In fact, even UEFI is visible for me - FYI my card is MSI Radeon RX 6700 XT MECH 2X 12GB OC. Also, if I block the amdgpu
kernel module from loading, I still get output from the card, with the same reduced functionality as from the integrated GPU (i.e. a simple framebuffer device and software rendering).
Similar to you, everything works fine except that I discovered one annoying issue - my display has only 3 inputs, but I juggle 4 devices with it. The Orion O6 board (well, the external GPU on it) is connected via a HDMI cable, but I have to detach it to use one of my other computers. However, shortly after I disconnect the cable the machine locks up and needs a reset. The only symptom that I noticed were the following messages that appeared in the kernel log (I was connected via SSH from another computer):
[ 4710.788374] arm-smmu-v3 arm-smmu-v3.0.auto: event 0x10 received:
[ 4710.788398] arm-smmu-v3 arm-smmu-v3.0.auto: 0x0000c30100000010
[ 4710.788414] arm-smmu-v3 arm-smmu-v3.0.auto: 0x0000020000000000
[ 4710.788419] arm-smmu-v3 arm-smmu-v3.0.auto: 0x0000000000000000
[ 4710.788422] arm-smmu-v3 arm-smmu-v3.0.auto: 0x0000000000000000
[ 4710.788426] arm-smmu-v3 arm-smmu-v3.0.auto: event: F_TRANSLATION client: 0000:c3:00.1 sid: 0xc301 ssid: 0x0 iova: 0x0 ipa: 0x0
[ 4710.788433] arm-smmu-v3 arm-smmu-v3.0.auto: unpriv data write s1 "Input address caused fault" stag: 0x0
I am currently on kernel version 6.17-rc2
, but it was the same on some versions of the 6.15
and 6.16
releases. If I keep the cable connected and leave the machine alone, even for more than a day, it still remains responsive afterwards. It is probably another dark corner of the amdgpu
code that needs a fix…