SOLUTION:

I was missing this package sudo dnf install rocm-hip-devel as per instructions here: https://fedoraproject.org/wiki/SIGs/HC

Hi, I'm trying to get GPU acceleration on AMD to work in Blender 4.1 but I can't seem to be able to. From what I've seen it should be working with ROCm just fine but I had no luck with it.

I'm using Fedora 40 GNOME with Wayland and my GPU is RX 6800 XT.

System is up to date. I've also installed all these packages:

sudo dnf install rocminfo

sudo dnf install rocm-opencl

sudo dnf install rocm-clinfo

sudo dnf install rocm-hip

and restarted system after.

rocminfo gives me this

rocm-clinfo gives me this

___``___

I put my laptop into sleep by just closing the lid and sometimes after wake up the clock shows wrong time and date which results into web sites not loading until I change it back and then I also lose logins to my previously logged in websites and have to re-enter the logins.

I don't have dual boot so it's not Windows messing with the Linux time issue I was able to find.

I use automatic time and date with my time zone selected. Not sure what I should do about this.

I've tried to enable RTC in local time zone but that made things even worse so I disabled it again.

When they are already finishing each others sentences.

I'm unable to pair my Wacom tablet to my notebook via BT. I think I know the root cause of the issue which is the version of Bluez being 5.64 and the behavior of the issue is the same as I had on Fedora on my main PC few months back when I've tried to pair PS5 Dual Sense controller.

I was able to fix the issue on Fedora by downgrading the Bluez version with "sudo dnf downgrade bluez" but on Pop!_OS this does not work and neither I was able to upgrade to one of the newer versions 5.65 or 5.66 (it says I need to compile these?).

I'm new with Linux so I'm not sure how I would go about compiling something to make it work. The 5.65 version seems to have bugfixes for my issue but I'm perfectly fine with downgrading to anything older as long as it works if it's simpler to do.

To describe the issue:

When I put tablet into a pairing mode it will appear in the Pop!_OS system and I can select it to pair. It will attempt to pair but then fails.

I've tried to force it to pair with "bluetoothctl trust/pair/connect MacID" commands and it managed to say "connected" but the tablet was still not responding and the LED was blinking as if it was still in pairing mode. I've also tried the Blueman GTK BT manager with no luck.

Note: It works perfectly well via USB connection. It also connects instantly to my main Windows PC via BT so the tablet is not at fault.

EDIT;

FIGURED IT OUT

by default laptop was non-discoverable and there is no system setting on either Pop!_OS or Fedora to make it discoverable. Had to use Blueman to make it so. I've only noticed that when doing "sudo systemctl bluetooth" for XYth time and there was "Disoverable: No" next to one of the MacIDs. After figuring out the ID belongs to my laptop I just had to figure out how to make it discoverable and turns out Blueman can do it.

• RED = Thermal Grizzly Carbonaut
• GREEN = PTM 7950 (no idea if the original or some fake but it works, bought from random seller on Aliexpress)
• before Arctic MX-4 was hitting TJmax 110C at 312W (good temps comparable to Carbonaut for first few days after application then drastic increase in short time due to pumpout issue)
• stock paste was hitting TJmax 110C at 255W (when I got the card it was used and already almost 2y old)

Unfortunately I don't have any data for the MX-4 and stock paste to put in the graph.

Testing was done with TimeSpy GPU Test 2 on a loop for 10 loops.

Anyone can help me? I wasn't able to find any solution for this. The controller works via USB-C just fine but I only have a very short cable, I borrowed the controller from a friend to try it and don't have the original cable but I was intending to play via Bluetooth anyways.

Basically, I can find the controller via pairing mode but when I try to pair it I get error:

The Setup of Dual Sense wireless Controller has failed.

After that I can see it in available wireless devices but when I try to connect to it it will immediately disconnect again, checked with bluetoothctl.

Using on-board Intel AX200 wireless controller.

I switched from Windows to Fedora last week and I'm monitoring the stats with Mangohud when playing games. I used to run HWinfo on 2nd monitor when using Windows 11.

I have 6800X ( default voltage) . The card maintains higher clocks at lower power most of the time. I've set the same OC as on Windows with a 2700MHz max clock and in games I'm sitting pinned at 2670MHz-2700MHz almost all the time in Linux when I don't hit power limit (312W) while on Windows the actual clock barely went over 2600MHz and card was almost always bouncing off of power limit resulting in massive clock drops to 2300-2400MHz. On Linux the drops go down just by like 100MHz-130MHz at most in the same scenarios.

Unfortunately I'd need to install Windows again and do proper testing to compare but I wonder if anyone else can confirm/deny this to me.

At least on idle I can confirm for a fact that the card uses less power, usually around 30-35W while on W11 was like 40-50W.

But I've spent most of the time tweaking and setting up and downloading stuff rather than actually playing. Games seem to work really well. Not doing benchmarking but I really like how stable the framerate is when frame cap is in place. So far everything I've tried was absolutely buttery smooth.

I've got a 2nd hand RX 6800 XT about 3 months ago and immediately had to repaste it as the thermals went well above 100C even at 255W. I used a new tube of MX4 which I kept as a backup and the temp went down significantly for about 2 weeks but then hot spot started to creep up until the difference between GPU Temp and TJ Max was 30C+ at times and with my OC at 300W it was often hitting 105C-109C on TJ Max in Time Spy (but even in games the hotspot would sometimes randomly jump to over 100C even if the game was using 200W).

So my theory was that there is a thermal paste pump-out due to thermal cycling which would explain why the temp was going up so fast after repaste but it took me until now to try the Carbonaut pad which I assumed could fix the issue.

I've used Time Spy GPU Test2 on 5 loops to get these results for comparison. GPU was set to 300W and 2600MHz at stock 1150mV. GPU fan speed fixed with side panel on the case as well.

I've started in the morning so room temp went only up until I've got all my results which means that the pad results are slightly better than what I've measured.

After replacing the paste with a pad the TJ Max did go down by about 6C-9C and I was only hitting about 100C at most BUT the core went up significantly by almost 20C from around 78C to 95C.

This was definitely disappointing as this affected the GPU clocks quite significantly and resulted in around 250MHz drop.

But because the hotspot went down this made me think that there just must be insufficient contact or cooler pressure so I was able to find some rubber washers or O-rings or whatever those are in a garage and I took off the retention plate and installed them. I tried to screw the plate back on as evenly as possible with just a normal screw driver and I hoped I wont crack the die by using too much pressure.

Results are absolutely stellar as I've got almost 20C drop on hot spot vs paste (around 9C improvement vs pad without washers) which makes my Time Spy max out at 91C on hot spot. Also the GPU temp went down by more than 20C vs pad without washers and around 5C lower than vs paste to around 73C in Time Spy.

So all in all I'm quite happy with the results. Washers probably did the most as I think doing washers + paste would get me similar results or maybe even better but I'm not going to try.

If you decide to go for the pad I recommend to get larger than 32x32mm one as I did as it's just big enough with almost no room for error if it moves during installation.