Low GPU FPS, gets fixed only after restart

[kaladinbg]

Hey there, I have the following issue:

When I boot up my PC and start playing a game or run a benchmark my results/FPS are lower than the average for the hardware.
When I restart the PC and try the same things everything runs as expected.
If I put the PC to sleep and then wake it up and try again my performance is worse just like in the first scenario.


Some concrete numbers:
3D Mark TimeSpy for GPU score specifically runs at about 24-25k when performance is lower and at 28k when everything works fine (after a restart)
Red Dead Redemption 2 - in the bad scenarios I get about 100-110 FPS average, in the good scenario I get about 160-170 FPS


Specs: Ryzen 7 7800X3D, 32GB 6000Mhz RAM, RTX 4080 SUPER


The issue is GPU only, my CPU score when running TimeSpy stays the same in all 3 scenarios while the GPU runs well only after a restart. I've also noticed that the watts of the GPU are lower on the bad scenarios hence the worse performance I suppose.


I've tried both DDU and fresh windows install, fast boot is off in both windows and bios. I've tried with EXPO on/off.

I've attached two TimeSpy results. The only difference between the two is that one is after a cold boot and the other is after restart, no other programs or background tasks are running, basically everything else is the same.

 

[Yamoima]

I have the same issue on my setup with Ryzen 7 7700X, RTX 4070Ti, 32GB DDR5 6000MHZ CL30 ram sticks. I tried every step you did too but still couldn't find a solution. My 3DMark results are same situation as you experience. Have you found a solution by any chance yet?

 

[Lew Zealand]

Keep GPU-Z running during both tests and monitor it, have a look at the graph and see if the GPU in the 2 conditions is changing its:

Power usage
Core clock speed
Memory clock speed
Core voltage

See what the limit flags are in each condition, they could/should be different. This won't fix anything but may point us in the right direction.

 

[kaladinbg]

The issue is constant now, no restarting fixes it now. What I notice is that in GPU the bus stays at x1 instead of x16. This is both in idle and when running a game/benchmark.

Hey there, I haven't found anything yet and it's driving me crazy. Have you had any success?

 

[Youseethis]

Use a bios with agesa 1.0.0.7c. That will fix your problem.

 

[Yamoima]

The issue is constant now, no restarting fixes it now. What I notice is that in GPU the bus stays at x1 instead of x16. This is both in idle and when running a game/benchmark.

Hey there, I haven't found anything yet and it's driving me crazy. Have you had any success?

Recently I have been checking background apps if there is any power limit or bug but I didn't find anything related to the issue. I will check the bus speed on my pc and see if there is something wrong. Thanks for your answer by the way

Use a bios with agesa 1.0.0.7c. That will fix your problem.

Are you sure about it? Currently I am running AGESA 1.1.0.2b on Gigabyte B650M DS3H BIOS F22 version. If I need to revert to AGESA 1.0.0.7c , I need to go back all the way to BIOS version F9. I am not sure if it is safe to revert the BIOS version... I would be happy if you can elaborate the situation a bit more.

 

[Youseethis]

I had this problem myself when I installed a bios with 1.1.0.2b so I searched online. There are few reports about pcie being stuck on 8x or less but no one knows why, what configuration triggers it or how to replicate the bug. Every agesa for ryzen 8000 series has this bug (1.0.8.0, 1.0.9.0, 1.1.0.0, 1.1.0.1 and 1.1.0.2). Allegedly amd knows about it and they are working on a fix. According to this chart even the newer agesa 1.1.0.3 is probably affected but the agesa 1.1.7.0 for zen 5 might have a fix. Take a look at the smu numbers, there is an update for ryzen 7000 series:

(Also I think it's safe to rollback these bioses... at least I did a rollback from 1.1.0.2b to 1.0.0.7c because this bug was annoying. I had to test my ram settings though)

 

[Yamoima]

I had this problem myself when I installed a bios with 1.1.0.2b so I searched online. There are few reports about pcie being stuck on 8x or less but no one knows why, what configuration triggers it or how to replicate the bug. Every agesa for ryzen 8000 series has this bug (1.0.8.0, 1.0.9.0, 1.1.0.0, 1.1.0.1 and 1.1.0.2). Allegedly amd knows about it and they are working on a fix. According to this chart even the newer agesa 1.1.0.3 is probably affected but the agesa 1.1.7.0 for zen 5 might have a fix. Take a look at the smu numbers, there is an update for ryzen 7000 series:

(Also I think it's safe to rollback these bioses... at least I did a rollback from 1.1.0.2b to 1.0.0.7c because this bug was annoying. I had to test my ram settings though)

Thanks for clear explanation. Now it makes more sense why we have this particular issue. I am gonna rollback to a BIOS with AGESA 1.0.0.7c and see if the issue still exists.

 

[wNotyarD]

Every agesa for ryzen 8000 series has this bug

Isn't every current Ryzen 8000 Phoenix-based? Could that be a reason behind the PCIe bus getting limited?

 

[Youseethis]

When they released the bios update to support 8000 series, they changed something for the 7000 series (smu 84.79.222>84.79.223). Evidently something broke. They might have neglected to test pcie 16x connectivity since ryzen 8000 series only support 8x on the main slot + (4x+4x) on nvme for phoenix or 4x main+(2x+2x) nvme for phoenix2. Who could have guess that something like pcie connectivity would break anyway? In any case, I am waiting for agesa 1.1.0.3.

 

[Yamoima]

I finally managed to figure it out. The issue happened again when I was playing Cyberpunk 2077. I checked the bus interface using GPU-Z and as mentioned earlier here, the bus speed was stuck at "PCIe x1 4.0" causing to get lower fps and power consumption than intended. I think I am going to wait until they release a new BIOS with AGESA 1.1.0.3 . So far, restarting the setup fixes my problem temporarily.

 

[kaladinbg]

I did revert to F2 which is with AGESA 1.0.0.7c but it didn't fix it for me. I'll try swapping for the MSI Tomahawk to see if this does it.

 

[Youseethis]

For me the main pcie gets stuck at 8X after a cold boot or after waking from sleep. I also tested again with the latest bios (1.1.0.2b) 2 rtx 4090s, one gigabyte and one msi and both dropped to 8x after waking from sleep. I have not seen it drop during gameplay or going below 8x.

 

[Yamoima]

I did revert to F2 which is with AGESA 1.0.0.7c but it didn't fix it for me. I'll try swapping for the MSI Tomahawk to see if this does it.

Have you checked the issue with new motherboard yet?

 

[Youseethis]

I tried to test with agesa 1.1.7.0 but windows wouldn't boot. It just just hard crashes and resets while loading with the windows symbol on screen. I rolled back yet again and it worked just fine again. I have seen something similar on a pc when I updated from agesa v1 1.0.0.6 to agesa v2 1.2.0.3c to support ryzen 5000. I did a clean install back then and everything works fine since then. For now I am done testing my main pc. It's too annoying

 

[Yamoima]

I tried it with AGESA 1.1.7.0 on B650M DS3H rev1.0 with F24c BIOS. The issue still exists. Bus Interface gets locked at "PCI-Express x 1".

 

[Yamoima]

Downgraded to BIOS version F9 with AGESA 1.0.0.7c and the issue is fixed. Bus Interface runs at 4.0 x 16 all the time. Including cold boot, sleep and restart.

 

[Ruru]

I've had this issue only once and that was with an old GTX 560 Ti 448. IIRC it was a driver issue on my case.

 

[Youseethis]

The pcie 8x bug is fixed on msi boards with agesa 1.1.8.0 (stable).
Be careful though, there is another bug that affect nvme ssd's since agesa 1.0.9.0 bioses. The can't be detected at all on cold boot and you need to restart the pc in order to be detected. Some affected models are wd850x, samsung 990 pro and msi m570 spatium. This one is not a specific agesa bug. It's an msi bug as far as we know but they have identified it, replicated it and they are fixing it (probably agesa 1.2.0.0 bios releases will have the fix).