I recently received my new GeForce RTX 4070 Ti graphics card, and I’ve been trying to get it to perform at least better than my old card to make sure everything is ok before doing anything drastic. The primary symptom is that the 4070 is performing 100% identically to my GTX 960 according to two Phoronix benchmarks on two of my Steam games and a 3rd test of just me going into one of my games that’s pretty intense.
Phoronix benchmarks were using Portal 1 and X-Plane 11. Portal 1 ran at about 723fps on both cards, X-Plane 11 ran at 62fps (low detail), 48fps (mid detail), 30fps (high detail) and 22fps (full detail). Then I have a medium park in Parkitect that just sitting watching the park from a certain spot takes the rate down to around 13fps.
Once I learned that the CPU could be causing a bottleneck, I went to a couple of bottleneck calculators and yeah, they both said my CPU will cause a bottleneck, but they say my CPU will be at 100%, and that’s where I’m not sure that’s my current problem. My CPU is sitting around 55% while the NVIDIA driver says the PCIe Gen2 usage is less than 5%, and the GPU usage is less than 40%.
I was expecting to see at least some improvement over the 960 even if I couldn’t get full capacity without a motherboard & CPU upgrade. I just read in the consumer forum that the FX series of processors had a bug that didn’t let them fully support a high performance GPU. Is this bug what would be causing my CPU to only be 55% use? The thing that’s just a completely weird coincidence is that the FPS stayed identical between the two cards. However, I could understand a bug throttling the speed and my 960 was never used a full capacity either.
System:
CPU: AMD FX-8350
GPU: msi GeForce RTX 4070 Ti (old GPU: ASUS GeForce GTX 960)
RAM: 32GB
Motherboard: Gigabyte GA-970A-DS3P (AM3+ socket, no built-in GPU, I just use a card anyway)
Driver: 525.89.02 (tried the beta 530.30.02 with no change)
OS: Linux Fedora 36 (upgraded from Fedora 35 yesterday to see if that would help)
I think it’s really just the cpu acting as bottleneck. You have to take into account that many/most games and graphics benchmarks are single-threaded or only have few threads so none will ever be able to max out the 8 cores of your cpu. Furthermore, the amd cpu was known for bad single-thread performance.
The second factor is your monitor, based on your benchmark values I suspect you’re only using a FHD monitor. Gpus are pixel-pushers, if there aren’t much pixels, there’s not much to do.
Regarding the “bug”, I don’t really know what that should be about but I remotely remember the Windows scheduler had issues with single-threaded games jumping between cores on some amd cpu, the workaround being pinning the thread to a fixed core. Could have been a different cpu, though and nothing that affected Linux.
Please run nvidia-bug-report.sh as root and attach the resulting nvidia-bug-report.log.gz file to your post.
I’ll do that when I get home tonight. Do I need to be running anything GPU intensive when I do?
I made sure vsync was off for the tests so it wouldn’t limit to 60fps (60Hz monitors), and I could see that all 8 cores were being used mostly equally. I think the Phoronix tests might be different since it effectively plays some of my Steam games for its benchmarks, but it was also the first benchmarking tool I downloaded so I don’t know if it’s at all unique.
As I was trying to resolve this issue and stumbling through driver settings that I’ve never touched, I hit the one that shows FPS on everything. Now that I know about that, I’m going to reinstall my old card tonight and play some games that don’t have a Phoronix test suite so I can just see a wider variety. I think I may have found a game last night after posting that actually did better, but I didn’t run it with FPS before, so it might just be wishful thinking. If even 1 game is faster, then I’ll be happy that it’s fine, and it’s just time to upgrade my MB and CPU. It was just the fact that the fps on my admittedly limited test was identical before and after that had me concerned.
Sorry I didn’t get that output last night, but I decided to do a few more tests, and it looks like everything is ok. I’ll post this follow up in case anyone else is having what seems like weird issues.
It ended up being that yes, my current MB and CPU are underpowered for the new GPU; however, the new GPU is able to do better than my old GPU (what I was trying to see). The unfortunate problem was that the 3 games I decided to originally benchmark were 3 of the only games that don’t seem to fully utilize the graphics card. So I played every graphics intensive game I have and recorded the FPS numbers. Most of the games were able to run with the highest graphic detail on the new card. Then when I switched back over to my old card, the very first game I played was unplayable because the old GTX 960 couldn’t handle it at ultra detail.
The game that showed this best was Dirt Rally 2. New card on ultra detail was between 60 and 90 fps. old card was around 5fps and couldn’t keep up enough for the system to provide force feedback on my wheel.
So now I feel comfortable that the card isn’t defective and the Linux driver is fine. So despite my CPU not being at 100%, there is a bottleneck, but it’s not as bad as my original fears, and I am getting some decent benefit from the new card before I upgrade the rest of my system.