- Apr 27, 2022
- 21
- 6
- 1,515
Hi!
I have a home-built PC that I set up about 2 years ago for 3d art with Blender. It includes a Ryzen 9 5950x CPU, RTX 3080 ti GPU. This week, I was rendering an animation, and I noticed that every frame of the animation was taking longer and longer to render, despite the frames being pretty similar. The first few were taking around 2-3 minutes to render, and by frame 27, I was killing the render after it had gone for 2.5 hours. I considered the possibility of a memory leak (even though I had previously done much longer renders without issues), so I tried closing blender, and even entirely logging out of my account on the computer, then starting the render up again, but I was still not able to make any progress when I started up again.
So, I rebooted. When I shut down, it took a pretty long time to even start booting up again, and then I got the bios message saying that I had a new CPU, and needed to set up the BIOS again. (I don't remember the text of it, but I think it's pretty standard when you change your CPU). All I had done was reboot, but I went in, and reset my BIOS settings (pretty much defaults except that XMP is on), then booted up. This time, restarting the render at frame 27 went great, and it was back to ~2.5 minute renders for a couple frames, and then started slowing down. By around frame 35, I killed it because it had been going for over an hour, and was 10% through the frame.
Rebooted again. This time, no weirdness in the reboot, but when it came back, the renders were still not moving at anywhere near a reasonable speed. There were a couple times that when I started a render, it quit claiming to be out of GPU memory within a couple minutes, but when it didn't have GPU memory issues, it was clear once it started doing samples that we were looking at a multi-hour frame, so I cancelled the render. Cancelling a render hung pretty badly and usually required using Task Manager to kill Blender completely.
So, I switched blender to rendering with CPU instead of GPU. Works fine -- 12-15 minutes per frame, which sucks compared to 2.5 minutes, but is pretty reasonable for CPU rendering. It is able to do plenty of frames in a row without any significant change in per-frame render time, so it doesn't seem to be having issues with that.
I also was able to move the blender file to a similarly-spec'd computer and verified that the other computer was able to do 30+ frames in a row in GPU render mode without significant change in per-frame render speed, so the slow-down seems to be machine-specific.
So, it *appears* to me that this is a GPU failure, or maybe a GPU RAM failure? Does that seem like a reasonable conclusion? Is there anything else that I'm missing that could cause this type of failure? Or, is there anything else that I should be trying/testing before I go through the 3+ week RMA process? (GPU is still under warranty.... as much as I'd love the excuse to upgrade, I'd rather get a free replacement).
I honestly just replaced the CPU in this machine about 2 months ago due to a CPU failure, so I'm wondering if there might be something else b0rk3d in the machine that is causing repeated hardware failures, although I admit that with 3d art, I am pushing the computer pretty hard and frequently do long renders / animations. I do have pretty good cooling, and whenever I check, the cpu and gpu temps are generally within normal ranges. (I have an iCue water cooling system for the CPU, and the case has a total of 10 fans...). I use a Corsair HX1200 PSU, and it seems to be fine. I admit, though, I'm not particularly good at hardware, and this is the first machine I ever built, so if there's something I could've done that would cause two components to fail so close together, 2 years after building, I'd like to fix it!
Thank you so much for the help!
I have a home-built PC that I set up about 2 years ago for 3d art with Blender. It includes a Ryzen 9 5950x CPU, RTX 3080 ti GPU. This week, I was rendering an animation, and I noticed that every frame of the animation was taking longer and longer to render, despite the frames being pretty similar. The first few were taking around 2-3 minutes to render, and by frame 27, I was killing the render after it had gone for 2.5 hours. I considered the possibility of a memory leak (even though I had previously done much longer renders without issues), so I tried closing blender, and even entirely logging out of my account on the computer, then starting the render up again, but I was still not able to make any progress when I started up again.
So, I rebooted. When I shut down, it took a pretty long time to even start booting up again, and then I got the bios message saying that I had a new CPU, and needed to set up the BIOS again. (I don't remember the text of it, but I think it's pretty standard when you change your CPU). All I had done was reboot, but I went in, and reset my BIOS settings (pretty much defaults except that XMP is on), then booted up. This time, restarting the render at frame 27 went great, and it was back to ~2.5 minute renders for a couple frames, and then started slowing down. By around frame 35, I killed it because it had been going for over an hour, and was 10% through the frame.
Rebooted again. This time, no weirdness in the reboot, but when it came back, the renders were still not moving at anywhere near a reasonable speed. There were a couple times that when I started a render, it quit claiming to be out of GPU memory within a couple minutes, but when it didn't have GPU memory issues, it was clear once it started doing samples that we were looking at a multi-hour frame, so I cancelled the render. Cancelling a render hung pretty badly and usually required using Task Manager to kill Blender completely.
So, I switched blender to rendering with CPU instead of GPU. Works fine -- 12-15 minutes per frame, which sucks compared to 2.5 minutes, but is pretty reasonable for CPU rendering. It is able to do plenty of frames in a row without any significant change in per-frame render time, so it doesn't seem to be having issues with that.
I also was able to move the blender file to a similarly-spec'd computer and verified that the other computer was able to do 30+ frames in a row in GPU render mode without significant change in per-frame render speed, so the slow-down seems to be machine-specific.
So, it *appears* to me that this is a GPU failure, or maybe a GPU RAM failure? Does that seem like a reasonable conclusion? Is there anything else that I'm missing that could cause this type of failure? Or, is there anything else that I should be trying/testing before I go through the 3+ week RMA process? (GPU is still under warranty.... as much as I'd love the excuse to upgrade, I'd rather get a free replacement).
I honestly just replaced the CPU in this machine about 2 months ago due to a CPU failure, so I'm wondering if there might be something else b0rk3d in the machine that is causing repeated hardware failures, although I admit that with 3d art, I am pushing the computer pretty hard and frequently do long renders / animations. I do have pretty good cooling, and whenever I check, the cpu and gpu temps are generally within normal ranges. (I have an iCue water cooling system for the CPU, and the case has a total of 10 fans...). I use a Corsair HX1200 PSU, and it seems to be fine. I admit, though, I'm not particularly good at hardware, and this is the first machine I ever built, so if there's something I could've done that would cause two components to fail so close together, 2 years after building, I'd like to fix it!
Thank you so much for the help!
Facebook
Twitter
Instagram