RuntimeError: CUDA error: no kernel image is available for execution on the device

[BunyaminEfe]

Why am I getting this error you see?
python3 joint_monkey.py command works fine. isaacGYM works fine

[cypypccpy]

Hi @BunyaminEfe ,

This is always due to a mismatch between the pytorch version and your GPU cuda driver:

According to the above error message, you can try to upgrade the pytorch version corresponding to the cuda driver.

Hope this can help you.

[BunyaminEfe]

English version of the fault text in the last line
Parçalama arızası (çekirdek döküldü) = Shredding failure (core spilled) What is the solution of the line here?

[cypypccpy]

Hi @BunyaminEfe ,

This may be because of this line that reports the error:

MESA-INTEL: warning: Performance support disabled, consider sysctl dev.i915.perf_stream_paranoid=0

I found a post similar to your error on the official forum of Isaac gym, You might be able to refer to the answer in it:

Anyway, since Isaac Gym is still in development, empirically it will have kinds of issues. Sometimes I also look for answers on the official forums I mentioned above, which are very helpful.

Hope this can help you.

[BunyaminEfe]

I looked at the place you mentioned at the moment, on the 470 installed computer, but now I'm upgrading to 520. I'm thinking of trying this shape.

[BunyaminEfe]

Upgrading version 520 was not a solution unfortunately @cypypccpy

[cypypccpy]

How about using this setting? It looks like the questioner above solved it this way:

Oh, I understand,
Using this setting :
export VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json

[BunyaminEfe]

i solved that problem but now i am facing this problem

[cypypccpy]

Oh, we have also encountered this problem before. It seems that you are using Cloud Container, Our experience suggests that Isaac Gym doesn't seem to support this well, so we solved it by switching to Cloud Hosting.

[BunyaminEfe]

@cypypccpy I am not using Cloud Container. My computer is giving the error right now.

[cypypccpy]

Okay, sorry because we've only encountered this at Cloud Container before. I'm sure this is an environment configuration problem, can you show me the result of typing nvidia-smi on the command line? I hope to help you solve this problem. @BunyaminEfe

[cypypccpy]

How about these post? These errors are similar to yours, I don't know if it helps: https://forums.developer.nvidia.com/t/weird-bug/210419, https://forums.developer.nvidia.com/t/cuda-and-driver-version/209140, https://forums.developer.nvidia.com/t/ubuntu-20-04-libpython3-7m-so-1-0-error-segmentation-fault/163948

Or you can try to use this my cuda driver that works well, maybe 520 has some problems because it is too new：

[BunyaminEfe]

I updated it to version 520 because it gave error on version 470 but it's still the same
GPU = NVIDIA GeForce RTX 3050 Laptop GPU

[cypypccpy]

Hi @BunyaminEfe ,

I'm so sorry that I couldn't solve your problem, but we and some other people use bi-dexhands well in different computer environments, so I think it may not be a problem with the code. Maybe you can go to the official NVIDIA forum and ask a question about this issue, or try another computer.

[BunyaminEfe]

Hi @cypypccpy ,
Thanks for your help. I wish you well in the project you are developing. :)

[qinge0]

Do we need to use cuda when configuring the environment for this project?

[BunyaminEfe]

@qinge0 yes cuda is needed