InternalError: Blas GEMM launch failed : a.shape=(65536, 63), b.shape=(63, 256), m=65536, n=256, k=63 [Op:MatMul]

[Eric-Morcy]

I'm running python run_nerf.py --config config_fern.txt And I get this kind of issue, any big brother knows how to fix this issue?

ps: I have the environment right, I have a big GPU memory(16G 4070 ti super), I have a compatible cuda version, I have the latest Graphic driver. And I dont know what to do

[atat1010]

Have you solved it? I have the same problem,could someone help me?

[windsky9998]

I'm using a 4070Super with 12G GPU memory and I'm still having this issue. I even used AI to generate a script for testing, but the script worked without any errors. The normal operation of this script shows that my graphics card driver is compatible with the cudnn version number, and the cuBLAS library can also be called and perform matrix calculations normally, and the video memory is not full at all, all possible problems have been eliminated, and I am beginning to wonder if this code has been abandoned by the times for too long. I'm trying to reassemble the code with an existing new environment, but I don't have any expectations.

[Ga1ahad]

I have the same problem,but I have successfully ran NeRF (pytorch version). I still want to understand and solve it

[Emily1112334]

i have the same problem,i don't kown what is the reason

[AsakuraKumoumi]

It seems that this issue is caused by a mismatch between the TensorFlow version and GPU model. TensorFlow 1.x version has stopped updating, therefore it does not support Nvidia GPU devices from the 30 series and above (including the 40 series) any more. Try installing TensorFlow in this repo https://github.com/fo40225/tensorflow-windows-wheel.git which compiles a higher version of CUDA to solve this issue.