Cuda compile errors

[himat]

I'm getting errors when compiling by running cmake --build build --config RelWithDebInfo -j 16

The build error log is long, so I put it here: https://pastebin.com/6tRsjYfM

Is this due to some incorrect cuda version incompatibility? I'm on cuda 11.2

(tensorflow2_p38) ubuntu@ip-172-31-40-250:~/instant-ngp$ python -V
Python 3.8.12

(tensorflow2_p38) ubuntu@ip-172-31-40-250:~/instant-ngp$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Feb_14_21:12:58_PST_2021
Cuda compilation tools, release 11.2, V11.2.152
Build cuda_11.2.r11.2/compiler.29618528_0

(tensorflow2_p38) ubuntu@ip-172-31-40-250:~/instant-ngp$ cmake --version
cmake version 3.22.1

(tensorflow2_p38) ubuntu@ip-172-31-40-250:~/instant-ngp$ echo $PATH 
/usr/local/cuda-11.2/bin:/home/ubuntu/anaconda3/envs/tensorflow2_p38/bin:/home/ubuntu/anaconda3/condabin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/home/ubuntu/anaconda3/condabin:/home/ubuntu/.dl_binaries/bin:/usr/local/cuda/bin:/opt/aws/neuron/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin

(tensorflow2_p38) ubuntu@ip-172-31-40-250:~/instant-ngp$ echo $LD_LIBRARY_PATH 
/usr/local/cuda-11.2/lib64:/usr/local/cuda-11.2/extras/CUPTI/lib64:/usr/local/cuda-11.2/lib:/usr/local/cuda-11.2/efa/lib:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/usr/local/cuda/lib:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64:/usr/local/cuda/targets/x86_64-linux/lib:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/usr/local/lib:/usr/lib::/home/ubuntu/anaconda3/envs/tensorflow2_p38/lib/python3.8/site-packages/tensorflow

[mmalex]

at a quick glance i can see an error about unsupported old cuda architecture. perhaps the gpu auto detection has failed. what gpu is on the build machine, and is it the one you intend to use? anyway, you can set an explicit gpu architecture as an environment variable before running cmake. from the readme: TCNN_CUDA_ARCHITECTURES enivonment variable for the GPU you would like to use. Set it to 86 for RTX 3000 cards, 80 for A100 cards, and 75 for RTX 2000 cards. let me know if that helps!

[himat]

Ah I see. I'm trying to run this on an AWS ec2 instance and it has a K80 GPU. (there's only one GPU on the machine I'm trying this on so I thought I wouldn't need to specify the architecture)
There are also instances with a V100.
Do you know what numbers to use for either of these? Or what aws instance you'd recommend?

[Tom94]

Unfortunately, both K80 and V100 GPUs are unsupported.

This codebase requires at least compute capability 75+, which means RTX 2000-series GPUs, A100 GPUs, or RTX 3000-series GPUs.

See also #13. I will momentarily update the README to reflect this more explicitly

[Tom94]

Apologies, but it's currently not in our scope to add dedicated implementations for older GPUs.

In part, because earlier GPU's performance would benefit much less from a fully fused implementation due to them being more compute rather than memory-bound than newer GPUs on the small-MLP workload.

All that said, I would still be more than happy to merge code contributions that improve compatibility.