Help setting up CUDA.jl and cuTENSORS

Specific Domains GPU
, , , ,
1

Hi all,

I am having trouble using CUDA.jl and cuTENSOR and would appreciate your help. I’m looking to use the ITensors.jl library with GPU acceleration (via ITensorGPU), which uses CUDA and cuTENSORS. I am running this on a NVIDIA Jetson AGX Xavier. On my first few attempts at installing all dependencies, I got internal errors with cuTENSOR. I backed up a few steps, and first want to verify that CUDA.jl is working propertly.

Here is my system information:

┌ Info: System information:
│ CUDA toolkit 11.6, artifact installation
│ Unknown NVIDIA driver, for CUDA 11.4
│ CUDA driver 11.4
│
│ Libraries:
│ - CUBLAS: 11.8.1
│ - CURAND: 10.2.9
│ - CUFFT: 10.7.0
│ - CUSOLVER: 11.3.2
│ - CUSPARSE: 11.7.1
│ - CUPTI: 16.0.0
│ - NVML: missing
│ - CUDNN: 8.30.2 (for CUDA 11.5.0)
│ - CUTENSOR: 1.4.0 (for CUDA 11.5.0)
│
│ Toolchain:
│ - Julia: 1.7.3
│ - LLVM: 12.0.1
│ - PTX ISA support: 3.2, 4.0, 4.1, 4.2, 4.3, 5.0, 6.0, 6.1, 6.3, 6.4, 6.5, 7.0
│ - Device capability support: sm_35, sm_37, sm_50, sm_52, sm_53, sm_60, sm_61, sm_62, sm_70, sm_72, sm_75, sm_80
│
│ 1 device:
└   0: Xavier (sm_72, 10.727 GiB / 14.558 GiB available)

I ran test CUDA, and about half of the tests failed. When I run julia runtests.jl cutensor (from the CUDA.jl package), I get the following error:

ERROR: LoadError: UndefVarError: has_cutensornet not defined

Do you have any tips on how I can approach troubleshooting this?

Thank you!

2

Please don’t immediately double-post like that, you asked the same on Slack. Copying my comment from there: which version of CUDA.jl are you using? IIRC I fixed that recently. Try the master branch if the latest version doesn’t work. This error is also related to the CUQUANTUM library, not CUTENSOR, which should still work.