Open fostiropoulos opened 11 months ago
Having tried the same code in different machine with different GPUs the error seems to be related to the GPU model. The code that produces a bug is from RTX 2080 with nvlink, while the same code does not produce an error for V100.
Thanks. This is due to an inadvertent ABI break in the 535 driver, which will fixed in the next patch release.
Thanks for clarifying. Should there be a check somewhere and an error raised for this particular version or documentation of it?
@wence- Patch release for what? nvidia? cuda? Or can the pynvml package be updated to fix this?
this may be the same problem as my issue here: #50
@erikhuck how I solved it was to uninstall driver version 535 and install the preceding release until the newer version is released.
| NVIDIA-SMI 525.125.06 Driver Version: 525.125.06 CUDA Version: 12.0 |
For my current system
The processes running on cuda are read incorrectly. There is an additional size_t element when reading the strut that is not documented but when ignored causes the response to be malformed.
The following code illustrates the error:
Output:
Expected: