2D class average run error

If the number of classes in my 2D class average run was anything more than 128 classes, at the start of iteration 0, I got the following error message:

Traceback (most recent call last):
File “/home/cryosparc/cryosparc/cryosparc-compute/sparcjob.py”, line 359, in
cuda_devs = current_experiment_params[‘cuda_devs’])
File “/home/cryosparc/cryosparc/cryosparc-compute/class2D.py”, line 270, in abinit
num_threads_per_dev = 2)
File “/home/cryosparc/cryosparc/cryosparc-compute/engine/engine.py”, line 759, in process
models_cudaarray = cuda_core.transfer_ndarray_to_cudaarray(all_models)
File “/home/cryosparc/cryosparc/cryosparc-compute/engine/cuda_core.py”, line 162, in transfer_ndarray_to_cudaarray
devarr = cudrv.Array(desc)
LogicError: cuArrayCreate failed: invalid argument

Any number of classes, 128 and below, did not produce an error message. I tested numerous class numbers above 128 until I reach 200 classes. All produced the same error. Thoughts?


Hi @Cosmo, can you confirm that this is still happening in v0.5.3? We had changed the GPU memory layout to handle many more 2D classes.