Dear all:
I am trying to execute deformetrica on a Tesla V100. I can execute it on a P100 without any issues.
If I run deformetrica on a V100 (default settings; test case examples/registration/image/3d/brains) deformetrica exits after 1 iteration with the message:
CONVERGENCE: NORM_OF_PROJECTED_GRADIENT_<=_PGTOL
(optimizer: L-BFGS). If I switch to gradient ascent I get a line search error.
If I run the same problem on a P100 the solver converges / runs until max iter (more precisely, I get the output: ">> STOP: TOTAL NO. of ITERATIONS EXCEEDS LIMIT").
I use CUDA/9.2.88 in combination with python/3.6. I compile using the default options (python setup.py build; python setup.py install --prefix PATH).
Is this a SciPy issue? Do you have any suggestions?
Let me know if you need additional specifications.
Thanks
Andreas
Output of nvidia-smi:
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.48 Driver Version: 410.48 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla V100-PCIE... On | 00000000:0D:00.0 Off | 0 |
| N/A 25C P0 25W / 250W | 0MiB / 16130MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+