SUPPORT / SAMPLES & SAS NOTES
Problem Note 69549: Training a SAS U-Net model with one or more GPUs might result in an error
 |  |  |  |  |
When you train a SAS U-Net model with one ore more graphics processing units (GPUs) and the image height is not equal to the width, the following error message might occur:
ERROR: CUDA: an illegal memory access was encountered
ERROR: The action stopped due to errors.
ERROR: The action stopped due to errors.
Here is an example that shows the DLPy fit function where the width and height are different and gpu= is specified:
model = UNet(s,
n_classes=4,
n_channels=3,
width=512, height=256,
bn_after_convolutions=True,
output_image_type='BASE64') # 'BASE64', 'PNG', 'WIDE' (default)
model.fit(data='data_train',
valid_table='data_valid',
optimizer=optimizer,
data_specs=dataspecs,
n_threads=4,
force_equal_padding=False,
gpu=Gpu(devices=[0, 1]))
valid_table='data_valid',
optimizer=optimizer,
data_specs=dataspecs,
n_threads=4,
force_equal_padding=False,
gpu=Gpu(devices=[0, 1]))
Click the Hot Fix tab in this note for a link to instructions about accessing and applying the software update. The hot fix is for Linux x64 only.
Operating System and Release Information
| Product Family | Product | System | Product Release | SAS Release | ||
| Reported | Fixed* | Reported | Fixed* | |||
| SAS System | SAS Visual Data Mining and Machine Learning | Linux for x64 | 8.5 | Viya | ||
| Microsoft® Windows® for x64 | 8.5 | Viya | ||||