On the Orin Nano 4G device, running the command ./bin/trtexec --onnx=./data/resnet50/ResNet50.onnx under the /usr/src/tensorrt/ path results in an error: Cuda failure: operation not supported. However, it runs normally on the Orin Nano 8G device.
ERROR_INFO.txt (4.5 KB)
Hi,
We are testing this command internally.
Will provide more info to you later.
Thanks.
Hi,
We test the command on Orin Nano 4GB with JetPack 6.1.
It can work as expected.
...
[01/16/2025-11:43:53] [I] === Performance summary ===
[01/16/2025-11:43:53] [I] Throughput: 105.714 qps
[01/16/2025-11:43:53] [I] Latency: min = 9.47241 ms, max = 9.5658 ms, mean = 9.49409 ms, median = 9.49329 ms, percentile(90%) = 9.50439 ms, percentile(95%) = 9.50806 ms, percentile(99%) = 9.52295 ms
[01/16/2025-11:43:53] [I] Enqueue Time: min = 0.488647 ms, max = 0.582321 ms, mean = 0.521172 ms, median = 0.520767 ms, percentile(90%) = 0.548218 ms, percentile(95%) = 0.554077 ms, percentile(99%) = 0.571472 ms
[01/16/2025-11:43:53] [I] H2D Latency: min = 0.0554199 ms, max = 0.0796509 ms, mean = 0.0597083 ms, median = 0.0594482 ms, percentile(90%) = 0.0620117 ms, percentile(95%) = 0.0629883 ms, percentile(99%) = 0.0686646 ms
[01/16/2025-11:43:53] [I] GPU Compute Time: min = 9.40552 ms, max = 9.49744 ms, mean = 9.42601 ms, median = 9.4249 ms, percentile(90%) = 9.43604 ms, percentile(95%) = 9.43811 ms, percentile(99%) = 9.45435 ms
[01/16/2025-11:43:53] [I] D2H Latency: min = 0.00561523 ms, max = 0.0106201 ms, mean = 0.00836225 ms, median = 0.00842285 ms, percentile(90%) = 0.00952148 ms, percentile(95%) = 0.00994873 ms, percentile(99%) = 0.0102539 ms
[01/16/2025-11:43:53] [I] Total Host Walltime: 3.02704 s
[01/16/2025-11:43:53] [I] Total GPU Compute Time: 3.01632 s
[01/16/2025-11:43:53] [I] Explanations of the performance metrics are printed in the verbose logs.
[01/16/2025-11:43:53] [I]
&&&& PASSED TensorRT.trtexec [TensorRT v100300] # ./bin/trtexec --onnx=./data/resnet50/ResNet50.onnx
Thanks.
Hi,AastaLLL
The system on my device is JetPack 6.0
Hi,
Based on your log, you are using TensorRT 10.3.
Could you try to upgrade the environment to JetPack 6.1 which by default supports TensorRT 10.3?
Thanks.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.