Michal Guzek comments

Results 93 comments of


                                            Michal Guzek

TensorRT 10.3 is 3+ times slower than p ytorch when running inference on Gpus A30 and 4090

@CallmeZhangChenchen , sorry for the late follow up - is this on Windows 10 or 11? If not, could you provide a specific OS version for us to reproduce?

TensorRT 10.3 is 3+ times slower than p ytorch when running inference on Gpus A30 and 4090

I've instanced an internal bug, thank you.

TensorRT 10.3 is 3+ times slower than p ytorch when running inference on Gpus A30 and 4090

@CallmeZhangChenchen , could you provide pytorch inference script as well? The issue is about comparison with pytorch, could be that TRT has bug or could be that the pytorch script...

TensorRT 10.3 is 3+ times slower than p ytorch when running inference on Gpus A30 and 4090

Thank you, I'll pass the info on

TensorRT 10.3 is 3+ times slower than p ytorch when running inference on Gpus A30 and 4090

@CallmeZhangChenchen the slowness here comes from the "CumSum" op, which is known to be super slow. We plan to fix this issue in TRT 10.6.

Where is wheel file of TensorRT python bindings for Jetpack 4.4 and TensorRT 7.x?

Can you try this way: https://forums.developer.nvidia.com/t/how-to-install-nvidia-tensorrt/250891

[Issue still exists in TRT 10.1] Fail to build the engine if input tensor number > 2^31

Closing in favor of https://github.com/NVIDIA/TensorRT/issues/4123

DCHECK(mem_span <= data_->size()) failed. mem_span: 2559986, data_->size()320000

I will follow up with people from Drive OS once they're back next week.

DCHECK(mem_span <= data_->size()) failed. mem_span: 2559986, data_->size()320000

Apologies for the delay, I was told to instance an internal bug for this to reproduce. Can you share `./trt_bev/bevf-864-1536-200-200_trt_plugin_fixed_20240919_prev_bev_opt_15.onnx` ?

[Issue still exists in TRT 10.4] Fail to build the engine if input tensor number > 2^31

@dzzhang96 can you provide your OS and the exact command/snippet you use to build the engine? I'll instance an internal bug at that point, thanks.