Cuda Error in findFastestTactic: 700 when using tensorrt in fp16 mode on Xavier NX with Jetpack4.4


When I converted an onnx-model to trt-model in fp16 model on Xavier NX with Jetpack4.4, the cuda error came out like:
[06/05/2020-17:10:15] [E] [TRT] …/builder/cudnnBuilderUtils.cpp (423) - Cuda Error in findFastestTactic: 700 (an illegal memory access was encountered)
[06/05/2020-17:10:15] [E] [TRT] …/rtSafe/safeRuntime.cpp (32) - Cuda Error in free: 700 (an illegal memory access was encountered)
terminate called after throwing an instance of ‘nvinfer1::CudaError’
what(): std::exception

But I can do it in fp16 mode on TX2 with Jetpack4.4 successfully.


ONNX IR version: 0.0.3
JetPack Version: 4.4
GPU Type: Jetson Xavier

Moving to Jetson Xavier NX forum so that Jetson team can take a look.



Thanks for reporting this to us.
Do you have an onnx model so we can reproduce this in our environment?


The network consists of two identical parts. If I only convert the first part on in fp16 model on Xavier NX with Jetpack4.4, it can be successfully converted. But the whole network fails.


We can reproduce this issue in our environment and will pass it to our internal team.
Just in case you don’t know, the float32 version can be inferenced without issues.

$ /usr/src/tensorrt/bin/trtexec --onnx=./checkpoints_0.9_ghost_320_780_part_v5.onnx


Thank you for your reply. I know that the float32 version is normal.
Waiting for your good news.


This issue is fixed in our internal TensorRT package.
We will let you know once it is available for the public.



Hi, is the JetPack 4.4 - L4T R32.4.3 production for NX the new version that the issue is fixed?

No, it’s not fixed at this version, but in the coming release,

OK, thanks.

I am facing the same problem.
the newest TensorRT is 7.2. Was it fixed ?

I am too facing the same error. Has it been fixed in 7.2 release?


Please noted that the TensorRT package in JetPack4.4.1 is v7.1.3.
You will need to wait for our next major release for the fix.

Sorry for any inconvenience.

As of today i was notified about JetPack 4.5

This version is including amongst others TensorRT 7.1.3 - it seems there is nothing new with that!
When going for the documentation of TensorRT it will as of now report versions up to 7.2.2.
Not sure if any relevant fix has made it being part that series of releases.

see this web pages for more details:

Nothing changed so far. I had this issue and it still remains. I guess we have to wait for version 7.2+.

1 Like

Dear Nvidia Team,

I am using float 32 bit mode in order to fix this issue as short term solution. When we can expect bug fixed version of TensorRT with the support of float 16 mode?


We will update to new version in next JetPack release. Thanks