AI Model Optimization & GPU Recognition Issues – Win11, CUDA 12.8, ZED SDK 5.0.7

[Andrej]

Hi everyone,

I’m encountering two related issues with the ZED SDK 5.0.7 on Windows 11 (CUDA 12.8, Pascal GPU):

1. AI Model Optimization Failure

   • Initial diagnostics pass, but the model optimization step fails.

   • Attached: Error logs from the optimization process.

2. GPU Recognition Issues

   • Some SDK examples fail to recognize GPU support, despite my GPU (Pascal) being within the supported range.

   • Attached: Error logs from example runs.

System Details:

• OS: Windows 11 (build: 26200.6584)

• CUDA: 12.8

• ZED SDK: 5.0.7

• GPU: Titan xp

Questions:

• Has anyone resolved similar issues with model optimization or GPU recognition in this setup?

• Are there known compatibility issues with CUDA 12.8 and ZED SDK 5.0.7?

• Any recommended troubleshooting steps or workarounds?

Thanks in advance for your guidance!

Attachments:
optimization_error image
example_run_error image

[Myzhar]

Hi @Andrej
Welcome to the Stereolabs community.

Please run ZED Diagnostic and post the full content of the report that it generates.
Have you installed the ZED SDK for TensorRT 10 or TensorRT 8?

[Andrej]

Hi Myzhar
Thank you for your reply.
I did download the sdk for cuda12.8 trt 10..9

Do I have to use sdk for TRT8?
from Downloads page:
” TensorRT 8 (legacy)

Pascal GPUs must use TensorRT 8

[Myzhar]

@Andrej

Do I have to use sdk for TRT8?

Yes, you are indeed using a GPU with Pascal architecture.

[Andrej]

after installing zed sdk5.0.7 win10_11_cuda_12.1_trt8.6 on win 11
consequently, downgrading the cuda to 12.1 to match the sdk
given cuda 12.1 comes with driver version 531 the zed sdk won’t work saying all good but driver needs updated to latest. which is misleading
obviously cuda 12.1 doesn’t support latest driver neither 5.0.7 sdk listed for cuda 12.1 supports the driver that cuda 12.1 supports
so, after upgrade let say we got cuda 12.6 with driver 560
now again the sdk doesn’t recogfnize the GPU as the cuda version mismatches what declared for the zed -12.1 and what got with updating the cuda 12.1 driver to get a newer driver
let me know which exactly version of cuda to install which version of the driver that should work with PAscal please
from that version of your choice, I can get diagnostics log gathered
one more question - zed x zed streaming only allows a few fps is there a way to get it to stream with more fps? although depth viewer won’t run the zed_reciever does somehow

[Andrej]

with that one zedfu says no GPU thus perhaps because now cuda version shifted due to driver upgrade
there is no model optimization error - it got addressed.

but zedfu or depth viewer won’t run perhaps because they cannot get that combination of cuda +driver supported with that version of zed sdk

[KEIGO]

Hi !

I’m experiencing a similar issue when trying to optimize the AI model using ZED Diagnostic.

My setup is a bit different, but the same error occurs during the optimization process:

• GPU: RTX 5060 Ti (Driver v581.57)

• VRAM: 16GB

• CUDA: 12.8

• ZED SDK: 5.0.7 (CUDA 12.8 & TensorRT 10.9)

• Windows 11

The optimization fails every time, and I’ve already tried reinstalling the SDK and using different CUDA versions, but no luck.

[Myzhar]

Please send the Diagnostic report

[Andrej]

@Myzhar Thank you for following up
Could you be more specific, please?
From which exactly combination of Zed sdk version + cuda version + driver version to gather the Diagnostic ?

[Myzhar]

The current configuration that you are testing… or every configuration that you would like to be analyzed.

[Andrej]

Hi @Myzhar thank you for following up
I gave up testing and deleted zed sdk; and installed latest cuda and drivers not supported by zed sdk
If you let me know which configuration exactly is supported, I can deploy specified by you to run the Diagnostics
How about the other thing - why FPS is so low 1-2-3-4-5 on GMSL zed X network streaming? Is it possible to get more fps processed over the network?

[Andrej]

@KEIGO you may like to share DIAGNOSTICS file saved from your configuration for troubleshooting