Trying to use a CUDA GPU.

[3suns-org]

As the subject says, trying to use a CUDA GPU. Logs say its enabled, but CPU usage changed by none and GPU is minimally used. CPU is rather high hence trying to offload. Python seems to be my biggest offender. Installed CUDA Toolkit 11.8 with Drivers for a Geforce M2000 card. Any ideas?

[MikeLud]

The issue is your GPU is to old and CUDA 11.8 does not support it,. You need a GPU with CUDA compute capability 3.5+, yours is 2.1.

https://en.wikipedia.org/wiki/CUDA#:~:text=*%20b)-,GPUs%20supported,-%5Bedit%5D

[3suns-org]

Okay I am now trying an M2000. Same behavior. Processing time is generallly between 500 and 1200 ms. Gpu seems to be mostly unused showing 0.30GB in use and 2%. Cpu continues to spike to 100% due to python.

[MikeLud]

Try uninstalling the module then reinstall the module using Do not use download cache

[3suns-org]

I have done so with no change.

20:34:18:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...abfcaa) ['No objects found'] took 2101ms
20:34:18:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...6d44df) ['Found car, bench'] took 2091ms
20:34:18:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...010f7b) ['No objects found'] took 2124ms
20:34:18:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...2f2229) ['No objects found'] took 2204ms
20:34:18:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...5d68c4) ['No objects found'] took 636ms
20:34:19:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...2dcae6) ['No objects found'] took 693ms
20:34:19:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...d43d6c) ['No objects found'] took 639ms

[MikeLud]

Try disabling Half Precision

[3suns-org]

Seems to have not had an effect. Stats are similar 07:06:48:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...9734de) ['Found truck, bench'] took 651ms 07:06:49:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...7e74fe) ['Found car'] took 856ms 07:06:49:Response rec'd from Object Detection (YOLOv5 6.2) command 'detect' (...db64f5) ['Found bench'] took 681ms

…

________________________________ From: MikeLud ***@***.***> Sent: Wednesday, April 9, 2025 9:19 PM To: codeproject/CodeProject.AI-Server ***@***.***> Cc: Cory Mckee ***@***.***>; Author ***@***.***> Subject: Re: [codeproject/CodeProject.AI-Server] Trying to use a CUDA GPU. (Discussion #322) Try disabling Half Precision image.png (view on web)<https://github.com/user-attachments/assets/60f8048e-9359-45c7-ad15-68a8e72592ef> — Reply to this email directly, view it on GitHub<#322 (reply in thread)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQ6PTEWEISAZIDOGCPOUR3D2YXWM3AVCNFSM6AAAAAB2K65426VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTENZYGYZTANQ>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[3suns-org]

I tried installing Yolo V8 and response times look great (100 MS or less), but I get an error when I try to download any of the models.

I also see that v8 shows GPU(CUDA) and 6.2 shows CPU.

[3suns-org]

It looks like I can only get the GPU to pick up for Object Detection but not for the other options, even though I have selected use GPU