I have GPU == GeForce GT 730 with compute capability 3.5 ,
I have Nvidia Driver >= 471
I have CUDA == 11.4
It is still not able to run yolov8 model on my system…
Please give me direction to solve this issue.?
Did you compile the software yourself? If so, ensure that is is built for GPU target architecture sm_35.
If you are using a pre-built binary, it may have been built without support for compute capability 3.5. This may have been done on purpose, check the vendor’s stated minimal hardware requirements.
Side remark: The GeForce GT 730 has 1GB of on-board memory per the TechPowerUp database. This is unlikely to make YOLO v8 happy; most GPU-accelerated machine learning frameworks require more memory than that. Again, check the vendor’s stated minimal hardware requirements.
I actually used pytorch versioon:—>
torch == 1.7.1+cu110
tochvision ==0.8.2+cu110
ultralytics == 8.0.124
I am not familiar with your software stack. You would want to ask for help on the support sites of the vendors of the various components in the stack.
CUDA uses fat binaries. When building CUDA software, the programmer can specify for which target architectures machine code should be embedded and for which virtual architectures PTX (an intermediate code representation) should be embedded. When running a CUDA program, the CUDA runtime first looks for machine code that matches the GPU the program needs to run on (sm_35 in your case). If it cannot find macting machine code, it looks for relevant PTX code that it can JIT compile for this GPU. If that is not found either, it signals the error you encountered.
From this we know that somewhere in your software stack there is some CUDA code which has been compiled to a binary without PTX and without machine code for sm_35 (compute capability 3.5). Your task now is to find that component and build it for sm_35 if possible.
As I said, the vendor of whatever CUDA-accelerated component that is may have a minimum GPU architecture that they require, for example sm_50. If that is the case, you will need to find hardware that satisfies that requirement.
Ok , but I am really stuck here.
compute capability 3.5 devices haven’t been supported by pytorch for some time now. There are many questions like this on various forums, you can get an idea what the options are from those. If you still need help, you may get better help by asking or researching on a forum dedicated to pytorch. There probably aren’t many pytorch experts here. For example, here is a relevant pytorch forum topic.