Vulkan-NCNN Performing better on CPU than on IMG-GPU

RISC-V Ecosystem
1

Hello Team,

I’m having a RISC-V Dev platform which has a IMG-GPU, and I’m able to successfully build the Vulkan-NCNN Framework and performing Yolo object detection what we noticed is we’re observing better with CPU as it’s taking around ~3.0 seconds for performing Yolo v8 objection detection, while GPU is taking ~6.2 seconds for performing the same object detection.

Can any of you let me know if anyone around here has already done something on this on any of the VF2 or other RISC-V Platforms?, If there’s a way to optimise or improvise the performance rates on GPU over CPU. Also it’ll be highly appreciated if you can help me with sharing more details on this issue.

Regards,
Ravi Kiran

1 Like
2

I got the same result with llama.cpp, CPU is faster than GPU on VF2.

1 Like
3

Hello @cwt, Thanks for you’re reply.

Do you have any similar data as how much time difference you’re noticing on CPU over GPU.? Also can you let me know if are seeing similar behaviour for Yolo Applications?

Thanks…

4

Ah, I cannot help you on that, somehow the latest change on llama.cpp make it stop working on VF2, and I never try Yolo on VF2.