Darknet: if i'm using AMD, what should i install so that i can use GPU to train?

[jiaswee1995]

if i'm using AMD, what should i install so that i can use GPU to train?

[ahsan856jalal]

This is maximum , I know from darknet perspective apart from software installation for your graphic card

Tesla V100

ARCH= -gencode arch=compute_70,code=[sm_70,compute_70]

GTX 1080, GTX 1070, GTX 1060, GTX 1050, GTX 1030, Titan Xp, Tesla P40, Tesla P4

ARCH= -gencode arch=compute_61,code=sm_61 -gencode arch=compute_61,code=compute_61

GP100/Tesla P100 ?DGX-1

ARCH= -gencode arch=compute_60,code=sm_60

For Jetson Tx1 uncomment:

ARCH= -gencode arch=compute_51,code=[sm_51,compute_51]

For Jetson Tx2 or Drive-PX2 uncomment:

ARCH= -gencode arch=compute_62,code=[sm_62,compute_62]

[jiaswee1995]

i m sorry, I couldn’t understand lol

[jiaswee1995]

So what I need to do is to install driver for my graphics card?

[TheMikeyR]

You can't use AMD cards with this darknet repo, since AMD don't have CUDA and it's only available for nvidia cards. There are some ports available but I don't know how great they work, e.g. this one https://github.com/ganyc717/Darknet-On-OpenCL

[jiaswee1995]

@TheMikeyR Okay, thx a lot!

[Takaklas]

Workaround: use this
https://github.com/thtrieu/darkflow
it's darknet written in tensorflow and also this
https://github.com/ROCmSoftwarePlatform/tensorflow
which is tensorflow on top of RocM for AMD gpu's (rocm is kinda like cuda for amd cards)
Hope it works for you , I am still having problems installing rocm but keep me informed if you try this
Only downside is probably that it is limited to yolo v2 and tensorflow 1.3 for the moment i'm writing this

[sowson]

My project had some issues but now seams to the most of them have been solved at https://github.com/sowson/darknet it is Darknet in OpenCL... that works! :D. I am looking for contributor who help with pull request that add Windows setup possibility.