ROCm 4.0 on RX 580 - Time to change your card?
by Mathieu Poliquin
Rocm 4.0 arrived a couple of weeks ago and it seems AMD stopped officialy working on supporting Polaris cards such as the RX 580 althought they haven’t removed them yet.
So I was curious if my RX 580 was still working and it seems it still does, at least the resnet-50 benchmark and the bandwidth test.
It’s unfortunate they will stop supporting the RX 580 as it’s one of the best performance/price cards out there and much cheaper than the other NVIDIA 8GB cards of the same generation. Probably AMD wants to focus their man power on the new architectures.
Moreover GPU prices are at sky high prices these days because of high demand by Cryptominers and a shortage of supply by manifacturers.
So I would say it’s the right time to sell your RX 580 if you have a replacement meanwhile GPU prices go back down. I plan to sell mine and depending on the state of ROCm in the futre I will go for one of AMD’s 16GB cards and/or the RTX 2060 super / RTX 3060ti, Nvidia’s best bang for buck ML cards.
- RX 580
- i3 4130 @ 3.4Ghz 2 cores / 4 threads
- 8GB DDR3
- Ubuntu 20.04
- tensorflow-rocm 2.4
- python 3.8
Installing is same as with previous versions except you need to add your user to the render group if you are using Ubuntu 20.04
Details can be found here:
tags: AMD - Rocm - RX580 - tensorflow - machine learning - review