Try this to install ROCm v 4.something. Use Kernel up to 5.4 to be safe (5.4 started in 2020)
forum of DEBIAN TEAM WORKING ON ROCM PACKAGING
AMD's repo of Pytorch. This is the branch I succesfully built.
GeForce RTX 3060 12GB is generally minimum recommended for AI inference and some training
NON-passmark comparisons of GPU. Vega 8 is better than Vega 7.
SHOWS EXACTS SPECS OF CPU OR APU INCLUDING GRAPHICS
2x16GB $89, or $80 for open box. This is close to my current memory, but is Ripjaws vs my TridentZ
$104 for 2x16G. Open box for $94. This is the closes to my current memory.
$249, at Canadacomputers, they also have open box for $224. BUY THIS, NOT THE AMD Ryzen 5 5600GT(VEGA7), BECAUSE THIS HAS VEGA8, FASTER THAN VEGA7, AND SAME AS VEGA8 ON LAPTOP 2500U.
Compare Distinct graphics cards with AMD APUs. Result: Buy Ryzen 5700G, or AMD Ryzen 5 5600GT as GPU is the same. BUT Ryzen 5700G has 25% better CPU
Buy the Ryzen 5 5600GT (6 graphics cores, $189CDN) but also consider AMD Ryzen 7 5700G (8 graphics cores, now on sale $245CDN)
$189 in Canadacomputers
Install Pytorch on ROCm from AMD git repo. TRY THIS!!!
Describes flags that build PyTorch locally - maybe I should try that!!
This guy claims he got both gfx900 and gfx902 working. I asked to try my test.
Another attempt failed for me.
Database of AMD GPUs. Here we can see that 2500U is Vega 8 mobile , which is GCN 5.0. This is only supported in ROCm up to 4.5.2!!
Rusticl seems equivalent of rocm.
This site seems to claim there is a known fix. Looks like someone is fixing it???
Installing PyTorch - this is a less interesting document, but it does show validation steps