The patches to enable it have just been committed and an updated driver is expected to be available for Windows tomorrow.
How to know which devices are supported? For example integrated GPUs
yes, that would be interesting.
rdna and rdna2 are supported but i have no clue what graphics cards or apus use that. amd’s naming scheme is the most confusing i have ever seen.
RDNA and RDNA2 are the names of the GPU microarchitecures https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units
Navi__ are the names of the specific dies / designs within those architectures.
RX5xxx / RX6xxx are GPU models you can buy that contains specific dies from specific microarch.
If you dig around Nvidia and Intel you will find that they have similar naming convention, so nothing to be outraged about. Just for an example: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
RDNA and newer means everything from RX5000 series and newer will be supported.
Also devs are investigating if older GPUs can be supported too.
@YAFU APUs naming convention is more convoluted because of mobile market, and I’m not following it, but usually product descriptions should contain what architecture integrated GPU have - for AMD things like “Vega” or “Navi” for example. Currently Vega is not officially supported.
Ok, I have seen the kernel compiler instructions for Blender and these are the supported architectures:
And according to this:
This includes GCN GFX10 (RDNA 1) and GCN GFX10 (RDNA 2)
At the moment this is bad news for users of old AMD cards and APUs, but I am not sure if they will be supported in the future. (Edit: See the post below this, probably more cards will be supported in the future)
@bsavery , Can you confirm if what I say here is correct?
covers this. The driver team committed to support only RDNA2 cards (6X00 series). We’re working with blender to test and enable other cards as well.
We’re basically working backwards for obvious reasons. Hopefully by 3.0 release more will be enabled.
can we hope for vega support or is this unrealistic? it seems like most apus still use vega?
Those igpu is too weak and won’t be much faster than CPU rendering if at all.
I remember running 1700x against rx470 and the CPU been way faster.
yes, it’s weak but maybe could help the cpu a bit.
With Luxrender my RX480 (Polaris10) appears to be considerably faster than just with the CPU (Intel i9-11900), so I expect the same could happen with Cycles HIP if/when it will be supported. Rendering with both the CPU and the GPU at the same time can also be useful (albeit not necessarily energy-efficient) for decreasing render times even if both devices have the same render speeds.
Are there any commits that show what devices are enabled / disabled? I have a Vega 64 and it would be great to use it!
You see my post above. Apparently Vega 64 is gfx900 architecture and it is not supported yet.
i feel pretty burned having invested in a vega64 that has has been buggy and lacking support the majority of my ownership of it
You said it. It has been my worst MAC experience since I started with the Mac Plus.
Don’t count your Vega out just yet. Pablo said, during the livestream, that they are going to work backwards, starting from 6000 series and move to 5000 and so on. So, fingers crossed, given some time, the 500 series and the Vegas will receive support.
Anyone got any preliminary benchmarks?
One of the blender devs posted this:
(Basically in line what we see but I can’t make specific performance claims).
Interesting, hes using a 5500XT, which isnt supported…
I can confirm, that HIP in Blender 3.0 Beta works with my Radeon 5600XT ( Gigabyte ). It’s 2x faster out of the box. I’ve noticed a bug in translucent material, however ( lamp shade )
I’m using SID denoiser in both versions. 800 samples. 2.93 - 10 min ( tile size 240x240 ), 3.0 Beta - 5 min ( full frame ).