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 ).
.
3 Likes
That’s a nice boost in performance. How about the viewport rendering, its responsiveness, speed?
It’s noticeably faster.
Nice. Looking forward to the full release.
I rendered some more tests. Agent 327 didn’t open well in 3.0 ( gemotery “exploded” everywhere ). Cat render is wierd and it shows intel i7 processor - my rig is Ryzen 1800x 
Fox and one of Party Tugs were rendered with Eevee.
Here are some images:
Windows 10 Home 21H1
Gigabyte Radeon 5600XT Gaming OC 6GB Windforce
AMD Ryzen 7 1800x ( not overclocked )
16GB Ram
ADATA SX8200 500GB Nvme SSD
GPU settings:
Is it possible to render using both GPU and CPU? If so, what is the “tile” situation?
It is possible to check both:
But it is MUCH slower then. With tile size set to 2048 ( full frame ) it took 6:30 min to render. GPU was not working on it’s full potential ( based on stats view in AMD utility ).
With tile size set to 32 it rendered with just 1 tile ( I didn’t want to wait for that ). CPU only takes about 9 minutes. Still better than 2.93. I believe we should wait for full HIP support in a release version of Blender. Also, i think Cycles X doesn’t like small tiles too much 
I believe devs are still trying to deprecate the tile system. In the Blog they talked about removing it, but there’s no news from that front AFAIK.
@psvedas @liquidstudio @stargeizer
You shouldn’t suppose to change tile size in 3.0. Tiles were repurposed to render very high-res images like 4k,8k or 16k and bigger. They aren’t used for performance tuning anymore. You can leave tile size at 2048. In new Cycles all devices are cooperating on one tile. Basically unless you are doing hi-res images you can forget about them.
You can report bad performance with AMD GPU+CPU rendering as a regression. I suppose it wasn’t tested well.
There will still be tiles for when you need to render a massive image like 8K or something.
Been just rendering 0 tilesize because its the fastest, doubt that has changed
I see. Well, it is nice to know the feature is available and has the potential to improve.
Are multi-GPU configurations supported?
Interested to know if it will use GPU/CPU/iGPU at the same time. That would be awesome for laptop users.
1 Like
Hybrid CPU/GPU rendering makes no sense. It is a waste of developers time.
Radeon RX 6800XT close to RTX 3080/3090:
https://www.reddit.com/r/radeon/comments/r0l2ts/rdnardna2_gpus_get_massive_performance_gains_in/
2 Likes
I wonder if HIP takes advantage of AMDs Ray tracing features, or is it just brute force? In any case, great result matching the 3080 at a substantially reduced power consumption and an extra 6 gigs of ram.
Hardware raytracing is not supported yet. It’s on the To-Do list:
https://developer.blender.org/T91571
1 Like
Wow. So with it, we can expect the AMD cards to comfortably outperform their Nvidia counterparts.