Itâs not looking good for scaling of Adaptive Sampling! (Which is what I reported to the devs a few days ago).
Itâs taking longer with 2 GPUs than 1 GPU but the result is seems less noisy with 2 than 1 despite the same settings. This was with the Nissan scene.
So yeah⌠it still needs some work.
I also still get tearing on the preview with Adaptive Sampling with 2 GPUs where one âhalfâ of he screen is one noise level and the other half is totally different.
It is still best to have one very fast GPU for this stuff. But for Final Rendering without Adaptive Sampling 2 GPUs or more really does work.
I was saying to be able to use Medium as an output for projects will be a godsend. I use Eevee a lot more than I ever thought I would and with the extra ray tracing of Medium it will be invaluable.
While I have managed to render out for the first time I am having crashes to desktop. I think it may be due to when I apply smooth shading to a reflective sphere which is hovering over a reflective plane it crashes. I will try and make this a repeatable crash and let you know the outcome.
In my extremely crude test scene adaptive sampling works with multi-GPU for sure now and there was an 80-90 render speed improvement.
I hope youâll support the Intel Open Image Denoiser and provide the necessary denoising passes. I appreciate itâs tech from a competitor but your CPU guys are giving them such a whooping and the RPR denoiser is still giving me a black viewport.
Is the Nissan Scene you speak of available to download please?
I only have my own crude test scenes and I wonât be able to share any of my client work I test with RPR so Iâd appreciate a half decent scene to test which I can report bugs with.
Are there any test scene from AMD themselves anywhere?
I canât get the Hybrid modes to render today. A scene with default cube will crash to desktop.
It was working last night briefly then I got a crash after changing the shading on a chrome sphere to smooth and now none of the hybrid modes will render. They work in the viewport fine but during the syncing process I get a crash to desktop. Note. Full Mode renders out fine.
Are there any cache files RPR makes during Syncing that I need to manually clear out?
Any help gratefully received.
Viewport Bug with Full Mode - Direct Illumination
Reflective surface renders reflections black in viewport (left) renders correctly in final render (right).
Thereâs no way posting bugs on a forum will get picked up by the devs.
I canât help with non Mac features. Mac doesnât have the fancy Hybrid mode yet, sadly. So weâre stuck with full render only. Final Renders with RPR on Mac with Metal are slightly quicker than RPR on Windows on the same hardware though so itâs not all bad
Must be a macOS feature!! I think Windows RPR Denoiser uses AMDâs own AI. You can still use Blenderâs built in Intel Denoise the same as Mac but youâll have to connect up the AOVs manually.
Regarding the time on the Nissan that would be 40 seconds faster than anyone else has ever rendered the Nissan scene.
What were your final render settings?
The âbench testâ settings are:
Fastest time 2m08
For a Resolution of 1080P full size.
I didnât change the render settings but I donât think the HDRI was included in the file so I used the RPR Sky default setting. Iâll rerun with 25 samples per view update which should be marginally faster.
Dual 1080TIs. 1.45.97 Total time, 1.33 Render time.
Single 1080TI 2.35.74 Total time, 2.22 Render time.
The HDRI should have been â4DstudioLight_2048x1024.hdrâ which must have come with the file at some point.
Your final render time 1m45 is extremely good and just proves that there is no AMD advantage AT ALL with Pro Render because in all our testing all the NVIDIA GPUs win the testing.
Your dual 1080TI setup is better than any single GPU so far.
And of note is that back in April 2019 the same scene was being rendered:
12/04/2019 RPR2.0.52 GTX 1080Ti GUI Windows 10 Time : 05:17
So if you wanted proof of âscalingâ or performance updates with RPR you have it there and no bias towards AMD hardware whatsoever.
The first time I used ProRender 1.0 was in C4D and it was dreadful, a joke. V2.0 in Blender has been a huge improvement, I choose to use ProRender 2.0 because itâs good not because itâs free and it keeps getting better. I just wish those Hybrid modes would work for me reliably as theyâd make a huge impact in my workflow with certain projects.
What is really interesting in your database is how fast the 5700XT is compared to much more expensive nVidia GPUs around it. Iâm hoping that Big Navi will deliver some serious performance at a fair price. Iâll happily put 2-3 Big Navi GPUs in each of our workstations if AMD can deliver the goods.
Itâll also be interesting to see how much performance ProRender will gain from ray tracing hardware that Big Navi is alleged to have. RTX has made a big difference to Cycles after all.
Pro Render is a young engine so it will naturally take a few versions before it can be considered fully mature.
Where Pro Render is integrated into the app it does seem to work better than where it is a plug-in.
The down side on Blender with RPR is that RPR is unlikely to ever be able to be compatible with every single Cycles node so you are limited to doing full texture work rather than procedurally generated stuff that can work so well with Cycles.
The 5700 XT is all over the place on benchmarks. Sometimes itâs better than a RADEON VII and sometimes itâs worse than an RX 580. In RPR it does appear to do quite well, as does the 5600 XT.
Weâll are all waiting in anticipation of what the rumours will bring. Canât wait.
I just hope AMD can implement some more thorough software testing as they do seem to shoot themselves in the foot with reliable hardware and buggy software here and there.
That is very true. Motion is faster with the RX 580. With RPR it seems only slightly faster than the 5600 RX. I gave back the 5700 RX so I cannot test it. The 5600 XT just runs well and is very silent.
That is indeed an issue when you rely on those nodes.
But I think for many it will also be sufficient - architecture/product renderings.
Optix seriously does have a raging advantage but so I also think will the hybrid mode - something eeVee does not offer.
Speed might also be an issue. RPR is simply compared to cycles and CUDA not at the same level - at least when I tried to do somewhat comparable tests.
AMDs big advantage (great marketing strategy) is that RPR is free and as it seems also highly adapted by many software packages.