Do you know if the patches also make the combined Vram via NVLink possible? Do you have a NVLink setup to test this?
And thx for your awesome benchmarks! <3
combined Vram via NVLink possible
I’d like it too but don’t hold your breath about that, it’s quite an advance feature. (i have one and no it was not working 
)
first they need to adapt the last cycles features (baking) and correct all the crashes (i had a lot of errors)
then they need to finish the AI viewport denoiser as promise in the blender dev post.
i don’t know if it is a special nvidia dev who was working on the integration or an official blender dev. In both cases i don’t think they will waist time on such niche feature. The publicity stunt is done for nvidia and blender devs have more important things to do.
we may need to pay a dev ourselves to do it. the market is too niche for such a feature.
1 Like
questions for curiosity:
eCycles, with these patches for RTX would have advantages, or are they replacement?
can anyone make speed comparisons between eCycles and Cycles + RTX?
1 Like
You know how to tease, dont’ you bsavery 
Still SIGGRAPH sill going, but haven’t noticed any solution that won’t lock me into hardware. But still few days to go, fingers crossed.
Your results agree pretty well with mine. Yeah, if a GPU is a pure compute card without driving any monitor can play a role.
That’d be nice. And comparisons between OptiX and CUDA with CPU + GPU.
Tested CUDA vs RTX on an archviz scene
CUDA : 1.53m
RTX: 1.02m
Same with the progressive rendering , feels twice as fast
2 Likes
I got a 1.8x scaling speed for both.
But it’s quite per case dependent
edit: hot stuff
e-cycles + RTX is going to be mad fast. We can hope for a 3x speed scaling.
5 Likes
HOLY ***** !! 
I am a great tease.
But here’s a demo of hybrid raytracing fast interactive viewport in blender:
https://twitter.com/bhsavery/status/1156238317718872065
Oh and it works on any vendors hardware and takes advantage of any hw raytracing acceleration available… #noclosedstandards
7 Likes
really interesting solution !
where can we find infos about it ?
do you know if amd will show their RayTracing hardware some days ?
My optics renders are consistently taking more time than cuda, no matter what the scene, samples or tile size. I’m using an RTX 2080 Ti. Any ideas why this is happening?
1 Like
i’d wait for an official release, its still really buggy. (from my pov at least).
So it will use the CUDA cores and RT core on a RTX card? Maybe it’s just me or the compression on twitter but it doesn’t look that impressive, there still some noise I was expecting something else. There’s also “RedShift RT” I got high hope does this.
I’m getting exactly same render time in cuda and optix. Titan RTX, 256x256 tile size.
ProRender had always terrible update rates. I am kind of surprised they remain terrible even in the low, rasterized mode. There’s no point going all the way down to rasterization if the update rate still stays in single digit FPS. Especially now, during the rise of path tracers running at 60+ FPS while still managing to denoise on top of that
In my experience, and from the other benchmarks I’ve seen posted, 256x256 is the worst size to use. I’d try 64x64 and see if it improves.
2 Likes
Do you mean Prorender also makes use of NVIDIA OptiX for a speed increase?
According to BD3D it is apparently not.
<-> CUDA <->
1x 2080ti 16x16 3:59:80 ++++
1x 2080ti 32x32 3:45.56 ++
1x 2080ti 64x64 3:43.92 + (=1x)
1x 2080ti 128x128 3:45.95 ++
1x 2080ti 256x256 3:52.61 +++
1x 2080ti 512x512 3:59.64 ++++
2x 2080ti 64x64 2:10.41 (2xGpu) (Dual Gpu Scaling =1.7x)
<-> Optix <->
1x 2080ti 16x16 3:59.29 +++++
1x 2080ti 32x32 2:27.27 +++
1x 2080ti 64x64 2:16.13 +++
1x 2080ti 128x128 2:12.73 ++
1x 2080ti 256x256 2:09.82 + (Optix =1.7x Faster)
1x 2080ti 512x512 2:10.22 ++
2x 2080ti 256x256 1:10.78 (2xGpu) (Dual Gpu Scaling =1.8x)
1 Like
i don’t think its universal, it depend on the gpu model and the render engine. for example e-cycles prefer tiny tiles it seems