Is anyone else noticing very fast adaptive subdiv performance + shader displacement with last nights build? Its blowing my mind right now. Instantaneous BVH building on render too. AND multi GPU working. This is moving faster than I anticipated!
4 Likes
For me BVH building was long gone with cycles X. thats why i used it from first released day coz it helped alot on renders.
1 Like
I have two GTX 1070Ti
I tested this and I do not see any time difference - how do you know that both GPUs are being used again - there are no tiles to visually see it and as I said the render time is the same compared to an older build ?
1 Like
ah fantastic !
1 Like
Where did you get your build from? None of the build on https://builder.blender.org/download/experimental/ have the multi-device changes in yet so I wouldnāt expect any change.
2 Likes
I just looked at Task Managerā¦ it must have been misreporting or I misinterpreted it somehow. Whoops! Only means faster times ahead tho!
@giacometti777 hopefully - for me since some time I get only cuda erros with both GPUs on.
So i stoped using both - but I am hopeful thats this will work again.
1 Like
Sergey was working on it in the developer stream yesterday, if IĀ“m not mistaken. Did not watch it myself, but it seems to be on the way for initial implementation atleast https://www.youtube.com/watch?v=atPY1Qgw--s
2 Likes
Oh wow they video record therm coding and stream it ?
1 Like
Hehe, Yeah, wellā¦, Sergey started streaming from time to time before the blender today show on mondays a while ago, not sure if it is gonna be a regular thing yet, I guess it could be a good resource for people wanting to start programing for Blender.
3 Likes
Thatās quieter fascinating to see providing a cool opportunity look behind the curtain
4 Likes
Multi device rendering seems to be implemented, but Iām not sure that working is the term. I tested on three of my own scenes, exactly as loaded, using e3ca9ccc98af (20 June) against cf028a8f6806 (30 June) as testing platforms.
The results I got were (old version first):
Scene 1 04:11.06 02:19.66
Scene 2 5:53.76 2:55.89
Scene 3* 0:57.79 0:45.88
Scene 3 is an odd one. It uses an awful lot of adaptive subdivision, and so is very memory hungry, pushing the limits of my GPUs. The above timings were at 1080. Rendering at 1440 either fails (out of memory) or, strangely, takes half as long again on multi GPU.
I also found that GPU/CPU (whether using both GPUs with the CPU, or just one) was painfully slow, so much that I didnāt have the patience to finish a render.
Dual GTX1070 8Gb/ Ryzen 1800X.
So the new experimental build has it in ?
From: https://builder.blender.org/download/experimental/
if you check the build hash number (underlined above) and compare it to the commit logs: https://developer.blender.org/diffusion/B/history/cycles-x/
You can see that the current daily experimental build is fully up to date with all the commits in the cycles-x branch.
2 Likes
Ah thatās very good to know - thank you for the explanation
1 Like
Iām sorry if this was already asked, but I tried some renders with Cycles Xā¦ and while itās incredible faster, specially on heavy scenes, on my end, it produces a lot more noise with the same settings, specially on the grass.
Iāve even tried the following settings:
*Increasing indirect clamp, and glossy filter, with no improvements. *
On both renders, caustics are disabled.
Here is a comparasion:
Cycles X render:
Cycles render:
Any thoughts? Thanks!
2 Likes
The main difference I can see is that the lighting isnāt the same in both images.
For the Cycles render, you appear to have additional light hitting the columns of the gate (the lighting pattern on the column is different) and there are more highlights on the sign above it too and some of the rocks in the background have different shadows. All of this suggests to me that the lighting in the two scenes isnāt quite identical.
On the Cycles-X render, the candles look different too. It looks like there is more/brighter sub surface scattering going on in the wax close to the flame.
I wouldnāt spend too much time stressing over the details at this stage. Cycles-X is still at pre-alpha phase of development and not everything is working as expected (if at all). I wouldnāt expect to be able to run a true and fair comparison between Cycles and Cycles-X image quality until we are into Beta at least.
To be honest, absolutely no changes on the light were made. Iāve just opened the same scene and rendered.
Load both images in Compositor and mix node set to Difference and you will see what the observation is about - even if no changes made, there is significant difference.
1 Like