The february update it now available. It’s about 5% faster in mean, up to 10% on top of the existing speedup. It also has a new option to reduce noise, allowing to render about 20% faster for the same noise level. The original idea is from solid angle, the implementation from Lukas Stockner, updated and fixed by @Dragon.Studio and me.
Here an example of architecture fly through. 1min54 per frame at full resolution
Original scene from Evermotion. So it works perfectly with animations, even with complex light path through glass.
You can get the 2.8x or 2.7x version until this midnight UTC+1 time at reduced price. If you prefer smaller rates, you can get 2.8x here.
As usual, if you find any bug, you can send me a file to reproduce per Email or PM and I’ll have a look.
i’ve been getting that error a load on my WIn10/GTX970, flat out will not render my scene - it has baked fire stuff in it, which may be a factor. tried optimising the materials etc, oddly enough though it run no prob at all on my Win7/GTX980 machine.
Ok, thanks for the report. Then you can provide a file per PM with OS, graphic card and driver version and I’ll have a look. @wolfie138 maybe it’s indeed the same bug, if you can provide a file, it will help.
Bugs are expected with new feature updates
i’ll try the new release over the weekend, if i still get the same/similar i’ll see if i can narrow anything down (eg, just have a file w/ only the domain in) and let you know.
Seems that is my fault =) I was playing with GPU overclocking and it causes to CUDA bug. I’ve disabled overclock and everything works as should. Thanks, now I have 2x faster renders.
One suggestion I’d have for those animations is that any time you render anything that moves, at all, I mean even if it’s just 2 frames… as long as it’s not a still image, always use motion blur, always
I usually cringe when I see CG animations in motion with motion blur off, because they immediately get that stuttering, fake CG motion look It’s especially common with archviz people. They tend to forget motion blur the most, and then wonder why their animations look so off
By the way, do you use static or moving noise pattern? You probably already know it, but just to make sure - if you use moving noise pattern then the flickering of denoised animation is much stronger. With static noise pattern it’s nearly non existent
Thanks for the tip, I indeed used static seed. I really just made 6 camera placement and launched anim rendering, so it was luck evermotion had it off Animation denoising is coming soon, so the problem will be soon even more subtle, I find it really ok actually as it is. As you say, it’s nearly non-existent, I guess Lukas is working so hard so that it’s also stable at very low sample count.
Yes, sure. It does look really good considering they are minute rendertimes, but I don’t think it’s the quality one could hand over as final to a client Non the less, if this is 1:15, then 5:00 per frame would probably be good enough, which is still really great for interior anim
Are you discussing semantics he never say he will merge the changes in master
bliblubli say he sending the code to the blender foundation if and if it merged is no more his responsibility
This may sound like a rough question to ask, or even a tricky one, but please don’t read this the wrong way.
What is the guarantee that these changes would be merged into master? I mean, this is great stuff, but I’m a bit reluctant to focus my workflow on a separate Blender version which development can be stopped at any time.
Even if you know how, maintaining an ongoing branch while master makes significant changes isn’t easy. Any time Blender developers touch the same regions of code that has been changed in one of the patches you applied to your version, you better understand how to code and resolve the conflict. It takes time to merge conflicts.
The probability of BF merging the changes is a big deal. If he is working for a year only to dump his changes on them ‘as is’ the probability of them accepting it is near zero.
Because by saying he will send it to BF, he is giving the impression that people helping him are helping the whole community. This simply isn’t the case. He isn’t talking with the developers at all so chances are many of his changes won’t be accepted. There are plenty of valid reasons not to accept a patch, but he could have fixed this by just being open with the community and BF devs about what changes are being made and why, which would have a big influence on both how many changes will be merged, and the realistic understanding of such by his customers.
Yeah he culd be doing this and still sell his service. People don’t really build their own binaries and especially don’t build their own with submitted and unaccepted patches. It couldn’t hurt to open the dialog with the developers now and get the process going. If they’re not accepted, no big deal, he’s still providing binaries for customers that offer something the main tree doesn’t. It also gets these kinds of things on the developers radars instead of just popping up a year later when developers might have already spent time working on the same issues but in a different, possibly incompatible but equally good way.
There’s nothing stopping anyone from getting a month of service or the class just for the patches and submitting them but it feels really icky because they’re not the developer and it would show up under their name instead of blibuli’s.
Animation denoising is coming soon, so the problem will be soon even more subtle, I find it really ok actually as it is. As you say, it’s nearly non-existent, I guess Lukas is working so hard so that it’s also stable at very low sample count.