Daily Builds 1.0 - 1.4

[Ludvik Koutny]

AFAIK denoise uses GPU, so if CPU temp jumps up, maybe GPU starts crunching, creating heat, and perhaps GPU fans somehow blow hot air onto CPU heatsink or somewhere around CPU temp sensor :)

[juang3d]

Maybe is what rawalanche says, a not optimal air flow may be provoking this, all depending on how do you have your refrigeration configured.

Do you see the same temperatures if you render with iRay for example? In that case you would be in the same situation using CPU+GPU, so if the GPU is the problem you should see the same temps you see now in the denoising phase I think.

Cheers

[Ondra]

Corona denoising is 100% CPU. It looks more like you did not test your overclocking solution well enough in burn tests - denoising does a lot of math operations on coherent data, those are usually the most efficient dataloads for modern CPU. A lot of instructions executed = a lot of heat produced.

[antanas]

@snakebox 
It seems like the turbo boost kicks in on top of your regular overclocking, that happened to me too when I was oc'ing my 3930k (which is still doing pretty well on 4.5ghz) but that was some years ago and, sadly, I don't remember how I've dealt with that but still I did that somehow )) - the question is why that happens only during denoising stage and not during the regular render itself, later I'll try to test that too.

[snakebox]

@snakebox 
It seems like the turbo boost kicks in on top of your regular overclocking, that happened to me too when I was oc'ing my 3930k (which is still doing pretty well on 4.5ghz) but that was some years ago and, sadly, I don't remember how I've dealt with that but still I did that somehow )) - the question is why that happens only during denoising stage and not during the regular render itself, later I'll try to test that too.

My cpu is locked and running a constanty speed. The turboboost is basically active 100% of time. So there is no "more boost" happening.  It's running 4.5 or 4.6 or 4.7 or what ever constantly. 

@Ondra   very possible that there are cpu computations that are heavier than others, but I don't understand is simply just how 100% is different from another 100% load, that's all.   Also are you telling me that rendering corona isn't utilizing the full cpu power then ;)

[Ryuu]

but I don't understand is simply just how 100% is different from another 100% load, that's all

It's not that unheard of, really. All the reports of 100% utilization really mean that there is some thread scheduled on the CPU 100% of the time. And that thread may be doing some heavy computation on the data available, or it can be mostly just waiting for the memory controller to fetch the data. In both cases, you'll still see 100% utilization in the task manager (or anything else measuring the utilization), but the effects on temperature will be vastly different. There are also other funny factors like CPU switching off power for unused parts of the pipeline and so on.

[DelightingPictures]

First Test with Denoise 1.4 Corona Render. 2 minutes for time render in Intel 5960x. 1500x1300 resolution.

[antanas]

@Ondra @snakebox @Ryuu - you are right - did some testing and yeah, during regular rendering for at least 15 minutes highest cpu core temperature was somewhere around 74 degrees C yet for some short time during denoising process it jumped up to 79 degrees, so that's definitely not a turbo boost related - it's more like the behavior of rendering (using any renderer) vs prime or OCCT testing - so most probably denoising really does produce some higher cpu and memory controller loads indeed + it also shows what my cpu's thermal grease maintenance is a long time overdue - 4 years without changing it is a little too much it seems ))

[romullus]

I would like to rise the question about splotchiness after denoising. Rawalanche did open ticket about it in mantis, but it was closed with explanation that it's really hard to fix it. You can see this effect  quite clearly in renders uploaded by jesuscaballero (on the blanket and in less degree, on curtains). Flat textureless areas are especially prone to this splotchiness and from my experience it shows itself even at as low noise treshold as ~3%.

I'm not complaining much, i just want to know if it's something that possibly will be looked at in the future, or it's completely unfixable?

In any case, to my great surprise, i find denoising very useful and i use it and like it very much!

[pokoy]

Similar to the experience from last week with a render node, I am facing a problem with a completely new render node where I can't install the new dailies because installation of the ms vis c++ redist fails... Is there any fallback solution or workaround to such a problem? Also, I can't run the benchmark since it seems to install similar files behind the curtains.
I wonder if I'm the only seeing this and how I managed to see the same problem on 2 different machines.

EDIT
Installing the redist worked after ~10th attempts, after I posted the above question here - just like the last time on the other node where I've struggled with the problem and it magically resolved itself after I've posted it here. Strange coincidence, looks like black magic to me :D

[maru]

looks like black magic to me

rendering is black magic

[Ludvik Koutny]

I would like to rise the question about splotchiness after denoising. Rawalanche did open ticket about it in mantis, but it was closed with explanation that it's really hard to fix it. You can see this effect  quite clearly in renders uploaded by jesuscaballero (on the blanket and in less degree, on curtains). Flat textureless areas are especially prone to this splotchiness and from my experience it shows itself even at as low noise treshold as ~3%.

I'm not complaining much, i just want to know if it's something that possibly will be looked at in the future, or it's completely unfixable?

In any case, to my great surprise, i find denoising very useful and i use it and like it very much!

Actually they seem to occur mainly on the curved surfaces. Flat surfaces seem mostly fine. There may be some internal angular threshold value that may need some tweaking. I am not very happy about current state myself, as denoiser produces quite obvious splotches even on scenes that are converged quite far.

[romullus]

Sorry, by saing flat, i ment a plain, featurless surface. My english is very meh.

Interestingly, i didn't noticed that there's correlation with geometry curvature at first, but now that you mentioned, i took a second look at my worst examples of splotchiness and i think i agree with you on that.

[Noah45]

Love and use Denoise everyday, but those blotches do not animate well.

[alexyork]

Denoising is amazing. We're using it now a lot on stills and animation. It's saving us a huge amount of time.

But what we do need is for Corona to optionally output 2 files for every output - one with no denoising and one with the specified amount. This way we can blend between them in post as needed.

One thing that would be even better would be to have denoising with include/exclude list, so you could include only big white walls for example, and avoid denoising fine wood textures where you'd lose detail, or glass/mirrors, where denoising doesn't work properly.