So I still am still unsure of the BEST workflow…
I have been using VRay since 2007 and I still don’t really seem to feel confident that I know what the best work flow is.
when it comes to final renderings and trying to produce a clean, noise free image, that is the most confusing area . BUCKET or PROGRESSIVE?
it seems that I have been getting decent results when using the bucket option, with all settings at there defaults except noise threshold and LC.
no here is where the confusion sets in…
last night I was submitting some stills to the Rebus farm with these settings and I get the warning message that I should switch to Progressive and my Image Sampler is set high (I think that is the warning)
So I would like to switch to progressive if that is going to give me better, quicker results, but I am not sure what the correct settings are
my confusion starts with the render time, how do I know how long to set this for, to get a clean image? or do I set the time to 0 which is not limited to time, so what controls the amount of time or passes so the Rebus farm knows when to stop and doesn’t bankrupt me
Sorry if these are stupid questions but it really is confusion to me
Yeah, I wouldn’t really trust what Rebus says about using progressive instead of buckets (especially from a company that benefits from longer renderings), but that’s my two cents
Progressive is generally slower than a well-set-up bucket rendering.
Also when timelimit is set to 0. Rendering stops when it has reached the noise threshold you determine.
Hey Ashley thanks for replying back (at least somebody did) but what do you adjust to create your well-set-up bucket rendering
I thought, according to various post going with most of the “defaults” is the way to go
Progressive is much better for look development, that’s for sure.
But Buckets are certainly more efficient for DR.
I think it also depends which is straight faster on one machine, no DR, whether you are using GPU or CPU.
Notice that when you switch between buckets and progressive, you have different settings you can save - so you can have your higher quality rendering ready for buckets and your look dev for development.
There are multiple situations where each has advantages and disadvantages though, and it’s not a case of which is better for one final fast render, but what is faster for what you are doing - preview rendering? making apreview render animation that you want to improve later? distributed rendering a still?
Progressive has issues though, like if you have a large frame with empty space and then a small but detailed object taking only lets say 5% of the frame. In that case with buckets you could set the bucket size to be small and get all cores working on that area but progressive will not utilize cpu efficiently in that way. How to get a clean render, depends on your conditions in the scene though its hard to say. Vray has shifted to a single method on relying more on the sampler to get clean image rather then individually tweaking certain settings. So lower noise threshold higher max subdivs will do the job.
You have to separate the two issues. Progressive helps clean up noise. Gi BF vs IM is a different type of issue (solution). While high max subdivs in the dmc sampler help clean up the noise (all kinds of noises) including brute force they don’t affect irradiance map. This is why im is still better for interiors (speed wise). Here is a quick test, dmc sampler at 1 / 1, you can see where bf is really noisy because it didn’t get any samples from dmc sampler.
I think the 1/24, 0.01 setting is actually quite strong in terms of how it engages the dmc sampler. It may be good for a final quality setting but if you don’t use brute force you might get away with half that, especially if you use denoiser. Depends on what the quality you are after. Obviously using high max subdivs, low noise threshold and brute force will give you best result but also longest render time.
And here is another quick example. While the dmc is generally good at cleaning up noise, it does not mean you should rely just on it. I mean its fine if you have lots of time and render power its certainly easy. But here is a quick example where I can get same amount of noise using local subdivs and increasing subdivisions from shaders and brute force while keeping the dmc low, vs having local subdivs off and having the dmc do all the work. The two images are very close in noise level however render time is nearly double for the image using purely dmc sampler.
Fortunately with the latest 3.6 (not version NEXT), I usually don’t have to change any settings to get a quality render - just switch renderer from Vray to any other renderer, then back (resets the vray settings to default) then switch from progressive to bucket. The more I have used Vray across the 3. lifecycle, the less I have had to fiddle around with render presets or changing settings - add a VrayDenoiser render element to your render elements if you do have noise too
that’s what I had read and thought as well, go with the defaults of a fresh VRay reset, switch to bucket and your good to go.
sound like you ice_creative have not switched to NEXT as well as myself?
I have made use of the NEXT trial, the best part was GPU bucket rendering on that, didn’t notice huge speed improvements on some CPU render projects I worked on with it but I think in a few months it may become essential to move to NEXT if you are using a GPU for rendering (we aren’t yet) and they will keep improving/supporting that version of Vray too . Anyway, happy rendering with the Vray default settings, I still chuckle thinking about the massive folder I had of render presets for 3ds Max that I used to depend on so much - hats off to Chaos Group team for making the software much simpler to use.
This is only valid for simplistic scenes.
It’s been shown, and measuredly so, that in an average scene the defaults are the best possible surefire solution.
Regardless of compute power, to a given, uniform noise threshold defaults can be considered as optimal as it could be achieved, regardless of scene contents.)
We do not recommend activating local subdivs and returning to low-AA workflows, or changing MSR, because of a number of issues (arbitrary oversampling and undersampling, uneven noise levels, loss of fine geo detail, and so on and so forth.).
On bucket Vs. progressive, well, buckets are progressive. just the progress isn’t shown until the bucket finishes.
There should be little to no difference in speed or completion time between the two engines (one has the stuck bucket syndrome, the other renders for a long time in the same areas. One checks adaptivity inside the single bucket, the other for the whole frame at once, but then bucket edges need more AA, the smaller the buckets, the more the work, which often compensates for the difference.).
Should the two be very different in rendertimes, say more than 10% either side, it ought to be reported as a bug.
Well, progressive always renders much slower on our end.
Also, if buckets are progressive, can we make them visible? That would come in handy. Maybe there is a stuck bucket on a headlight (happens all the time) and I cancel the render, I can still use the result. Right now I have an empty bucket. I would take a slightly noisy result over a black bucket all the time.
Sure. Head over to the beta group though, so not to confuse the rest of the users.
It’s really been thoroughly tested and demonstrated, and i’d rather not restart a closed argument in public. @kosso_olli , read above, if it’s more than around 10%, it’s likely an issue, report it please.
Here is a test from a production scene. 4K resolution, one domelight and an area light in the scene, some VrScans, DOF. Bucket mode took 47 minutes, while progressive took 64 minutes, so progessive was around 36% slower. Noise threshold was set to 0.01. Noise pattern is different of course, but they look identical. One thing that I liked about progressive is a little highlight in the DOF, which came out much cleaner for progressive. However, a 36% render time hit is to much for us. I am sure it will get worse the higher the render resolution gets.
It’d be great if you could save a scene with either render setting, and send it over to support for profiling.
I can imagine a bigger size render suffers more of the all-image noise level detection than a smaller one (that’s also why you get a slightly better DoF than with bucketing), but that shouldn’t be enough.
*Many* REs *do* slow down progressive compared to bucketed, but then we’re talking 10-30 and more REs at once, is this the case?
Super!
I’ve meanwhile written a script to measure the “slack” in progressive versus bucket, and output a CSV, with a few conditions (with and without tiled textures, or proxies, which need to unload; with and without many REs, at small and bigger image size, and so on), to help figure out possible bottlenecks.
I’ll update it as i’ll get your scene to better reflect a production setup case.
Would you be so kind to CC me in the mail to support?
