Progressive VS. Bucket for Final

[danio]

I guess you haven't seen his video explanation of the Rockwell Retro Encabulator

https://youtu.be/RXJKdh1KZ0w?t=17

I didn't know Lele was on YouTube...!?!

[kosso_olli]

To get back to the topic, I just made some more tests. The image I rendered was this, lots of motion blut on the entire image. Bucket took 48 minutes, while progressive took only 29 minutes. I have to say that bucket had immense "stuck bucket syndrome" on the headlights, though. So there really is a benefit for progressive.

[Alex_M]

Interesting observation. Wouldn't it be cool if Vray knew upfront which parts of the image would need lots of sampling and dynamically split big buckets into smaller ones in these particular places? But now that I think of it, maybe it hasn't been done so far because it's impossible or has more downsides than upsides?

[kosso_olli]

Interesting observation. Wouldn't it be cool if Vray knew upfront which parts of the image would need lots of sampling and dynamically split big buckets into smaller ones in these particular places? But now that I think of it, maybe it hasn't been done so far because it's impossible or has more downsides than upsides?

If I remember correctly there was some research done in that direction.

[Muhammed_Hamed]

Vlado talked on CG Garage some time ago, about using LC for dynamic bucket splitting to solve this issue. GPU buckets are much faster though, and with hybird rendering you get even smaller buckets for CPU
I haven't ran into stuck buckets for a long time, will probably stick to bucket mode for now
Still it is impressive that in Vray it is totally up to the user to choose between progressive/Bucket, they are both fast enough and no major limitations
.. In RS, progressive mode doesn't use any kind of adaptivity, no biased GI ,all sampling settings are ignored, point cloud SSS will render as Raytraced SSS(will not look the same)
In Cycles, progressive mode is terribly slower compared to Bucket mode, but both will give the same result,
Other GPU renderers are only progressive so you will need to render high res. images in tiles..

[kosso_olli]

V
I haven't ran into stuck buckets for a long time

I run into stuck buckets all the time. Mostly on parts with high refraction depth like headlights.

[JuhaHo]

I run into stuck buckets all the time. Mostly on parts with high refraction depth like headlights.

I'm having that problem too at the moment. My current project is a boat animation, where the frame times are something like 15 hours or more on 1920 x 1080 resolution when the Phoenix FD foam is on. Without the foam it's more like 15 minutes or less. The CPU usage goes down when the non foam buckets are rendered. The last one or two buckets require most of the 15 hours of render time.

Anything I can try, or should I call the customer to forget the animation and settle for stills only?

[Muhammed_Hamed]

I was talking about Bucket with Cuda engine, They never get stuck..
If you are using CPU, maybe try smaller buckets,
or for animation and 1080 res (if you don't use DR), progressive should work nicely and you can use a time limit for each frame, say 30-40 minutes depending on your setup

[JuhaHo]

On the latest frame, I got over 18 hours rendering time with bucket rendering. I switched to CUDA progressive rendering which gave frame time of 26 minutes. However there was no foam on the rendering and I saw some artefacts too. The quality of the rendering is not usable at all when compared to CPU bucket rendering results.

I will test next iterative mode.

[JuhaHo]

My first attempt with progressive rendering set to 10 minutes took 1 hour and 10 minutes with horrible image quality.

[^Lele^]

I hope that my criticism didn't bring any hard feelings.

None whatsoever, the arguments you make are sound, albeit some of the replies quoted were conversational, rather than educational.
The criticism is however well aimed, and worthy of working on for me.

[^Lele^]

I was talking about Bucket with Cuda engine, They never get stuck..

We're currently talking strictly CPU here, as the GPU bucket engine, we established, is way too young to perform as it should, and hopefully soon will.

To get back to the topic, I just made some more tests. The image I rendered was this, lots of motion blut on the entire image. Bucket took 48 minutes, while progressive took only 29 minutes. I have to say that bucket had immense "stuck bucket syndrome" on the headlights, though. So there really is a benefit for progressive.

kosso_olli i managed to hijack one farm slave to profile on, and it's still going (i tried base res, half and a quarter, progressive and bucket, with and without the REs set.).
Interesting that progressive did better in your test, what was the original render resolution?

[glorybound]

For quick setup, progressive is awesome on my main machine, but once I tick distributed rendering, my rendering times soar. I just stopped one that was going to take 1 hour 40 minutes. I set it to a bucket render and it'll take 12 minutes. I don't know why this happens and if progressive is even supposed to be used with DR, but it doesn't work well here. Vlado mentioned that it might be elements, which I haven't tried turning off yet.

[Alex_M]

Always buckets for final renderings here. Progressive still can't top buckets off for that, at least for my use cases. Better speed, CPU and RAM utilization. I only use Progressive for previews because I can stop it at any time and still have the full image.

[^Lele^]

For quick setup, progressive is awesome on my main machine, but once I tick distributed rendering, my rendering times soar. I just stopped one that was going to take 1 hour 40 minutes. I set it to a bucket render and it'll take 12 minutes. I don't know why this happens and if progressive is even supposed to be used with DR, but it doesn't work well here. Vlado mentioned that it might be elements, which I haven't tried turning off yet.

Element can add to the rendertime, but not in that magnitude (unless maybe you have a few hundred MultiMattes, who knows...).
Often defaults catch people out, as bucket is set to 0.01 NT, while progressive at 0.005, meaning it would take four times the base rendertime, all else being equal.
Second important quirk: defaults also have a different max AA, higher for progressive. This, coupled with the lower noise threshold, will lead to much higher rendertimes.
So, as a general rule of thumb, remember to match AA and noise threshold between engines.