500x1 samples vs 1x500 samples - time difference

[maru]

I've rendered two images using bucket mode but it could also be achieved with progressive:

500x1 - 500 path tracing samples rendered with 1 initial sample (equivalent of 500PTS x 1 pass in progressive mode)

1x500 - 1 path tracing sample rendered with 500 initial samples (equivalent of 1PTS x 500 passes in progressive mode)

I understand why there is a big visual difference but my question is: why is there such difference in rendering time? Why is sampling one pixel 500 times with 1 path tracing sample slower than sampling one pixel 1 time with 500 path tracing samples?

[rafpug]

Hi maru,
nice ^ _ ^

As soon as I have some time I start seeing this method bucket

[Ondra]

easy: there is additional overhead for primary rays. If there was none, having PT samples parameter would make no sense

[maru]

Ok, thanks. I'm just trying to understand as much as I can. :)

[Stan_But]

Maru, PT does not equal In.sample - it is your main mistake.
I use BM too and i'll show example for you later

[maru]

Maru, PT does not equal In.sample - it is your main mistake.

Your main mistake is that you didn't understand what I wrote.

What I meant is: [path tracing samples] x [initial passes] in bucket mode is the same as [path tracing samples] x [passes] in progressive mode.

[maru]

Ok, Headoff, you were right. But not entirely. Turns out that:

in progressive mode [path tracing samples] x [1 pass] = in bucket mode [path tracing samples] x [2 initial samples in bucket mode] x [pass]

Here is proof. If you view these images in some viewer and you can jump from one to another, even the noise is the same:

[maru]

One more test showing that rendering a scene with

progressive: [X pt samples] * [Y passes]

is exactly the same as rendering it with

buckets: [X pt samples] * [X*2 initial samples]

phew

[Stan_But]

as i understood when BM set 1/1/X - AA is OFF.  It's why we see such image without anti aliasing.
And i tested it until become clean image. I mean that settings pt 200  vs. 400/1 did same clean result. Like as you showed in your latest post

[maru]

Ok, I'm glad we've come to and agreement. :)