Is pt suitable for simple real-time rendering? Can this be implemented?

[Splizard]

Is pt suitable for simple real-time rendering? Can this be implemented?

[fogleman]

Not really. Most renders can take quite a while. Most of the 2560x1440px examples on the README took 30+ minutes, for example.

[mattgodbolt]

Even using GPU resources and a very much simplified path tracer it takes a while to render with path tracing: this is the best I can find at doing "interactive" rendering. It's quite a lot to expect a pure CPU implementation could keep up.

[joeblew99]

Am wondering if cycle based rendering helps ?
This is when the precision is lowered.

The other way is to allow tile based rendering. At least this allows speeding things up by throwing more servers at the problem.

[fogleman]

"Throwing more servers at it" is already easily doable. Each server can produce its own image and then the images can be combined. Since each server starts with a different random seed, the sampling will be just as if a single machine did 2x, 3x, etc the work.

This is what the combine script in the repo is for. :)

[joeblew99]

Tiles based is different from what you do now ? To render one image 50
times faster, can I use 50 servers ?

I saw the combine script and was wondering...

On Fri, 5 Feb 2016, 03:00 Michael Fogleman [email protected] wrote:

"Throwing more servers at it" is already easily doable. Each server can
produce its own image and then the images can be combined. Since each
server starts with a different random seed, the sampling will be just as if
a single machine did 2x, 3x, etc the work.

This is what the combine script in the repo is for. :)

—
Reply to this email directly or view it on GitHub
#11 (comment).

[fogleman]

@joeblew99 See my slides on path tracing: https://speakerdeck.com/fogleman/path-tracing

Specifically the section on Noise. Maybe that will help you understand.

Here's the video if you want to watch: https://www.youtube.com/watch?v=x4oQsQ76OHY

[joeblew99]

Yes i saw it thanks. I see what it is
This is all super but can I please get a straight answer to the original
question pretty please ?????

On Fri, 5 Feb 2016, 14:29 Michael Fogleman [email protected] wrote:

@joeblew99 https://github.com/joeblew99 See my slides on path tracing:
https://speakerdeck.com/fogleman/path-tracing

Specifically the section on Noise. Maybe that will help you understand.

—
Reply to this email directly or view it on GitHub
#11 (comment).

[joeblew99]

It's just that I have used many rendering engines and yet to see anyone
solve the shared memory model aspects for distributed rendering of photonic
ray tracing.

On Fri, 5 Feb 2016, 17:35 Joe Blue [email protected] wrote:

Yes i saw it thanks. I see what it is
This is all super but can I please get a straight answer to the original
question pretty please ?????

On Fri, 5 Feb 2016, 14:29 Michael Fogleman [email protected]
wrote:

@joeblew99 https://github.com/joeblew99 See my slides on path tracing:
https://speakerdeck.com/fogleman/path-tracing

Specifically the section on Noise. Maybe that will help you understand.

—
Reply to this email directly or view it on GitHub
#11 (comment).

[fogleman]

@joeblew99

To render one image 50 times faster, can I use 50 servers?

Yes.

Tiles based is different from what you do now?

It is different.

If you wanted to run 256 samples per pixel, you could:

• Run 256 on 1 machine
• Run 128 on 2 machines
• Run 64 on 4 machines

Etc.

Then you literally average the images and get a result equivalent to the full 256 samples.

[joeblew99]

I would like to try it. Will have a look at see if the code has some sort
of render director.

On Fri, 5 Feb 2016, 23:27 Michael Fogleman [email protected] wrote:

@joeblew99 https://github.com/joeblew99

To render one image 50 times faster, can I use 50 servers?

Yes.

Tiles based is different from what you do now?

It is different.

If you wanted to run 256 samples per pixel, you could:

• Run 256 on 1 machine
• Run 128 on 2 machines
• Run 64 on 4 machines

Etc.

Then you literally average the images and get a result equivalent to the
full 256 samples.

—
Reply to this email directly or view it on GitHub
#11 (comment).