REGION RENDER extremely important

[abstrax]

Just tested octane max 2.0, so disapointed, region render is SLOWER than rendering whole framebuffer ?!

No it's faster. What you see in the samples per second is the samples per second per frame, i.e. averaged over the whole image. To do one sample per second per frame you have to do many samples per second per pixel inside the render region.

[Dom74]

Just tested octane max 2.0, so disapointed, region render is SLOWER than rendering whole framebuffer ?!

No it's faster. What you see in the samples per second is the samples per second per frame, i.e. averaged over the whole image. To do one sample per second per frame you have to do many samples per second per pixel inside the render region.

Ok, I think I understand, but in this mode, we can't use the region render to have the exact number of sample per pixel in the selected area, I think
the region render must respect the required samples amount to match the whole picture.

[whersmy]

Can you do a comparison with region render?

[Dom74]

Can you do a comparison with region render?

Well, when I select the entire framebuffer with region render, the sampling rate seems to be the same, maybe a little slower,
but when I reduce my selection, sampling rate goes down, but I have more samples in this region than indicated in the info bar.
If I have well understood, for example, I have a 1 million pixels framebuffer and a rate of 1 Msamples/sec, if I select a 250K pixel region,
I still have the info of 1 Msamples/sec (or so), but in my selection, sampling will be 4X more effective, so, instead of stopping at max. 100 samples, I'll obtain 400 samples refinement.
Not really sure about that, maybe Abstrax can confirm.

[abstrax]

Can you do a comparison with region render?

Well, when I select the entire framebuffer with region render, the sampling rate seems to be the same, maybe a little slower,
but when I reduce my selection, sampling rate goes down, but I have more samples in this region than indicated in the info bar.
If I have well understood, for example, I have a 1 million pixels framebuffer and a rate of 1 Msamples/sec, if I select a 250K pixel region,
I still have the info of 1 Msamples/sec (or so), but in my selection, sampling will be 4X more effective, so, instead of stopping at max. 100 samples, I'll obtain 400 samples refinement.
Not really sure about that, maybe Abstrax can confirm.

Yes, that's exactly right. If your region covers 1/4 of the whole frame and you render an additional 100 samples/pixel, you have actually added 400 samples/pixel to the inner region and 0 samples/pixel outside the feathered border and something in 0 .. 400 samples/pixel in the feathered region.

Now you may ask, why do I still see a drop in speed, because shouldn't it still do the same number of samples per second, independent of the region? In principle yes, but in practice not for these two reasons:

- The render speed is not equal over the whole image. Some parts of the image are easier to render than others. To refine rendering you will typically pick areas that are hard to render, because they require complicated paths and result in more noise because of that. So if you extrapolate this to the whole image, it would be as if the whole frame would be hard to compute and therefor you will end up with a lower sample rate.

- If the region is small, not enough threads can be run in parallel which means the GPU can't hide latencies so well, resulting in a slow down.

We have an idea how to mitigate/reduce the latter problem, but didn't have time to implement it yet. We will try it when we get to it.

[glimpse]

Abstrax, Thank You for this explanation -
helps a lot to undestand how things work.

[Dom74]

Yes, thanks for your precisions.

[JNDesign]

Hi Guys
I have noticed that region render also drops the gpu load
We have 4 titans and when we don't use region render the load is 99% on all 4 but region render drops them to say between 40% to 60%
I am sure there is technical answer for this

[abstrax]

Hi Guys
I have noticed that region render also drops the gpu load
We have 4 titans and when we don't use region render the load is 99% on all 4 but region render drops them to say between 40% to 60%
I am sure there is technical answer for this

That's because of the second point in my explanation above.

[dysfunctional]

My friend and I were incredibly excited for the area render feature in 2.0 because we thought it would be useful for production of our 3D motion comics. We use overlays of images within frames to present static images in a more dynamic way and make it seem like action is occurring, while still holding on to the 2D medium of comics.

At first, we saw that the time to complete our maxsamples setting per rendered area was exactly the same as rendering the entire frame. But if you're saying that the best way to do it is sort of eyeball your region render and cut the maxsamples back by the inverse proportion, then that is useful information. Up until now, my friend and I have been complaining in a different thread that the area render time to completion and samples/pixel fields were not updating to reflect the area render, but rather showing the length of time to complete the whole render.

So, if I understand you, if we typically render each and every frame of our motion comics to 2,000 samples/pixel, if we need to render only an area where a scenery change (such as a character moving within the frame or something) has occurred, we need only trace our new render area with the area render tool, make an educated guess what percentage of the whole this area is (say, for example, this area we are rendering is 1/4 of the whole), then we need to dial back our maxsamples to 500 (because 1/4 of 2,000 is 500). Then, the final area render result should be consistent with the previous rendered frame in terms of quality and noise.

Do I understand that correctly?