How much does tile size affect rendertime

[ultrasonic]

Hi guys,

another question i am just a little curious about...i know i could easily test this in a production scene but i would also like to generally know how much impact the tile size has on the final rendertime...let's not take memory usage into account...say i would render one frame with 16x16 and with 64x64 tilesize...will the large tilezize be much faster?

cheers
Oliver

[Metzger]

I'm a 16x16 guy.. I want instant feedback.. Not wait for a large chunk to finally poop out. So to speak.. Hehehe

[ultrasonic]

I'm a 16x16 guy too, but i was wondering if my choice would otherwise cost a lot of overhead in rendertime when rendering a sequence...so we could maybe add a preRender script which set tilesize larger when in batchmode...

[vlado]

Do you mean the render bucket size? Or the tile size for tiled OpenEXR textures?

Best regards,
Vlado

[ultrasonic]

render bucket size

[chuckie7413]

pretty sure they mean render bucket size. would like to know this too!

I think it is more efficient to batch render with 64x64. Normally you would use 16x16 to help reduce memory issues but also for faster feedback when wanting to watch an interactive render but when batch rendering at 16x16 especially when using distributed rendering I can imagine it creates a lot more overhead.

Interested to know.

Thanks,

Richard

[Dariusz Makowski (Dadal)]

In max Its a huge difference. A 16 bucket give u instant feedback but a 64-128 give u 2x more speed. Just do a test and have a look at statistics u'll see.

[vlado]

If you use an antialiasing filter, V-Ray needs to calculate a border of pixels around each bucket to make sure filtering is consistent from one bucket to the next; for smaller buckets this border might be a significant proportion of the number of pixels in the bucket itself. For example, suppose you use an Area filter with size 1.5 pixels (the default). This means that the additional border is 1 pixel thick on either side.

For 16x16 buckets, you get additional (18x18-16x16)=324-256=68 additional pixels, or about 26.6% more than the time you would have without filtering.
For 64x64 buckets, you get (66x66-64x64)=4356-4096=260 additional pixels, but they are only about 6.3% more than the time you would have without filtering.

Now, if you use 16x16 buckets, you will need to render 16 buckets to cover one 64x64 bucket, so V-Ray will compute ( 18x18 )x16=5184 pixels compared to the 4356 pixels if you use 64x64 bucket. In other words, with 16x16 buckets, you could expect about 5184/4356=1.19 times slower rendering or about 19% longer render times.

Best regards,
Vlado

[ultrasonic]

thx vlado...that was the best answer i could imagine! and i think 20% is a considerable speedgain! i know it also depends on imagesize and number of cores, but your answer could work as a general rule of thumb!

cheers
Oliver

[Neilg]

So does this only affect the speed when filtering is turned on?

[ultrasonic]

So does this only affect the speed when filtering is turned on?

If I understood it right then -> YES...only when filtering is in....

[vlado]

Yes, this is only when filtering is enabled. Also, my calculations were only for filter size < 3 pixels. For larger filters (e.g. mitchell-netravali or catmull-rom) things are slightly worse as then a larger border is needed around each bucket.

Best regards,
Vlado