On 8/27/25 08:34, William F Pokorny wrote:
> Aside: Puzzling performance results. I am unsure why yuqk is so much 
> faster with the scene set up for the attached image?
> 
> v3.7
> 577.29user 0.33system 2:42.95elapsed 354%CPU
> 
> v3.8 beta 2
> 572.11user 0.35system 2:33.45elapsed 373%CPU
> 
> yuqk
> 105.71user 0.13system 0:33.08elapsed 320%CPU

With global photon jitter on:

1) The number of stored photons for v3.8 beta 2 remains about the same 
whether the jitter is on or off. With yuqk there is an increase in 
stored photons of about 2.3% over yuqk run with global photon jitter off.

2) With global photon jitter on, the number of stored photons is exactly 
the same between v3.8 b2 and yuqk.

3) The performance given (2) becomes comparable too:

v3.8 beta 2
131.73user 0.20system 0:42.00elapsed 314%CPU

yuqk
131.10user 0.14system 0:40.54elapsed 323%CPU

---

So, the yuqk fix to the photon distribution where jitter is off seems to 
be the trigger for the many times better performance in that case. 
However, I'm left lost as to why a better distribution of stored photons 
matters so much to the 'render stage' performance.

Been a long time since I was in the photon code for the photon 
distribution fix, but IIRC, the global photon jitter is 'jittering' the 
spiral shooting pattern of forward tracing photon rays. The yuqk fix 
changed how the photons are distributed along each shot photon ray in 
the media.

Both the fix and the spiral photon shooting jitter make the distribution 
of stored photons more noisy / random in the stored media - and this 
apparently helps 'something' in the photon code work a lot faster.

Bill P.

Post a reply to this message