CoolColJ's test pics thread

[joedizzle]

real camera have way more diffraction at the same fstop as Indigo - me thinks the simulation needs tweaking

Maybe that (sadly still not released) Paper could be a good starting point:
Polynomial Optics: A Construction Kit for Efficient Ray-Tracing of Lens Systems

It's here, http://www.cs.ubc.ca/~hullin/#publications, interesting paper!

[CoolColJ]

Trying out an old exit portal test scene from Indigo v2 with v3.4.2
with an i7 970 - 6 core at 3.85ghz

Damn clean for only 4 mins and 363 samples

Ha - only 1 min, this new improved BPT is pretty fast... ahh memories of the old days with a Pentium 4 are no more...

Anyone with an x58 board, definitely pick up a second hand i7 970. Not much slower than a 3930k at the same clock speed, but it won't overclock as well. Runs much cooler than an i7 920 or 930 at the same clock speed!

[dcm]

2600k ftw

[CoolColJ]

2600k ftw

2 cores short of penile length

[StompinTom]

2600k ftw

2 cores short of penile length

Baha!

Coming from a 980X over here...

[FoXar]

2600k ftw

2 cores short of penile length

Baha!

Coming from a 980X over here...

Oh don't "Baha!" too much, also have a 980X, clocked at 3.8GHz and watercooled..
Let's see who is "Baha-ing!" next

[CoolColJ]

Trying to brute force this on my EVGA GTX 1070 FTW
and OpenCL path tracing - think I might resolve fairly well overnight

The underwater block seems to converge better than with MLT though

[Oscar J]

Oh man, classic thread. This is a gruesome task for single dir PT.

[lycium]

I'm so glad to see this classic thread again, after 4 long years

[pixie]

Epic thread is epic!

[CoolColJ]

So I have a Ryzen 5950x system now.... it's quite a bit faster than my previous Intel 3930k system
Also MSI RTX 3070 Trio

I decided try this classic scene out again, and the underwater bits were a bit slow to resolve, with a low sun angle, using the captured simulation, but started to show up after 10000 samples per pixels.

Now after 9 hours, at 6+ million samples per sec and 700k samples per pixel, it's actually fairly resolved

edit - I get 7 Million samples per sec with latest release candidate on the same scene

I also found setting the max path depth to higher numbers to be helpful, like over 32. I used 99
Especially in scene with high reflection and refraction. The caustics develop better and faster

[Headroom]

LOL, yes not a surprise. What render mode did you use for that image ?
The mention of max path depth indicates GPU use. But I would have imagined that the bidirectional path tracing, perhaps even with MLT converge much faster than brute force path tracing on the GPU.

I've got a number of renders where GPU rendering appears to be almost helpless. I can share those but don't want to spoil your thread with my images.

[CoolColJ]

LOL, yes not a surprise. What render mode did you use for that image ?
The mention of max path depth indicates GPU use. But I would have imagined that the bidirectional path tracing, perhaps even with MLT converge much faster than brute force path tracing on the GPU.

I've got a number of renders where GPU rendering appears to be almost helpless. I can share those but don't want to spoil your thread with my images.

Bidrectional MLT on CPU... so the max path depth has no effect on CPU then?

It's much worse on GPU

Feel free to post in this thread

[CoolColJ]

I have rendered this one before, but I always lost patience with it as it takes ages to converge.
Now with a faster CPU I redid it, and it took till 15k samples/pixel before the details started to show more on the red cyclinder, and +30K till more details showed up.

At 75+k S/P after 88 mins of rendering I think it's close to full converged

[CoolColJ]

ran this one for over 10 hours, restarted a few times, and the underwater block did eventually converge at 400k samples/pixel. I think another 5 hours will make it smoother.

So in 5 years time when CPUs are 10x faster, this will be no big deal, relatively speaking