RC5 25% faster than SP2 and little better quality!

[^Lele^]

Yes, same results as me.
I tested the sampler (with the math in my matrix exposimeter) to find the exact same sampling amount across the image.
In my case, 0.5 (ie. in between min of 1 and max of 16, on average).
The AA noise threshold to achieve that was around 0.0073 for dmc and 0.0072 for rQMC (well inside the error margin, so to be considered identical).

Turns out the DMC is slightly slower, but only ever so slightly (1-3%).
Not bad for being just the second public appearance .

[pailhead]

^

we need a "simple english" button like the one on wikipedia

[^Lele^]

lol.
let's put it this way: with the "show sampling" option on, the renderer returns a bluish image.
The darker the blue, the less work the AA has done (ie. min subdivs in the DMC AA).
The brighter the blue, the more work it has done(ie. Max subdivs for the brighter blue).
Say you take any image, and average out the color in the sampling pass.
What you'll get is the average sampling for that image.
in the exposimeter, I set a goal average -normally 0.5 float- and then let the exposimeter render a sampling pass and calculate multipliers for the noise threshold: if it's higher than the target, it's sampling more than desired, so it will raise the noise threshold.
If it's lower, the opposite is true.
By doing this, i ensure that even across VRay versions and samplers, the AVERAGE sampling for a given image is the same.
So on average the two images will have traced the same amount of rays (of course, give or take a small % of error).
Hence i can compare speed in an objective way, rather than in a subjective one (numbers, in other words, rather than perception of noise levels).