Ok, I had a look and the problem is that in 1.01 and 1.02 the glossy material in the (old) benchmark scene gets loaded with an IOR of 1.3 (which is wrong), while 1.0, 1.03, 1.04 and 1.05 load the material with an IOR of 0 (which is correct). Loading it with an IOR 1.3 reduce the specularity a lot, which means that less bounces will be traced and it renders faster. Obviously the rendered image looks fairly different:FrankPooleFloating wrote:I think we may have possibly taken a step backward with v1.05....
[2] EVGA GTX460's
v1.02 = 4.00 M/s
v1.05 = 3.58 M/s
Can anyone else confirm somewhere around a 10% decrease in performance?
And are we supposed to be using that same benchmark scene (which I think is from 2010)?
(yes - I do understand that changing the scene would defeat (at least one) purpose... I just
want to be sure I am using the right one nonetheless)
IOR of 0.0:
Code: Select all
version | IOR 0.0 | IOR 1.3
---------+-----------+---------
1.00 | 22.7 | 27.2
1.01 | 21.8 | 26.3
1.02 | 22.7 | 27.2
1.03 | 22.7 | 27.2
1.04 | 22.3 | 26.8
1.05 | 22.3 | 26.8
Conclusion: Always check that your benchmark result is the same, when you compare those. And if you want to speed up stuff, try to tweak your scene, often you can improve things a lot this way.
Cheers,
Marcus
PS: And yes, the grass is always greener over there ...
