Indigo Renderer Forum

Ok guys...
something that really shocked me hard after I just tested the Mac Build of 2.4.8:

Win 4x3GHz => 179k sps
Win 2x2.8GHz => 94k sps
Mac 2x2.8GHz => 143k sps

I used latest 2.4.8 beta on Vista 64 vs Mac OS 10.5.
Render Mode was pure PT, without fancy extra stuff.
iMac with Bootcamp is the 2x2.8Ghz machine.
3x3Ghz is a AMD PC, that scales fine with results of iMac@win !

2 possibilities for this I could imagine:
1) Mac Build is buggy and does faster samples by taking "shortcuts" that result in biased rendering
2) Win Build was compiled badly since a long time, and wastes a lot of potential!

Please investigate here! A "fixed" win build would give us a +52% speed bonus
without any new crazy scientist stuff to be integrated!!!

Could you send the scene to me? I have a similar processor to the one you used in your test (4 x 3.4Ghz) and I'm running windows.

Fix fix fix!

Godzilla wrote:Could you send the scene to me? I have a similar processor to the one you used in your test (4 x 3.4Ghz) and I'm running windows.

you got PM

2.4.7-
4m 16s, 410.41 SPP, 420k S/S

2.4.8
4m 13s, 271.30 SPP, 280k S/S


Not really a noticeable difference in noise, but some weird artifacts appeared on the 2.4.8 (And some artifacts disappeared from the 2.4.7 version).

Run on Windows 7 using a 3.38 Ghz Quad-core (Intel Q6700).

in mac vs win test you also see difference in noise!!
Here samples are way higher for 2.4.7, also new artifacts on the left side of the chrome bold.

Zom-B wrote:new crazy scientist stuff

Well, this will definitely be looked at today!

lycium wrote:

Zom-B wrote:new crazy scientist stuff

Well, this will definitely be looked at today!

Lol, your turn mr. Ludwig. I just made indigo 170% faster.

roflcopter

fused wrote:Lol, your turn mr. Ludwig. I just mad indigo 170% faster.

roflcopter

Do it for the Windows build

lycium wrote:Do it for the Windows build

Uh.... I currently have other stuff todo, err.... or so.

We have tested it on our office mac (2x2.4 Ghz) and we can confirm:

Win7: 95k samples/s
OSX 10.6: 123k samples/s

edit:
comparing indigo console:

Win7: 99k samples/s
OSX 10.6: 144k samples/s

edit 2:
2.4.7 vs. 2.4.8 on Win7 (no GPU):

2.4.7: 89k samples/s
2.4.8: 99k samples/s

Ok, I justed re-benchmarked my Win 2x2.8 System, with higher sps -.-
So the total Speed gain "dropped down" to +53%...
Sorry for that, I updated the first post!

Maybe somebody could do a Win vs Linux Benchmark on the same machine?!

I did a Win vs Linux benchmark back in 2.4.4 days or so, with Linux running in the VM, and it was still faster. Don't remember by how much, but the take-away point is that GCC is kicking Visual C++ compiler really hard!

We've done some tests at the office with different compilers and are considering making a change there, since 50% is really worthwhile (and it would be nice to use a single compiler on all platforms).

lycium wrote:I did a Win vs Linux benchmark back in 2.4.4 days or so, with Linux running in the VM, and it was still faster. Don't remember by how much, but the take-away point is that GCC is kicking Visual C++ compiler really hard!

We've done some tests at the office with different compilers and are considering making a change there, since 50% is really worthwhile (and it would be nice to use a single compiler on all platforms).

w00tles I'm assuming this speed upgrade has chances to boost the GPU beta too?

Borgleader wrote:

lycium wrote:I'm assuming this speed upgrade has chances to boost the GPU beta too?

This is something I hope too!
Also the GPU integration of BiDir could be retested after updating the compiler.
The OpenCl render Mode could benefit too!