VRay and SSE

[Brim]

When rendering with Max Scanline there is an option to turn on SSE instructions. I noticed VRay doesn't have this option in its render dialog. Does VRay have use SSE and if so is it enabled automatically? Does it matter with VRay?


-edit-- fixed some typos

[Brim]

Nobody? 53 views....

[flipside]

don't even know what sse instructions are...

[Psy]

@flispide
intel specific cpu instructions.

if a program is written to take advantage of these instructions, it will run faster.

look at lightwave for instance, when hardware sites do product comparisons, they rarely use lightwave to benchmark a p4 vs an amd for performance. reason? because lightwave is heavily optimized for sse and sse2 (which are intel specific instructions) amd has gotten in on it and is starting to write support for sse and sse2 into their chips however.

its a good thing, but a lot of people see it as product branding/favoritism.

@brim
I really have no idea, I'd assume it does since that gives it another claim to fame in its speed. PM valdo or peter and ask them, they gotta know, they wrote the program.

[Dynedain]

the SSE instruction checkbox is to turn on optimizations for particular processors that use SSE (Streaming SIMD Extensions). Its an extra set of 60 processor commands present in the P3, P4, and Itanium. Its what Intel was advertising like crazy when they said the P3 made the internet such a richer experience and were publishing "P3-only" websites and content.

[Intenzity]

the SSE instruction checkbox is to turn on optimizations for particular processors that use SSE (Streaming SIMD Extensions). Its an extra set of 60 processor commands present in the P3, P4, and Itanium

By chance would you know if there are any downfalls? Like if SSE may not work correctly or is there possibility it can slow down rendering instead of speed it up?

-Andy

[Dynedain]

discreet basically says in the help file if it speeds things up, use it, otherwise, dont


seems to me to be one of thos fuzzy, magic options

[Brim]

Open islands.max that ships with MAX (3dmax5/scenes/effects/islands.max), render with SSE off and then with it on.

I'm using an 1900XP with 1024MB PC2100 DDR, render @ 720x306 (defualt when you open scene):
SSE on - 19 secs
SSE off - 22 secs

This is with Max default scanline renderer. I'm still curious as to how VRay is affected by SSE - is it hardcoded "always use" or does it not use it at all. I think probably Vlado can answer this question.

[losbellos]

i used vray on an amd k6-2 500 without any problem, so... I guess it doesnt have such optimalizations or it will turn on or off automatically. It is also possible that just some part of the renderer use this feature for example the "ray shooter"
algorythm(). So then no need to worry about it in development.

[vlado]

Currently VRay doesn't use SSE instructions for compatibility reasons. The last time I tried to compile it with the Intel C++ compiler with SSE instructions enabled, the resulting code was slower than without them. There have been many new version of the compiler so it may hande SSE code better now, however judging by other projects I've been working on, SSE makes very little difference in speed (usually the code is a little faster, but there have been cases when it's actually slower). In addition, SSE instructions reduce the precision of arithemtic operations - everything is computed in 32 bits instead of the normal 80.

Best regards,
Vlado

[Psy]

what about sse2 or 3dnow?

[Brim]

Currently VRay doesn't use SSE instructions for compatibility reasons. The last time I tried to compile it with the Intel C++ compiler with SSE instructions enabled, the resulting code was slower than without them. There have been many new version of the compiler so it may hande SSE code better now, however judging by other projects I've been working on, SSE makes very little difference in speed (usually the code is a little faster, but there have been cases when it's actually slower). In addition, SSE instructions reduce the precision of arithemtic operations - everything is computed in 32 bits instead of the normal 80.

Best regards,
Vlado

Good to know Vlado. Thanks for responding.