Will Vray get faster with Intel Sandy Bridge AVX Instructions?

[ultrasonic]

Intel has stated that Sandy Bridge processors will include a new set of instructions known as Advanced Vector Extensions (AVX).[16] These instructions are an advanced form of SSE. The data path is widened from 128 bits to 256 bits, the two-operand instruction limit is increased to three operands, and advanced data rearrangement functions are included. AVX is suited for floating-point-intensive applications.[17] New features include mask loads, data permutes, increased register efficiency and use of parallel loads, as well as smaller code size. When AVX instruction is used in conjunction with these improvements, it provides double peak FLOPS performance compared to using SSE4 instructions on CPUs. Sandy Bridge will also have a new extensible VEX opcode prefix.[

Hey Guys, i just read that and was wondering if vray would benefit from this new AVX stuff....even though i don't know exactly what this is all about I'm on linux so this should already be supported by the OS...

cheers
Oli

[vlado]

Not sure yet; we will need to compile V-Ray specifically for this instruction set and see if it is worth it. In the past, all in all, SSE2 did not bring any significant improvements, so I'm a bit skeptical. However there are a few parts in V-Ray RT where we directly use SSE instructions and this is where the new ones may be very useful.

Best regards,
Vlado

[renderfarmer]

I was bout to post the same question... So if SSE optimizations aren't beneficial to Vray, what types of improvements to CPU architecture are? What specifically should we be looking for in upcoming CPUs?

Has there been any talk of created a Maxon C4D-like standalone benchmark for Vray? Getting sites like Anandtech to run a universal Vray bench on new CPUs/GPUs as part of their evaluations would benefit everyone.

[vlado]

I was bout to post the same question... So if SSE optimizations aren't beneficial to Vray, what types of improvements to CPU architecture are? What specifically should we be looking for in upcoming CPUs?

Faster memory It's by far the greatest bottleneck right now.

Has there been any talk of created a Maxon C4D-like standalone benchmark for Vray?

There is one... it's been gathering dust for a while now, perhaps we should do something useful with it

Best regards,
Vlado

[renderfarmer]

Faster memory It's by far the greatest bottleneck right now.

By memory you mean system RAM or CPU on-board L2/L3 Cache? Please elaborate if you can as I think this is a big insight to the community.

There is one... it's been gathering dust for a while now, perhaps we should do something useful with it

Yes! Please, please, please, release it! With a slew of new CPUs out there and many studios and artists pondering spending thousands on upgrades this would be of great help to everyone.

[super gnu]

if memory bandwidth is your thing...


http://vr-zone.com/articles/ivy-brid...mms/15320.html

[renderfarmer]

Thanks for that link, Super Gnu.

I think Vlado was speaking in general terms, because after his post I ran some informal benchmarks comparing RAM running at 1333MHz and 2133MHz on the same system and saw no difference in rendering speed whatsoever.

RAM speed in my experience has been a non-factor with vray over the past 5 years, which is why I was so interested in a follow up to his comment.

[super gnu]

that is a very good point, as actually, having upgraded from 6 gb to 24 gb last year, i had to downclock my overclocked ram to keep stability, and i also didnt see a difference in rendering speed.