corona renderer and Xeon Phi support

[prorender]

Dear Mr. Karlik,

do you see a chance to develop also a Xeon Phi version of your corona renderer?

Regards
Daniel Krüger

[Ondra]

hi,
certainly not for the current generation of Phi as it would not support 3dsmax lights and textures

[prorender]

Ok,

but if these limitations won't exist further how difficult is it compile your code to Xeon Phi? Think of the idea "corona renderer - first Xeon Phi based renderer". This would be a powerful tool on the market.

Regards
Daniel Krüger

[Ondra]

I don't know, I haven't researched much into the Xeon Phi. But from what I see, it has different memory space. Which means no unported code can be supported (standard lights, standard texmaps, ...), and every part of Corona would have to be ported. Plus the architecture is very different from desktop CPUs (it might be x86, but it employs different methods to get high performance), so basically Phi is in principle similar to CUDA GPU. But I heard they'll do next version with stared memory space, maybe then it will be feasible.

[prorender]

Let's wait and see how the development will advance. At the moment there is no known renderer on the market which is developped for co processor Xeon Phi. Perhaps a good chance for your futher development.

Regards
Daniel Krüger

[Juraj]

Jeez, if all "rumours" be true then, next year will be interesting. Both Phi and nVidia Maxwell should feature shared memory. But this has been promised few times before so who knows.

[J4CO3]

Hello,
One year later, does anyone know if Xeon Phi version of Corona can be expected any time soon or at all?

Thank you

[johan belmans]

https://forum.corona-renderer.com/index.php/topic,5859.msg40370.html#msg40370

[SaY]

Any updates on Phi support?
Newly released Knight Landing is basically a bunch of Atom cores, how difficult would it be to adapt Corona to use it?

[SHD]

yep, and will it make sense?

[Ondra]

sorry, it still seems we can speed up Corona more by focusing on optimizing the existing CPU code (think things like denoising)

[FrostKiwi]

yep, and will it make sense?

You can Copy paste Corona's code in the xeon phi Compiler, but that doesnt make up for the hours upon hours of actually optimizing the code to run in there.

Knights Landing will support only 8-16 GB of ram per core. We don't have clear numbers, but the 384gb collective ram may or may not be too slow, if each of the 72 cores starts pulling from it large data sets like in the case of rendering.

By supporting Xeon Phi Corona would tap into the very same pitfalls of GPU rendering, it tried to get away from.

-Limited Memory
-need for out of core shenanigans / sharing, even if it already is part of the architecture
-barely developed for outside of scientific research and data centers
-ultra slow development cycle, optimisations would delay features

[mandrake0]

just as a info. the Xeon Phi supports Windows Server! the functionality allows to run standard application with no modification of code.

if you are using TBB you should run it on the xeon phi. just the difference is the amount of cores and faster memory.

Article about Xeon Phi: http://www.nextplatform.com/2015/03/25/more-knights-landing-xeon-phi-secrets-unveiled/