POV-Ray : Newsgroups : povray.general : NVidia Tesla vs. Intel Knight's Corner Server Time
26 Nov 2024 03:56:50 EST (-0500)
  NVidia Tesla vs. Intel Knight's Corner (Message 1 to 4 of 4)  
From: jhu
Subject: NVidia Tesla vs. Intel Knight's Corner
Date: 7 Dec 2011 19:10:00
Message: <web.4edfffd153724149d19b0ec40@news.povray.org>
Lots of processing elements on one chip. Knight's Corner-based products (which
supposedly contain 50 processing elements with 512-bit SIMD and run x86 code)
are still at least one year away. So, the question is: is it worth porting
Povray to CUDA or is it better to just wait until Knight's Corner comes out?


Post a reply to this message

From: Christian Froeschlin
Subject: Re: NVidia Tesla vs. Intel Knight's Corner
Date: 8 Dec 2011 20:09:43
Message: <4ee15fd7$1@news.povray.org>
jhu wrote:

> Lots of processing elements on one chip. Knight's Corner-based products (which
> supposedly contain 50 processing elements with 512-bit SIMD and run x86 code)
> are still at least one year away. So, the question is: is it worth porting
> Povray to CUDA or is it better to just wait until Knight's Corner comes out?

Well, I think purchasing the new super computer chip for home use
will be quite a bit further off in the future, while most people have
or can get their hands on a graphics card supporting CUDA. And of
course graphics cards increase in performance over time, too.

I also wonder whether this architecture is just "compatible with
x86 application" or actually gives automatic performance boosts to
x86 applications that are not especially optimized for it.


Post a reply to this message

From: jhu
Subject: Re: NVidia Tesla vs. Intel Knight's Corner
Date: 9 Dec 2011 01:25:06
Message: <web.4ee1a9394c4b86a3d19b0ec40@news.povray.org>
Christian Froeschlin <chr### [at] chrfrde> wrote:
> jhu wrote:
>
> > Lots of processing elements on one chip. Knight's Corner-based products (which
> > supposedly contain 50 processing elements with 512-bit SIMD and run x86 code)
> > are still at least one year away. So, the question is: is it worth porting
> > Povray to CUDA or is it better to just wait until Knight's Corner comes out?
>
> Well, I think purchasing the new super computer chip for home use
> will be quite a bit further off in the future, while most people have
> or can get their hands on a graphics card supporting CUDA. And of
> course graphics cards increase in performance over time, too.
>
> I also wonder whether this architecture is just "compatible with
> x86 application" or actually gives automatic performance boosts to
> x86 applications that are not especially optimized for it.

It should be feasible to port to CUDA now that it supports 64-bit float and
recursion. Whatever happened to that guy who was trying to port Povray to CUDA?
Did he have any working code?


Post a reply to this message

From: jhu
Subject: Re: NVidia Tesla vs. Intel Knight's Corner
Date: 14 Dec 2011 11:45:00
Message: <web.4ee8d25a4c4b86a3d19b0ec40@news.povray.org>
More info on Knight's Corner:

http://www.tomshardware.com/news/intel-knights-corner-mic-co-processor,14002.html

"it is fully accessible and programmable like fully functional HPC compute node,
visible to applications as though it was a computer that runs its own
Linux-based operating system independent of the host OS."

If that's true, no need for CUDA (which BTW, NVidia is releasing the source code
to). Knight's Corner is supposedly going to plug into LGA2011, so just get a
dual-processor board, plop in a Knight's Corner, and enjoy 200 threads of
awesome.


Post a reply to this message

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.