Intel Larrabee physics white paper

[thilanliyan]

http://isdlibrary.intel-dispat...csOnLarrabee_paper.pdf

 

[Idontcare]

I just saw this over on XS...am reading it now.

Hey Nemesis - notice how many Russian co-author's are on this whitepaper? Keeps fitting together as you said.

 

[OCGuy]

Originally posted by: Idontcare


Hey Nemesis - notice how many Russian co-author's are on this whitepaper? Keeps fitting together as you said.

Dont egg him on! :laugh:

 

[Nemesis 1]

Would they be Russian Jews than lol ' Ya its amazing Compilers has become the big thing hasn't it.

 

[BenSkywalker]

Ya its amazing Compilers has become the big thing hasn't it.

This white paper shows you what to do to manually extract parallelism and code for optimal results on Larry, because the compiler can't do it itself. This white paper is actually direct proof from Intel that you have been very wrong Nemesis, they are relying on coders writing code, not chimps banging on a keyboard and a compiler will do something magical

 

[SunnyD]

Originally posted by: BenSkywalker

Ya its amazing Compilers has become the big thing hasn't it.

This white paper shows you what to do to manually extract parallelism and code for optimal results on Larry, because the compiler can't do it itself. This white paper is actually direct proof from Intel that you have been very wrong Nemesis, they are relying on coders writing code, not chimps banging on a keyboard and a compiler will do something magical

Any programmer knows that no matter what compiler you use (including an EPIC compiler), the developer still has to make at least some moderate effort to design the code to be effective in a parallel environment. I think compilers will slowly get better over time with this, but will always needs a dev with a clue in order to make it go.

 

[BenSkywalker]

Any programmer knows that no matter what compiler you use (including an EPIC compiler), the developer still has to make at least some moderate effort to design the code to be effective in a parallel environment. I think compilers will slowly get better over time with this, but will always needs a dev with a clue in order to make it go.

Of course, but Nemesis is in rather extreme denial about this. Even this document which explains how you need to manually alter your code approach he reads as evidence that compilers are somehow going to make the architecture(and this isn't even EPIC....).

 

[SunnyD]

Originally posted by: BenSkywalker

Any programmer knows that no matter what compiler you use (including an EPIC compiler), the developer still has to make at least some moderate effort to design the code to be effective in a parallel environment. I think compilers will slowly get better over time with this, but will always needs a dev with a clue in order to make it go.

Of course, but Nemesis is in rather extreme denial about this. Even this document which explains how you need to manually alter your code approach he reads as evidence that compilers are somehow going to make the architecture(and this isn't even EPIC....).

Ah, I see what you're getting at now.

In some ways he is (a little) correct though. For example, SSE wasn't implicitly compiled until very recently and even so only for a few specific relatively common cases that use it. In almost all other applications that support it, it requires explicit use on the developer's part. (This blurb was meant for Nemesis btw)

While it would be nice to see a compiler become more parallel in nature, unfortunately the gains to be made with automatic code profiling would be minimal compared to something specifically hand tuned. You want a perfect example of this even in today's serial world? Go look at the demo scene. Nearly every single thing they do is completely hand tuned, and there's very little parallelism going on in those demos.