Apparently Intel wants to remove the need for programmers to implement multiprocessor support for applications. According to this article posted June 18, 2007, Intel is developing an Exoskeleton technology which tricks the OS into thinking the multicore chips are actually just single processors.
So let's say you have two 16-core chips, Intel's new technology would fool the OS into thinking you have just one core while benefiting from all 32 cores.
Seems to me like this technology, if it works, would be extremely helpful for chess engines, anyone agree? No longer would a programmer need to optimise the code for multiprocessor support, the engine would get 100% benefit from all 32 cores, etc. Here's the article:
http://www.tgdaily.com/content/view/32503/115/
Intel to remove the need for mutiprocessor support?
Moderators: hgm, Rebel, chrisw
-
- Posts: 2071
- Joined: Thu May 04, 2006 3:40 am
- Location: Dune
Re: Intel to remove the need for mutiprocessor support?
I don't think so. I think it would just make it easier to add some sort of multiprocessing.Leto wrote:Apparently Intel wants to remove the need for programmers to implement multiprocessor support for applications. According to this article posted June 18, 2007, Intel is developing an Exoskeleton technology which tricks the OS into thinking the multicore chips are actually just single processors.
So let's say you have two 16-core chips, Intel's new technology would fool the OS into thinking you have just one core while benefiting from all 32 cores.
Seems to me like this technology, if it works, would be extremely helpful for chess engines, anyone agree? No longer would a programmer need to optimise the code for multiprocessor support, the engine would get 100% benefit from all 32 cores, etc. Here's the article:
http://www.tgdaily.com/content/view/32503/115/
To me it sound more like an automated OpenMP, wich would then mainly be used to fe unroll loops over several processors.
Nothing close (performance wise) to splitting your own tree.
Tony
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Intel to remove the need for mutiprocessor support?
It won't help a chess engine at all, unless the engine is parallel. This idea is not related to parallel search at all. It is an attempt to integrate lots of "things" (cpu core, graphics core, floating point core (accelerator) and such) into a unified "look". It won't help make a serial chess program use multiple CPU cores at all...Leto wrote:Apparently Intel wants to remove the need for programmers to implement multiprocessor support for applications. According to this article posted June 18, 2007, Intel is developing an Exoskeleton technology which tricks the OS into thinking the multicore chips are actually just single processors.
So let's say you have two 16-core chips, Intel's new technology would fool the OS into thinking you have just one core while benefiting from all 32 cores.
Seems to me like this technology, if it works, would be extremely helpful for chess engines, anyone agree? No longer would a programmer need to optimise the code for multiprocessor support, the engine would get 100% benefit from all 32 cores, etc. Here's the article:
http://www.tgdaily.com/content/view/32503/115/
-
- Posts: 4889
- Joined: Thu Mar 09, 2006 6:34 am
- Location: Pen Argyl, Pennsylvania
Re: Intel to remove the need for mutiprocessor support?
Leto wrote:Apparently Intel wants to remove the need for programmers to implement multiprocessor support for applications. According to this article posted June 18, 2007, Intel is developing an Exoskeleton technology which tricks the OS into thinking the multicore chips are actually just single processors.
So let's say you have two 16-core chips, Intel's new technology would fool the OS into thinking you have just one core while benefiting from all 32 cores.
Seems to me like this technology, if it works, would be extremely helpful for chess engines, anyone agree? No longer would a programmer need to optimise the code for multiprocessor support, the engine would get 100% benefit from all 32 cores, etc. Here's the article:
http://www.tgdaily.com/content/view/32503/115/
If it works anything OpenMP. so far I hsve seen limited returns for most chess engines , although I suspect that could be improved.
smp reagrds,
Michael B
-
- Posts: 12538
- Joined: Wed Mar 08, 2006 8:57 pm
- Location: Redmond, WA USA
Re: Intel to remove the need for mutiprocessor support?
I guess that you will get 30% of the processor power if it is taken out of your hands. That is fine if you want your program to coexist nicely with lots of other processes but a problem if you want the last ounce of strength.