Groups > comp.programming > #3984 > unrolled thread

Many-cores hit the memory wall

Started by	aminer <aminer@toto.net>
First post	2013-10-28 14:14 -0700
Last post	2013-11-03 21:05 -0800
Articles	3 — 2 participants

Back to article view | Back to comp.programming

  Many-cores hit the memory wall aminer <aminer@toto.net> - 2013-10-28 14:14 -0700
    Re: Many-cores hit the memory wall aminer <aminer@toto.net> - 2013-10-28 14:44 -0700
      Re: Many-cores hit the memory wall Robert Miles <robertmilesxyz@gmail.com> - 2013-11-03 21:05 -0800

#3984 — Many-cores hit the memory wall

From	aminer <aminer@toto.net>
Date	2013-10-28 14:14 -0700
Subject	Many-cores hit the memory wall
Message-ID	<l4m9h7$8mr$3@news.albasani.net>

Many-cores hit the memory wall

http://storagemojo.com/2008/12/08/many-cores-hit-the-memory-wall/


Thank you,
Amine Moulay Ramdane.

[toc] | [next] | [standalone]

#3985

From	aminer <aminer@toto.net>
Date	2013-10-28 14:44 -0700
Message-ID	<l4mb9s$cpg$2@news.albasani.net>
In reply to	#3984

"The performance is especially bad for informatics applications
—data-intensive programs that are increasingly crucial to the labs’
national security function."

http://spectrum.ieee.org/computing/hardware/multicore-is-bad-news-for-supercomputers

This is what i have said about memory bound applications and/or disk
bound applications such as database engines etc.


Amine Moulay Ramdane.



On 10/28/2013 2:14 PM, aminer wrote:
>
> Many-cores hit the memory wall
>
> http://storagemojo.com/2008/12/08/many-cores-hit-the-memory-wall/
>
>
> Thank you,
> Amine Moulay Ramdane.

[toc] | [prev] | [next] | [standalone]

#3994

From	Robert Miles <robertmilesxyz@gmail.com>
Date	2013-11-03 21:05 -0800
Message-ID	<36f153cd-31e9-4f31-b797-a5ca8c06413f@googlegroups.com>
In reply to	#3985

On Monday, October 28, 2013 1:44:13 PM UTC-5, aminer wrote:
> "The performance is especially bad for informatics applications
> —data-intensive programs that are increasingly crucial to the labs’
> national security function."

One possibility:  Give each core its own dedicated section of memory, with access to the memory of the other cores restricted and expected to be slow.

Intel's hyperthreading takes advantage of cores about twice as fast as the memory by giving every core two sets of registers, so it can give almost the same performance as two individual cores.

[toc] | [prev] | [standalone]

csiph-web

Many-cores hit the memory wall

Contents

#3984 — Many-cores hit the memory wall

#3985

#3994