Advanced search

Message boards : Number crunching : Unified memory by AMD. CPU + GPU.

Author Message
Profile Zarck
Send message
Joined: 16 Aug 08
Posts: 145
Credit: 328,473,995
RAC: 0
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 35570 - Posted: 9 Mar 2014 | 12:57:01 UTC

http://developer.amd.com/wordpress/media/2012/10/hsa10.pdf

The subject seems timely for everyone,

Le sujet semble d'actualité pour tout le monde,

@+
*_*
____________

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 35579 - Posted: 9 Mar 2014 | 21:01:27 UTC - in response to Message 35570.

In a few years, when the software is mature and the hardware is widely available, this could turn into a revolution for code which can not currently run on GPUs efficiently due to portions stil needed to be run on the CPU. This applies to almost all BOINC projects, except the ones which already work well with GPUs. It could, however, lead to new physical algorithms becoming possible with GPU acceleration or "just" faster ways to compute the same physical algorithms.

In any way, the code written for HSA will not even work on current AMD GPUs, so don't expect anyone to happily develop yet another code path for what is still a tiny amount of hardware in user hands. It's different if you can choose your hardware yourself, i.e. writing some software to be run on your own cluster.

MrS
____________
Scanning for our furry friends since Jan 2002

Post to thread

Message boards : Number crunching : Unified memory by AMD. CPU + GPU.

//