By: Maynard Handley (name99.delete@this.name99.org), October 23, 2014 2:07 pm
Room: Moderated Discussions
Mr. Camel (a.delete@this.b.c) on October 23, 2014 1:28 pm wrote:
> Symmetry (someone.delete@this.somewhere.com) on October 23, 2014 12:54 pm wrote:
> > Anyone have any thoughts on this press release? Apparently they just came out of
> > stealth mode. Mostly it's marketing but I'm hoping someone can tie it to a research
> > paper that can give more details about what the heck they're really doing.
> >
> > http://www.softmachines.com/wp-content/uploads/2014/10/SoftMachinesLaunchPR.pdf
>
> Could this be the first commercial implementation of speculative multithreading?
>
Speculative anything doesn't really get you "breakthroughs in performance-per-watt".
To me, it all sounds like an implementation of the Vector Thread architecture that's been talked up at Berkeley:
http://scale.eecs.berkeley.edu/papers/vta-isca2004-slides.ppt
But what they're saying is vague enough that it could be any of a dozen other things. The description also matches Apple's macroscalar patents (dynamically figure out independent loop iterations and run them simultaneously), and a variety of ideas that have spun out of the KIP stuff, which all boil down to aggressively figuring out streams of dependent instructions and splitting them off from other (independent) streams of dependent instructions. I guess we'll know more in a few hours.
Interesting that AMD is an investor. (Frankly I'm a lot more optimistic that they will do something with this research than I am about Samsung.)
> Symmetry (someone.delete@this.somewhere.com) on October 23, 2014 12:54 pm wrote:
> > Anyone have any thoughts on this press release? Apparently they just came out of
> > stealth mode. Mostly it's marketing but I'm hoping someone can tie it to a research
> > paper that can give more details about what the heck they're really doing.
> >
> > http://www.softmachines.com/wp-content/uploads/2014/10/SoftMachinesLaunchPR.pdf
>
> Could this be the first commercial implementation of speculative multithreading?
>
Speculative anything doesn't really get you "breakthroughs in performance-per-watt".
To me, it all sounds like an implementation of the Vector Thread architecture that's been talked up at Berkeley:
http://scale.eecs.berkeley.edu/papers/vta-isca2004-slides.ppt
But what they're saying is vague enough that it could be any of a dozen other things. The description also matches Apple's macroscalar patents (dynamically figure out independent loop iterations and run them simultaneously), and a variety of ideas that have spun out of the KIP stuff, which all boil down to aggressively figuring out streams of dependent instructions and splitting them off from other (independent) streams of dependent instructions. I guess we'll know more in a few hours.
Interesting that AMD is an investor. (Frankly I'm a lot more optimistic that they will do something with this research than I am about Samsung.)