By: Patrick Chase (patrickjchase.delete@this.gmail.com), January 8, 2015 7:02 pm
Room: Moderated Discussions
juanrga (nospam.delete@this.juanrga.com) on January 7, 2015 6:03 am wrote:
> Eric Bron (eric.bron.delete@this.zvisuel.privatefortest.com) on January 3, 2015 2:42 pm wrote:
> > thank you for the link, he was saying "going away" not "abandoned, which is factually
> > true speaking of the near future evolution of MIC, i.e. roughly same number of cores
> > in 14 nm KNL than 22 nm KNC, whatever your favorite marketing material says about it
> >
> > as others have already tried to explain you, it's the opposite to what was "predicted"
> > in a series of papers you have posted on this very forum saying that the future is with
> > more simpler cores
>
> "Abandonment" is a synonym of "going away".
No, it is not. Like many phrases in human language "going away" has a range of possible meanings, and you have to look at context to understand which the writer intended.
In general "abandonment" connotes completely giving up on something, while "going away" connotes a shift of emphasis or direction and is therefore more subtle.
That's exactly what Linus sought to convey when he pointed out that Intel is adding per-core complexity while holding core count constant, rather than adding cores while holding per-core complexity constant. In that sense they're moving in the opposite direction from what you and other manycore advocates have suggested, hence "going away". They are absolutely not "abandoning" manycore, rather the balance a bit.
> My point is that future applications will require the main processor to be a
> hundred/thousand manycore because multicores don't scale up and cannot provide the
> needed throughput --the goal is to obtain 50x more performance by about 2020--.
IMO that's an unrealistic goal as it corresponds to a speedup of about 2.2X per year. That never happened even in the heyday of Moore's Law. NO architecture will be able to provide that, so the fact that multicore can't doesn't prove anything.
> I asked Linus twice how he proposes to reach the target goals
> and he twice avoided the question, he only writes ridiculous rants against manycores.
No, he's pointing out that your goal is impossible via manycore as well, so it doesn't matter that multicore can't get us there. It's impossible, period.
> Reaching this goal will require novel hardware and software techniques now under
> development. Recently Intel's Crago and coworkers have published a set of latency
> techniques "to reach the energy efficiency goals of future 1000-core data-parallel
> processors" [1]. They evaluated and rejected a number of existent latency
> techniques, including Linus beloved OoO.
Let's get one thing out of the way: NOBODY (except maybe Linus) thinks OoO is an elegant solution. It's flagrantly wasteful on several levels, and architects have been desperately trying to figure out how not to use it ever since Tomasulo (1967). If you could care to pick a year, I'm sure I can find papers just like the one you cite above predicting the demise of OoO.
> Eric Bron (eric.bron.delete@this.zvisuel.privatefortest.com) on January 3, 2015 2:42 pm wrote:
> > thank you for the link, he was saying "going away" not "abandoned, which is factually
> > true speaking of the near future evolution of MIC, i.e. roughly same number of cores
> > in 14 nm KNL than 22 nm KNC, whatever your favorite marketing material says about it
> >
> > as others have already tried to explain you, it's the opposite to what was "predicted"
> > in a series of papers you have posted on this very forum saying that the future is with
> > more simpler cores
>
> "Abandonment" is a synonym of "going away".
No, it is not. Like many phrases in human language "going away" has a range of possible meanings, and you have to look at context to understand which the writer intended.
In general "abandonment" connotes completely giving up on something, while "going away" connotes a shift of emphasis or direction and is therefore more subtle.
That's exactly what Linus sought to convey when he pointed out that Intel is adding per-core complexity while holding core count constant, rather than adding cores while holding per-core complexity constant. In that sense they're moving in the opposite direction from what you and other manycore advocates have suggested, hence "going away". They are absolutely not "abandoning" manycore, rather the balance a bit.
> My point is that future applications will require the main processor to be a
> hundred/thousand manycore because multicores don't scale up and cannot provide the
> needed throughput --the goal is to obtain 50x more performance by about 2020--.
IMO that's an unrealistic goal as it corresponds to a speedup of about 2.2X per year. That never happened even in the heyday of Moore's Law. NO architecture will be able to provide that, so the fact that multicore can't doesn't prove anything.
> I asked Linus twice how he proposes to reach the target goals
> and he twice avoided the question, he only writes ridiculous rants against manycores.
No, he's pointing out that your goal is impossible via manycore as well, so it doesn't matter that multicore can't get us there. It's impossible, period.
> Reaching this goal will require novel hardware and software techniques now under
> development. Recently Intel's Crago and coworkers have published a set of latency
> techniques "to reach the energy efficiency goals of future 1000-core data-parallel
> processors" [1]. They evaluated and rejected a number of existent latency
> techniques, including Linus beloved OoO.
Let's get one thing out of the way: NOBODY (except maybe Linus) thinks OoO is an elegant solution. It's flagrantly wasteful on several levels, and architects have been desperately trying to figure out how not to use it ever since Tomasulo (1967). If you could care to pick a year, I'm sure I can find papers just like the one you cite above predicting the demise of OoO.