By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), December 6, 2007 12:53 am
Room: Moderated Discussions
DaveC (davecuny@gmail.com) on 12/5/07 wrote:
---------------------------
>That would make sense, but when has Sun's MPU group ever made complete sense? :-p
>However, you're not really "shrinking" N3 if you ADD cores.
It depends on the cores. If they drop the many-thread-per-core approach they can cut on the multiple contexts every core has to support (register files, etc...) and also share L1s like Rock is supposed to do.
>Thing is, they have a adequately-performing core that's engineered for 8 threads
>per core, so why abandon the many thread idea for dual thread? Why not just beef up the cores?
Higher single-threaded performance I guess. If Rock cores are able to yield higher single-threaded performance and the very high resource usage levels of Niagara it would be certainly a good idea to reuse them in the low end.
>My guess is that N3 is very similar to Rock but w/out the scout threads and SMP.
Rock cores w/o scout threading don't make much sense IMHO as they wouldn't have any way to absorb stalls, however I guess you might be right on the SMP though. They might also be 'cut down' versions. Rock is supposed to be 4-way superscalar, corresponding T3 cores might be 2-way (like the current ones which are sort-of-2-way-but-not-really-superscalar).
Gabriele
PS: All of the above is just pure, unfounded, wild speculation on my part so take it for what it's worth
---------------------------
>That would make sense, but when has Sun's MPU group ever made complete sense? :-p
>However, you're not really "shrinking" N3 if you ADD cores.
It depends on the cores. If they drop the many-thread-per-core approach they can cut on the multiple contexts every core has to support (register files, etc...) and also share L1s like Rock is supposed to do.
>Thing is, they have a adequately-performing core that's engineered for 8 threads
>per core, so why abandon the many thread idea for dual thread? Why not just beef up the cores?
Higher single-threaded performance I guess. If Rock cores are able to yield higher single-threaded performance and the very high resource usage levels of Niagara it would be certainly a good idea to reuse them in the low end.
>My guess is that N3 is very similar to Rock but w/out the scout threads and SMP.
Rock cores w/o scout threading don't make much sense IMHO as they wouldn't have any way to absorb stalls, however I guess you might be right on the SMP though. They might also be 'cut down' versions. Rock is supposed to be 4-way superscalar, corresponding T3 cores might be 2-way (like the current ones which are sort-of-2-way-but-not-really-superscalar).
Gabriele
PS: All of the above is just pure, unfounded, wild speculation on my part so take it for what it's worth
Topic | Posted By | Date |
---|---|---|
ISSCC 2008 Preview online | David Kanter | 2007/12/05 05:28 AM |
ISSCC 2008 Preview online | DaveC | 2007/12/05 11:35 AM |
ISSCC 2008 Preview online | Foo_ | 2007/12/05 01:34 PM |
ISSCC 2008 Preview online | David Kanter | 2007/12/05 01:40 PM |
ISSCC 2008 Preview online | DaveC | 2007/12/05 01:51 PM |
ISSCC 2008 Preview online | Gabriele Svelto | 2007/12/06 12:53 AM |
ISSCC 2008 Preview online | savantu | 2007/12/05 03:03 PM |
ISSCC 2008 Preview online | Gabriele Svelto | 2007/12/06 12:44 AM |
ISSCC 2008 Preview online | Michael S | 2007/12/06 08:42 AM |
ISSCC 2008 Preview online | Gabriele Svelto | 2007/12/06 09:09 AM |
ISSCC 2008 Preview online | anonymous | 2007/12/06 12:25 PM |
ISSCC 2008 Preview online | Michael S | 2007/12/06 02:20 PM |
ISSCC 2008 Preview online | Aaron Spink | 2007/12/06 07:36 AM |
ISSCC 2008 Preview online | Wes Felter | 2007/12/05 01:53 PM |
ISSCC 2008 Preview online | mas | 2007/12/05 03:42 PM |
ISSCC 2008 Preview online | Anonymous | 2007/12/07 12:40 AM |
ISSCC 2008 Preview online | Dan Downs | 2007/12/07 03:28 PM |
ISSCC 2008 Preview online | IntelUser2000 | 2007/12/09 12:58 AM |
Silverthorne | anon | 2007/12/09 01:40 AM |
Silverthorne | IntelUser2000 | 2007/12/09 03:09 AM |
Silverthorne | David Kanter | 2007/12/09 06:02 PM |
Silverthorne | Ian Ameline | 2007/12/10 01:48 PM |
Silverthorne | IntelUser2000 | 2008/01/09 12:03 AM |
Silverthorne | David Kanter | 2007/12/09 02:41 AM |
Silverthorne | anonymous | 2007/12/09 02:53 AM |