By: David Kanter (dkanter.delete@this.realworldtech.com), May 19, 2013 12:28 pm
Room: Moderated Discussions
Michael S (already5chosen.delete@this.yahoo.com) on May 19, 2013 6:30 am wrote:
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on May 19, 2013 6:00 am wrote:
> >
> > Anand explicitly mentioned ARM in his first sentence. However it's still a ridiculous statement when
> > you compare with Bobcat (and possibly VIA too).
>
> I'd say it's ridiculous when you compare to original P6. Slightly less so, when you compare to PM or K7.
> As to Bobcat, look at the date. AFAIK, Anand does not have time machine.
>
> > This is how small an Atom core is today. Remember
> > die size translates to cost, so this is why Atom never came out in quad-core variants, not even the
> > Centerton server parts (where quad-core would have made sense to try to compete with ARM).
> >
> > Wilco
>
> Not sure that quad-core Centerton would make sense. 8 threads is too much
> both for Saltwell's L2 cache and for its external memory interface.
> Similarly, I think that Cortex-A15 based SOC with 4 cores per cluster is not a good idea for servers.
> As to cost of silicon, I don't believe that in case of Centerton it even slightly mattered. 11 mm^2
> (the size of couple of Saltwell cores) is too little to be accounted for in a chip with >$50 ASP.
>
> Besides, quad-core would take non-trivial redesign - not something you want to do for a product that you launched
> not because you think that there is a meaningful market, but in order to keep Wall Street happy.
>
The problem with QC Centerton is that the FSB infrastructure wouldn't support it well.
David
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on May 19, 2013 6:00 am wrote:
> >
> > Anand explicitly mentioned ARM in his first sentence. However it's still a ridiculous statement when
> > you compare with Bobcat (and possibly VIA too).
>
> I'd say it's ridiculous when you compare to original P6. Slightly less so, when you compare to PM or K7.
> As to Bobcat, look at the date. AFAIK, Anand does not have time machine.
>
> > This is how small an Atom core is today. Remember
> > die size translates to cost, so this is why Atom never came out in quad-core variants, not even the
> > Centerton server parts (where quad-core would have made sense to try to compete with ARM).
> >
> > Wilco
>
> Not sure that quad-core Centerton would make sense. 8 threads is too much
> both for Saltwell's L2 cache and for its external memory interface.
> Similarly, I think that Cortex-A15 based SOC with 4 cores per cluster is not a good idea for servers.
> As to cost of silicon, I don't believe that in case of Centerton it even slightly mattered. 11 mm^2
> (the size of couple of Saltwell cores) is too little to be accounted for in a chip with >$50 ASP.
>
> Besides, quad-core would take non-trivial redesign - not something you want to do for a product that you launched
> not because you think that there is a meaningful market, but in order to keep Wall Street happy.
>
The problem with QC Centerton is that the FSB infrastructure wouldn't support it well.
David