Since BD cores are rather "cheap" in terms of total die area per CPU,I think that AM3+ socket will ,in time, get >8 core models on the same 32nm process node. Plus,all chances are that second iteration of Bulldozer chips,made on 22nm node,will work in AM3+ socket since all the electrical specs are in places(per module power gating,TurboCore 2 etc.) .And the 22nm version will definitely have much more than 8 cores.
As for clock speed,I wouldn't be surprised if we saw a 4Ghz 8 core Orochi next year.That's roughly 20% increase Vs Thuban on 45nm (which wasn't a high clock speed design to begin with,while BD is ).
So to sum it up,my expectation for Orochi(8 core AM3+) are: 20% higher per core performance Vs Thuban (10-35% range with 20% averaged ), 20-25% higher clock speed targets,33% more cores and 50-70% better performance Vs Thuban. Single thread performance,being a function of IPC and clocks can be as high as ~1.5x over Thuban,since we would have aggressive default clock and new Turbo,giving ~1.25x the clock effectively Vs Thuban and better per core perf.(my est. 1.2x on average),thus 1.2*1.25~=1.5x.
Multi core performance estimate : 3.3Ghz Thuban ~ 3.3*6*1=19.8 (clock * core count * relative perf. level) ; 4Ghz Orochi ~ 4*8*1.09=34.88 ( clock * core count * relative perf. level with counted scaling penalty via sharing the front end ). 34.88/19.8=1.76x ,or if the 4Ghz is a bit on the generous side for the Orochi,then with the 3.6Ghz as a def. clock we get a 1.6x throughput improvement with 33% more cores .
Bookmarks