Let's make some things clear here. 2 core intel CPUs with SMT have very aggressive Turbo and base clocks(yeah,"blame" intel's superior process tech). Apart from that, 4 "core" Trinity has 2 floating point units(2 threads capable each),same as 2 core+SMT intel solutions. So essentially,what you get is a rough parity in SSE/fp benchmarks if clocks were always equal (which they are not of course).
This is why 8 "core" bulldozer on desktop cannot beat 4 core SB/IB in floating point/SSE workloads. It essentially has the same number of floating point units and in workloads that are not FMA optimized it can only reach parity(or around that level roughly). That is AMD's choice,they targeted the design for server integer workloads leaving FP/SSE to heavy optimizations(FMA) to make up for reduced number of execution units per chip.
Numbers that Trinity shows at least give us clue about solid IPC boost for a "quick fix" (of around 5-10%) and a solid power draw reduction at similar clock as bdver1. If this gives us 4-4.6Ghz Vishera with 5-10% higher IPC then AMD may challenge 8T IB in quite a few workloads.
Reply With Quote
Bookmarks