Early tests conducted with Intel's upcoming Nehalem-based Xeons by TechRadar today show the processor running approximately twice as fast as its current equivalent. Based on the same basic design as Core i7 mainstream chips, a pair of 2.8GHz quad-core Xeon X5560 chips received a SPEC score of 160 in floating-point math tests versus 90 for two 3.4GHz Xeons from the current generation.
The performance is faster than AMD's already-improved Shanghai-era 2.7GHz Opterons, which score 105 in a dual-processor setup, and compares more closely to a quad-Opteron setup's 190 rating in the same SPEC test. Intel is expected to launch 2.93GHz and 3.2GHz Xeons and so should see faster-still results from those systems.
The Xeon X5560s are even faster in a Stars Euler3D fluid dynamics test and complete their calculations in less than half the time, taking just over 14.3 seconds versus the 2.7GHz Opteron duo's 30.3 seconds.
Much of the increase stems from the key change in memory architecture to the platform, which moves the memory controller on to the Xeon's chip die and establishes a point-to-point architecture that lets the processor cores communicate more directly with peripherals. This reduces lag time, eliminates the need for a traditional front side bus and substantially increases the memory bandwidth up to 35GB per second, mostly eliminating a number of memory bottlenecks for some calculations.
Extra assists also come from Hyperthreading, which sometimes allows two threads of programming to execute on a single core, and general optimizations that let the cores process more information per clock individually and in parallel.
The formal launch of Nehalem technology is due in early 2009 and should find its way into updated workstations almost immediately afterwards, including those from Apple, Dell and HP.
No comments:
Post a Comment