Intel's new Nehalem-based Core i7 desktop processor is nippy. And now we can exclusively reveal that the upcoming dual-socket server variant of the Nehalem architecture is so fast it's almost silly.
We know this because we're the first in the world to test it.
Unsurprisingly, our time with this top secret rig was rather limited. But we have managed to squeeze in a single full run of arguably the most important test of processor performance – SPECfp_rate_base2006. Full details of the testing scenario can be found here.
We got our filthy paws on a fully operational dual-socket Nehalem EP platform for a spot of guerrilla benchmarking. Our test system was trimmed out with a pair of 2.8GHz Nehalem EP chips, likely to be sold as Xeon X5560 CPUs when Nehalem EP launches in the first quarter of 2009.
For the record, the system is based on essentially the same Tylersburg chipset as used with the first "Bloomfield" Core i7 desktop chips. For our testing it was configured with 24GB of 1,066MHz DDR3 memory.
The really important figure for Intel here is the base floating point metric. That's a test existing Xeons struggle with largely thanks to the weaknesses of their ancient front side bus and discreet memory controller architecture.
Epic scaling
All of that is gone with Nehalem EP, replaced with the sleek new Quick Path Interconnect and on-die memory controllers. The result is truly epic bandwidth and performance scaling. The big number many will be waiting for is the SPECfp_rate_base2006 base rate. So here it is: 160. (Performance estimated using preliminary hardware provided by Intel; for details of testing see this link.)
To put that into context, the Intel's current Penryn-based Xeon dual-socket platform fails to hit 90, even running at 3.4GHz. Perhaps even more significantly, AMD's shiny new 45nm Shanghai (Opteron 8384) chips in dual-socket 2.7GHz trim score just 105 points. Even a four-socket Shanghai rig is only good for just under 190. (Competitive data retrieved from www.spec.org on 24 November 2008.)
And remember, Nehalem EP's 160 point score is for a pre-production system running at 2.8GHz. Models humming a 3.2GHz tune will be available when the chip launches early next year. What's more, our testing was carried out on Windows Server 2008.
Run the same benchmark courtesy of the leaner, meaner Linux OS and you'd be looking at an even higher score.
In other words, as good as AMD's new Shanghai chip is, it appears that Nehalem EP will slap it around with a wet fish.
Just to confirm that our eyes did not deceive us, we also gave Nehalem EP a quick going over with the Stars Euler3D benchmark. It's a computational fluid dynamics simulation that majors on floating point performance.
Sure enough, Nehalem EP roasts all comers in this benchmark, too – it's twice as quick as a pair of 2.7GHz Shanghai processors (14.34 seconds to complete five instances versus 30.32 seconds).
Oh, and the platform's memory bandwidth is borderline biblical, clocking up 35GB/s in SiSoft Sandra's quick and dirty bandwidth benchmark.
