CHIPMAKER Intel's Core i7 3960X, the first Socket 2011 processor, is a Sandy Bridge-based six-core CPU that right now represents the top of Intel's desktop CPU portfolio.
We found that the 3.3GHz processor has pretty good overclocking capabilities, and despite only three-fourths of the eight cores actually present on the die being enabled, the six-core, 12 thread combination using hyperthreading, or simultaneous multithreading, is quite sufficient for most current desktops.
The full eight-core die rollout is reserved for the server and workstation parts that are due for launch a month from now, so while the die has eight cores, all the chips Intel sells have only six cores enabled at present.
How is the base performance affected by the hyperthreading, sharing two threads simultaneously on a single core? In principle, if the applications are composed of threads heavily using the core resources, it's slightly better to turn it off. On the other hand, if the typical threads are somewhat lighter, hyperthreading does help boost total CPU performance in real usage situations. If the CPU has high enough performance, the difference will be even more obvious. Since it's an often overlooked aspect when benchmarking, why not have a look?
Here we run the Intel Core i7 3960X on one of the best X79 chipset mainboards in existence, the Asus Rampage IV Extreme, an extended ATX size mainboard from Asus' Republic of Gamers (ROG) series optimised for heavy overclocking and games. Besides the high end components for high power and clock speeds, the R4E, as it's also known, has a multitude of USB3, SATA3 and other new interfaces, plus Quad SLI and Quadfire capability.
The new UEFI BIOS with GUI supports fine-tuned overclocking, plus a couple of pre-set modes. The fastest of these modes, the 4.25GHz Turbo all-core automatic setting, was the one we chose. Intel's standard compact sealed liquid cooling kit made by Asetek seems more than good enough to sustain this speed setting.
Under the standard Windows 7 SP1 64-bit OS setup, we ran Sandra 2012 and CineBench 11.5, both tests supporting a high number of threads, to see the difference in performance and per-core scaling while enabling and disabling hyperthreading. In Sandra, we ran all key CPU integer, floating point, multimedia and inter-core tests, while in CineBench we ran multithreaded ray tracing.
The memory used was an A-DATA set of four 4GB DIMMs, running at DDR3-1666 CL9 settings. To eliminate dependence on the GPU - either in performance or power usage - we used the entry level AMD Radeon HD 6570 graphics card during these CPU focused benchmarks.
Next page: Benchmark results