The guts behind Skulltrail

CPU: Yorkfield Core 2 Extreme QX9775

The CPU behind Intel’s Skulltrail platform is the Core 2 Extreme QX9775. This Yorkfield-based quad-core CPU runs at 3.2GHz, boasts 12MB of L2 cache, 1600 MHz FSB, with a TDP of 150W. Clock-for-clock, the new core is 5-10% faster, but the addition of the extra L2 cache (that’s also faster) and SSE4.1 instructions means that real-world performance can be even faster. But this CPU is more than just a die shrink. It is the culmination of Intel’s most sophisticated engineering resources. Hyperbole? Not really. As an example of Intel’s innovation, these CPUs aren’t even made with silicon. They’re made with Hafnium. This advance in material science allows Intel to cut gate leakage by an order of magnitude while simultaneously reducing the switching energy. It’s faster and cooler.

Yorkfield is a dual-die quad core design with 6MB of L2 cache per die (12MB total), a FSB of 1600 MHz, and approximately 820 million transistors. The Extreme Edition processor has an 8x multiplier for a core clock of 3.2GHz, but as with previous “Extreme Edition” processors, Intel has removed the multiplier lock, allowing users complete oveclocking flexibility. With hand-selected CPUs, the Core 2 Extreme CPUs have maintained excellent overclockability: phase-change cooling has reached 5GHz, water cooling has reached 4.4GHz, and air cooled CPUs have reached 4.0 GHz. Alienware has shown off a 4GHz water-cooled system suggesting a comfortable performance envelope with production CPUs.

Yorkfield is more than just a die shrink of Clovertown. There have been several new features. One of these is a Radix-16 divider unit. As you know, computers operate in binary. The easiest way to divide a binary number is to work one-bit at a time by doing a series of subtractions, tests, and shifts. You can do this by hand but this approach is time consuming and slow. That’s Radix 2. Modern CPUs have used a radix-4 divider in which 2 bits are calculated at any time. Since you’re looking at two bits at any given time, you only need half the number of iterations. To do this, you need bigger comparators for your tests, adding complexity to the chip design. The Radix-16 approach in Yorkfield allows 4 bits to be compared at a time.

This latest generation of CPU from Intel also features new SSE4.1 instructions. Applications such as Adobe Premiere Pro are already taking advantage of these features to provide significant improvements in matrix math and SIMD instructions.

While Intel uses a Xeon-type socket for the Skulltrail motherboard, the Core 2 Extreme CPUs feature desktop-optimized prefetchers and so it’s not a rebadged Xeon. That said, the Skulltrail motherboard is fully capable of supporting Xeon processors and so power-users on a budget looking for 8-core performance can use an older Clovertown or a lower-speed Tigerton-core CPU.

CPU Cooler: Zalman CNPS-9700

The coolers have to be mounted in a rotated fashion in order to avoid the beefy heatsinks on the CPU power regulators. We opted to direct our airflow to help cool the FB-DIMMs.

Memory: Micron 2x2GB DDR FB-DIMM @ 800MHz 5-5-5-18

Unfortunately, the Xeon package means that FB-DIMMS are required rather than standard memory modules. FB-DIMM is not an inherently flawed design and modern engineering allows FB-DIMMs to reach almost the same speeds as regular DIMMs. By going with a fully-buffered design, however, the Skulltrail platform will run more efficiently with larger memory capacities including 16GB and beyond. Peak bandwidth is rated at 28.4 GB/sec with all 4 DIMMS in use – thanks to support for 800MHz memory support however our real-world measurements in a two DIMM were less robust (about 7 GB/sec). The real consequence of FB-DIMMs is not performance, but cost.

At the moment, 800MHz FB-DIMMs are few and far between. Our DIMMs were clearly marked as engineering samples from Micron and Intel has listed Nanya as another source for FB-800 DIMMS. Future availability shouldn’t be a problem though as the Apple Mac Pro also uses 800MHz FB-DIMMs.