Intel's Skulltrail supports HyperTransport links
NVIDIA IS REALLY happy with Skulltrail, or at least its chipset sales people are. Since Intel wanted SLI and, later, Quad SLI, the almighty Chipzilla had no choice but to bow to the current hotshot here. This bowing includes including two Nvidia chips on the Skulltrail motherboard.
These chips are no less than the fully-fledged Media Communications Processors (MCP) that were already used in previous Nvidia chipsets, like the 680i and 590, but with all of the chipset functionality disabled. The chips sit on the motherboard and do nothing but enable SLI and QuadSLI. So Skulltrail will not sport Nvidia's cool dual GbE controllers (or even Quad GbE, in this case), nor RAID5 features from Nvidia. All of the SouthBridge functionality remain with Intel's own chip.
The MCP chips communicate with each other through the HyperTransport interface, and serve as a quite good layout trial for Quick Path (FMACSI - Formerly Known As Common System Interface), the bi-directional interface of the future for Intel's Nehalem generation, since the company does not have a lot of experience outside GTL+ FSB (Front Side Bus) world. For communications between the MCPs and Intel's own chipset, the PCIe 2.0 interface is used.
We're not all that too sure what waves this will make in the world of latencies. We have seen on results of Nvidia's previous attempt - AMD's 4x4/QuadFX/FASN8 project for Athlon 64 FX/Agena FX series. That project had issues with latency between just two graphics cards, since the GPUs chips were not directly connected to the chipset, but rather to two different chips (680a chipset).
It will be interesting to see what will happen with SLI and QuadSLI performance-wise when the graphics cards are connected with the CPU through FSB, through the NorthBridge to the first MCP and then through HyperTransport from first to second MCP. µ
http://www.theinquirer.net/gb/inquirer/news/2007/09/26/intel-skulltrail-spots
[ 本帖最后由 4X4 于 2007-10-7 02:38 编辑 ] |