Last week, Chinese site Expreview.com published a story about the new generation of nForce chipsets for AMD processors.
We managed to find more details through course of the weekend. For starters, the lineup will consist out of three (not two) chipsets with various capabilities.
MCP82-S1, MCP82-S2 and MCP82-S3 will round the lineup, all targeting their respective markets (high-end, mainstream and entry-level). The S1 and S2 will support SLI, while S3 targets lucrative OEM/ODM deals – our sources indicate that this variant will be pitched as a successor to GeForce 6150 line that conquered many Dells, HPs and Acers out there.
MCP82-S1 targets the high-end, with Hybrid-SLI technology attaching up to three graphics cards in 3-Way SLI (one card connected via x16 slot, two via x8 slots). The chipset features 35 PCIe lanes and 7 links. You can expect that a typical MCP82-S1 mainboard layout for will include three PCIe x16, two x1 and one PCI 3.0 slot. Or two PCI and single x1 slot.
MCP82-S2 supports 19 PCIe lanes and 4 links, so typical layout will be two PCIe x8 slots and two x1 slots. Or, single PCIe x16 and three x1 slots. As you can see, this board does not target the multi-GPU gamer, but rather offers compelling single-card experience. HybridSLI is supported, of course.
MCP82-S3 should be available only in micro-ATX form factor, offering a cut-down cost effective version of S2. These motherboards will target customers that are entering the world of computing. If Nvidia gets the price down, you will see this chipset even in netbooks. We feel that combination between next-generation Tegra and MCP82-S3 would be quite interesting for netbook market. Of course, if Microsoft allows licensing of Windows Mobile for a “cut-down notebook”.
Integrated GPU in all three chipsets is based between G92 and GT200 chips, offering improved performance in CUDA applications. I haven’t been able to find out will it bring 32 or 40 shader processors (thus, two or three FP64 DP units). mGPU features Dual-Link DVI, HDMI and analog VGA connectors. If Nvidia really wants us to connect our displays to their mGPU, they should put two Dual-Link DVIs or two DisplayPorts. Only then they can start thinking about removing one DVI connector from their GeForce cards, not before (more about that later).
Storage-wise, Nvidia continues with hardware RAID support – RAID 0, 0+1 and 5 are all supported through six SATA connectors. Hardware network chips will continue to offer GbE or paired GbE speeds through two RJ-45 connectors. Of course, if you activate paired GbE network, you get single RJ-45 with 2Gbps speed.
Memory-wise, MCP8 series supports AM2, AM2+ and AM3 processor sockets, so you can expect to find DIMM slots for either DDR2-800/1066 or DDR3-1066/1333/1600. SLI Memory will continue to be supported from the box and there is little doubt that new SLI-Ready memory will not appear in time for launch. If AMD scores a big one with their 45nm line-up, we might even see the continuation of AMD-CPU-Only memory from OCZ, for instance.
Further integration with AMD processors includes full support for Advanced Clock Calibration, feature from latest Southbridge chips from ATI. With ACC, you can push the CPUs further than you were before. Usually, a gain is around 200-400 MHz (from 2.8-3.0 to 3.2 on air), and the overclocking capabilities of AM3 processors remain unknown.
The biggest question that lies in the air is – will the nForce brand survive? Destiny of S3 is already sealed and delivered (GeForce), but time will tell will we see nForce 890a and 870a, or will the naming convention swing the GeForce way. We expect these chips to launch in June, during Computex Taipei 2009.