The nForce Professional is once again, a single chip MCP, dubbed the 2200, in its most basic configuration much like the desktop variants of the nForce 3 and 4.

A quick specification overview

• 16 x 16 HyperTransport link
• Flexible PCI-Express with 4 controllers and 20 lanes
• NVIDIA Networking
  Gigabit Ethernet with TCP/IP Offload Engine (TOE)
• High performance storage solution
  4 SATA 3Gb/s
  4 ATA-133
  NVIDIA MediaShieldTM RAID
  Native Command-Queuing
• 10 USB 2.0 ports
• PCI 2.3

There's not too much different from the nForce 4 at this point but it is worth noting that the MCP with the nForce Professional also supports RAID 5 in addition to RAID 0, 1, 10 and JBOD. The nForce Professional is already certified for dual core processors when AMD releases them later this year, NVIDIA will be ready on the workstation and server front.

The big surprise however is not the 2200, but the 2050 which is billed as an "I/O Companion Chip." It is basically a stripped down version of the MCP 2200.

• Operates with nForce Professional 2200
  Provides additional I/O capability
• PCI-Express
  4 controllers
  20 lanes
  Flexible configuration
• NVIDIA Networking
  Gigabit Ethernet with TCP/IP Offload Engine (TOE)
• High performance storage solution
  4 SATA-2 3Gb/s ports
  NVIDIA MediaShieldTM RAID

Each processor is capable of connecting to a total of two MCP chips through a HyperTransport link - at this point it can be a single 2200 and 2050 or a pair of 2050s. The maximum number of 2050s at this point is 3, as NVIDIA is concentrating on single and dual processor configurations at this time but this may change in the future should they decide to go after the 4-way and 8-way markets.

NVIDIA's RAID technology allows for RAID configurations to span across multiple controllers and in a nForce Professional configuration with a 2200 and 3 2050s, there can be a total of 16 drives on a single RAID array which should allow for some interesting disk configurations. The argument for multiple 2050 chips on a single board is that it should reduce the overall PCB space needed when compared to separate, discrete boards.

Stand alone towers, blade servers and server rack configurations should be all available. Something that is missing from the server configuration however is integrated video. When questioned about this, NVIDIA replied with a terse "no products with onboard video will be announced today." It would of course make a lot of sense to pop in a GF4MX onto the like the nForce 2 IGP just for convenience's sake*. On a personal level however, I have worked a lot on nForce 2 IGP boards and they are the most finicky boards when it comes to memory compatibility issues. NVIDIA will probably have to resolve this as they really do not want to compound IGP issues with fairly strict memory guidelines for the typical server board.

*edit - this is not imply that any sort of 3D functionality is needed for a server setup - the GF4MX remark is just a convenient example of a product that NVIDIA already has in the integrated department. any sort of VGA capable output would sufficient. To further clarify this, NVIDIA has stated that there will be many OEMs that will provided integrated video as a seperate chip, whereas we were speaking of integrated video as something integrated into the MCP itself.

Announced server OEMs

Workstation boards are fairly similar to server boards - here are the three planned configurations.

High-end

• NVIDIA nForceProfessional 2200 + 2050 MCP
• Dual AMD OpteronProcessors
• Dual NVIDIA Quadro GPUs(SLI Capable)

Mid-range

• NVIDIA nForce Professional
• Dual AMD OpteronProcessors
• Single NVIDIA Quadro GPU

Entry-level

• NVIDIA nForceProfessional
• Single AMD OpteronProcessor
• Single NVIDIA Quadro GPU

The main difference is the SLI capability on the high end board - it consists of two FULL x16 slots unlike the 2 x8 configuration on the desktop SLI board. This is possible as the 2050 chip supplies an extra 20 PCIe lanes. Note that there is only one listed 2050 chip for the high end, unlike the three that are supported for the server version. The mid-range class still offers dual processor capability but loses the 2050 chip and therefore SLI support. The bottom end is a single chip, single processor configuration which may be a hard sell against a nForce 4 Ultra desktop board.

One of the issues that we brought up with current Opteron systems is the shoddy state of SATA RAID support for various add-on chips under Linux. NVIDIA claimed that one of the stipulations that OEMs have is full Linux support for everything - including SATA RAID which should grow in importance with a continual convergence of SCSI like features in SATA drives. NVIDIA stressed that the featureset is largely driven by OEM demands - if there is enough demand for it, they'll implement it.

NVIDIA once mentioned that their biggest competitor is Intel and it remains true here - the biggest challenge that NVIDIA is competition with Intel platforms, not existing AMD ones. AMD's own chipsets seem more to be a necessity rather than something that they are interested in pursuing as a money making endeavor. AMD quietly dropped out of the desktop chipset market once NVIDIA, VIA and SiS gained traction and it would not be too surprising to see them do the same on the server side.

The nForce Professional is a logical step for NVIDIA in the chipset market - we have seen what they have done to the professional graphics side of things with the Quadro lineup and I think it is safe to say that they will do much the same with the nForce. The feature set is much as what we would expect for server and workstation platforms - the only real complaint is the lack of onboard video on the server side. The expanded I/O capability through the 2050 chip is pretty innovative and it will be interesting to see how adoption will be compared to just the regular 2200 MCP.