IGP

IGP

In the nForce, the two chips that control everything are the IGP (Integrated Graphics Processor) and the MCP (Media and Communications Processor). Between these two chips, the IGP takes on the functions of the North Bridge, and the MCP acts as the South Bridge.

Integrated Graphics Processor, the name itself oversimplifies how much this chip actually does. In addition to performing the duties of a graphics processor, the inner workings of the IGP consist of the DASP, Twin Bank Memory Architecture, the HyperTransport bus, and of course a graphics processor.

IGP	The MCP and IGP

DASP

DASP stands for Dynamic Adaptive Speculative Pre-Processor. Aside from simply having an exceedingly long name, the concept is a good one. The role of the DASP is to predict what kind of instructions the CPU will want before it needs them. It does this by recognizing patterns in the CPU's behavior. This would be akin to the way the Pentium 4 has its deep pipelines. Except in this case, the CPU isn't the one doing the work.

According to NVIDIA, the performance gain associated with incorporating the DASP is on the 20-30% mark. We doubt these kinds of gains will be the norm though. In the benchmarks provided, the DASP only gave around a 7% increase in performance when going through real-world benchmarks.

Twin Bank

Twin Bank Memory Architecture is probably the single most stunning piece of silicon attached to this chipset. That's mainly because of the number associated with it - 4.2. My friends, the nForce gives 4.2 GB/s of memory bandwidth, a good 3 GB/s more than conventional SDRAM. And to the surprise of almost everyone, a full 1 GB/s more than dual channel RDRAM, and all of this while using PC2100 DDR SDRAM.

Twin Bank Memory Architecture (TBMA) consists of two independent 64-bit channels utilizing PC2100 DDR SDRAM. The dual channels help to reduce system wide slow down from onboard components such as sound and video.

TBMA	Reference nForce

The Crossbar Memory controller is essentially the heart and soul of TBMA. Crossbar consists of two independent memory controllers that are utilized via different slots. Dubbed MC0 and MC1, each one accesses its own memory bank and delivers the data to whichever requesting party. If you look at the diagram, you can see that this configuration does indeed require two sticks of RAM. Interestingly enough, the nForce specs only show three available memory slots. Odd, especially after seeing the previous diagram, the third slot is then controlled by MC1

You can use the nForce with just one stick of RAM also. Of course, when utilizing the system in this manner all benefits of the TBMA are essentially lost. Another interesting note is the ability to build unevenly. You can populate one slot with a 128MB DIMM, and the other with a 256MB DIMM. In actuality, because of the two independent memory controllers, you can even put in sticks of RAM that have different speeds and sizes.

GeForce2 MX

Most of us aren't too thrilled when the topic of conversation shifts over to integrated video cards. Generally the dregs of the video card scene, no self-respecting gamer would ever use such a solution. With some integrated solutions offering about as much 3D performance as the Diamond Stealth 3D(S3 Virge Chipset), we'd rather not think about it.

Now, the nForce changes the rules of the game in the integrated video card scene. The IGP essentially consists of an onboard GeForce2 MX. Not exactly a powerhouse by any means, but it actually makes 3D gaming possible. All the numbers we saw of the integrated solution showed it to be an MX that was slapped onto a motherboard. Interestingly enough, the onboard MX will be running on an 8X AGP interface. If you add an external AGP card, the interface is still only 4X. Like other onboard video solutions, the MX takes advantage of the system RAM. But in the case of the nForce, the overall performance hit is minimal because of the Twin Bank Memory Architecture.