NVIDIA’s New Motherboard Chipset

The hardware is similar to that found in Microsoft’s upcoming gaming console, Xbox, and it has features that make it suitable for high-end usage

NVIDIA, a behemoth in the video market, has now set its sights on another potentially lucrative market–motherboard chipsets–with its nForce chipset. There is no dearth of competition in this area with established players like Intel and VIA fighting for market share. So, what sets the nForce apart? 

The nForce is an integrated chipset with onboard sound and video. We can already see gamers frowning, given the reputation of such integrated solutions. Most can’t even compete with the cheapest AGP cards, and the sound quality isn’t worth writing about. However, as you’ll soon see, NVIDIA seems to have chosen its components very critically. In fact, the hardware offered in the nForce is similar to what will be available in Microsoft’s upcoming gaming console,
Xbox. 

First, let’s look at chipset basics. Most chipsets have two chips: a North bridge, and a south bridge. The north bridge interconnects the CPU, memory and your graphics card through the AGP bus. The south bridge handles most of the I/O that takes place in a computer like the IDE and USB controllers. The connection between these two bridges has traditionally been the PCI bus, but several new alternatives are being explored including the Intel hub architecture and VIA V-Link. The north and south bridges in the nForce chipset are called IGP (Integrated Graphics Processor) and MCP (Media and Communications Processor) respectively. 

MCP

This is nForce’s south bridge and includes support for Ultra ATA-100, 10/100 Mbps Ethernet, Home PNA 2.0 (home phone line networking) and six USB ports. However, the most exciting feature of the MCP is the integrated APU (Audio Processing Unit). According to NVIDIA’s own documentation, the sound subsystem of the nForce is the same as that of the Xbox. Most of the features offered by this APU are comparable, if not better than most PCI sound cards available, and it is the first audio device to be fully compliant with DirectX 8.0. It provides real-time processing of up to 256 voices, 64 of which are 3D and the rest 2D. What caught our fancy most was its capability to do on-the-fly Dolby Digital 5.1 encoding and give the output to two, four, or six speakers.

StreamThru architecture

Devices that need real-time delivery of data are called isochronous devices. StreamThru is a complete isochronous transport system and the nForce IGP and MCP support multiple virtual channels of isochronous data streams. The link controller on the IGP side dispatches isochronous as well as non-isochronous requests to an intelligent arbiter, which guarantees memory bandwidth and latency. According to NVIDIA, StreamThru not only enhances broadband and network performance, but other time-dependent applications like digital video, audio, and multiplayer games also. StreamThru is primarily implemented through
HyperTransport.

HyperTransport

The HyperTransport protocol, originally called LDT (Lightning Data Transfer), is AMD’s solution for connecting the north and south bridges of a chipset and is being used by NVIDIA in its nForce chipset. It compares favorably to both Intel’s hub architecture and Via’s V-Link technology, and offers a total bandwidth of 400 MB/sec upstream or downstream, and 800 MB/sec total. An 8-bit wide bus running at 400 MHz, it ensures high throughput between the two chips.

IGP

The north bridge plays a crucial role in determining a system’s performance, since all communication between the processor and memory takes place through it. It also determines what CPU and memory you can use with the chipset. The nForce IGP only supports AMD Athlon processors, and both DDR and SDR SDRAM. Like its name suggests, the IGP also hosts the integrated graphics core. This core is very similar to NV11, popularly called GeForce2 MX. It has a built-in transformation and lighting (T&L) engine, and will be the first integrated video solution to offer this. It has two pipelines, is clocked at 175 MHz and has a theoretical fill-rate of 350 megapixels/sec. It also includes a 4x AGP port for external video cards.

Unlike all north bridges we’ve seen for SDRAM, the nForce has two memory controllers with independent 64-bit buses. This lets the CPU or GPU use one or both of the controllers and leads to efficient usage of memory bandwidth. This is called TwinBank Memory Architecture in the nForce. It supports three DIMM slots, two of which are connected to one controller, and one to the other. To take advantage of the twin controllers, you will have to put in at least two DIMMs, one for each controller. If you decide to use only one memory module, the system will behave like any other north bridge with a single controller. To cater to all market segments, the nForce IGP will be available in two flavors: IGP-128 with two controllers, and IGP-64 with a single controller.

The next major improvement is the Dynamic Adaptive Speculative Pre-Processor (DASP). This tongue-twisting feature is similar to the current read-ahead prefetch hardware routines found on the Pentium 4 and Athlon 4 processors. This feature analyses the data requests being made by the processor, tries to guess what data it will need in the future, fetches the data from memory, and keeps it ready for the processor. You could think of this as a kind of third level cache, implemented on the chipset instead of the processor.

Getting ready to wage war

Currently, five manufacturers (Asus, Abit, Gigabyte, MITAC and MSI) are in the process of designing and making motherboards based on the nForce chipset. From what we know now, it looks like this chipset will give Intel and VIA a run for their money. 

Anuj Jain