nForce 4, Athlon 64 4000/FX-55, and the Industry

A Little Bit of Everything

by Josh Walrath

This all changed with the release of the nForce 3 250 series of chips. NVIDIA finally got it all right with this product line, and the marketshare that they have been gaining since the release of this product has proven its success. The first thing that NVIDIA did was to release this product in three different SKU’s. The first was the standard, and inexpensive, nForce 3 250. This had all the basic features of a standard Athlon 64 chipset, including SATA support and the 1 GHz HT link. The next level up was the nForce 3 250 Gb, which featured another two SATA ports controlled by an external PHY, integrated Gigabit Ethernet, and the integrated NVIDIA firewall. The final product was the nForce 3 250 Ultra, which includes all of the 250 Gb features, as well as 7.1 sound support (with an external PHY).

                These chips have been selling like hotcakes. Not only did NVIDIA cover all of the major price points with these products, but they all come in a single chip. While the single chip architecture may increase the complexity of the motherboard by a small amount, the ability to handle all of that functionality in a single chip does decrease the overall cost for motherboard manufacturers. At the end of Q2 2004, it was estimated by 3rd party industry watchdogs that NVIDIA had garnered around 45% of the total Athlon 64 chipset market. This was the breakout product that NVIDIA had been waiting for.
Motherboards from nearly every manufacturer now feature nForce 3 250 chipsets (with the noted exception of Abit, but that is a different story altogether...). These range in price from $73 to over $200, which easily covers the entire spectrum of marketable motherboards. While the nForce 3 250 series have been a knockout punch for NVIDIA, it still was based on the AGP standard, which is now slowly starting to fade.

nForce 4 Series

                NVIDIA could have taken the easy way out and merely replaced the AGP circuitry with PCI-E, repackaged it, and sent it out into the world. Lucky for consumers, NVIDIA decided to do far more than that.  The nForce 4 is essentially a new chip from the ground up. Unlike VIA, which does incremental changes with its north/southbridge combinations, NVIDIA took what it learned from the nForce 3 250 series and created an entirely new product in the nForce 4 series. Of course the first major change is the PCI-E support, but NVIDIA did significant work on the SATA/IDE controller, as well as put true dedicated firewall circuitry into the integrated Gig-E controller.

                The nForce 4 series of chips are made on IBM’s 130 nm process. The older nForce 3 250's were made on TSMC’s 150 nm process. There will be a PCI-E lock on the nForce 4, and because of the process transition, we can assume that it has the ability to clock higher than the current nForce 3 250 design. Many users are hitting 300 MHz with the nF3 250, so we can assume that overclocking with the nF4 may be just as successful, if not more.

PCI-E

                Intel introduced the world’s first PCI-E solutions with the 915/925 chipsets this summer, and VIA was the second to follow with the release of the K8T890 series. With the nForce 4, NVIDIA joins the PCI-E ranks. While there are still no shipping AMD PCI-E products, NVIDIA promises that by the middle of November there will be plenty of nForce 4 products on the market. The nForce 4 and nForce 4 Ultra products feature the standard PCI-E loadout when it comes to lanes. These both feature 1 x16 PEG (PCI-E Graphics) slot and 3 x1 PCI-E slots. There is also the standard complement of regular PCI slots for many current add-in cards.

                The standout product in this series is the nForce 4 SLI, which features 2 x PEG slots as well as the 3 x 1PCI-E slots. When both PEG slots are used, they do not support the full x16 complement of lanes, but in fact each utilizes x8 lanes for an aggregate bandwidth of 4 GB/sec for each card. If the user only wants to use one PEG slot, they can easily reconfigure the motherboard to accept the full x16 lanes (this is done by reversing a plug in card located on the motherboard) to get a total of 8 GB/sec of bandwidth for that single PEG slot. NVIDIA claims to have done a lot of testing on many different configurations, and they found that balancing them both out at x8 lanes creates the best possible solution.

                The nForce 4 has a dedicated SLI core controller, which utilizes a significant amount of circuitry to route data effectively to both cards. NVIDIA has also done a lot of work to make this solution both flexible and fast. It is a fully validated program, so while it is not recommended, the user has the ability to use cards from different manufacturers in SLI setups. The only thing that needs to be insured is that boards with the same family of chips are used (eg. a GeForce 6600 should not be used with the GeForce 6800, or a standard 6800 should not be used with the 6800 Ultra).

Here is a quick shot of MSI's SLI nForce 4 Motherboard.  Note the riser card between the two PEG slots.  This controls the amount of PCI-E lanes going to each PEG slot.

                From my understanding dual SLI can be used in two different ways. The first is as two separate cards featuring all available video outputs. This is essentially the same as a current setup with an AGP card with a second card in a PCI slot. The only advantage the user gets is more bandwidth to the secondary card, which could be important in high end graphics programs. The second, and what will undoubtedly become the most common setup, is the pure performance mode. Only one set of video outputs will work in this scenario, but the performance is doubled for that one card. This is the feature that most enthusiasts and gamers are drooling over. The ability to put two GeForce 6800 Ultras or GTs together and get nearly double the performance in games is nothing short of breathtaking for most of us. The idea of Doom 3 or Half Life 2 at 1600x1200 with all settings maxed at 4X AA and 16X AF makes most gamers weak in the knees. Having a 22" monitor of my own, this idea is very appealing! In the future NVIDIA plans on making this a bit more flexible by enabling both the high performance mode AND the ability to use all 4 video outs on the two cards.

Next:  SATA Controller

Copyright 1999-2004 PenStar Systems, LLC.