IntroductionIt's been an almost insufferable wait for the second generation of Athlon 64 and Opteron chipsets from AMD's chipset partners. While the initial products - VIA's K8T800 and NVIDIA's nForce3 150 spring immediately to mind - are competent and allowed the processors to breathe freely and perform well, there's always been a couple of annoying niggles, as is always the case with the very first revisions of such things.
That AMD dumped a lot of northbridge logic onto the processor was a reason for the early promise, so the task of building a competent supporting core logic set was much reduced. Integrating the memory controller onto the processor, along with easy interconnect with other connected system essentials with HyperTransport, means that chipset partners for Athlon 64 and Opteron (ALi added themselves to that list recently with the purchase of AMD's own 8000-series stuff) had time and effort to focus on getting the rest right.
This had the adverse effect of slingshotting the poorer parts of the chipsets into the limelight, with many online technology commentators quick to jump on initial flaws (while print magazines generally papered over the cracks, more fool them). The base criticisms with the early chipsets from VIA and NVIDIA centered mainly on one thing: the inability for either core logic set to lock the AGP and any primary or secondary attached PCI busses. While some boards claim to implement the lock, and indeed there might be the odd rogue board that does by means of clever engineering, it transpired that, as far as the base chipset was concerned, none did. K8T800 and nForce3 150 are physically unable to clock those busses discretely, instead deriving the clocks via a divider from the driven CPU frequency, itself derived from the base HyperTransport clock.
That leads on to the nForce3 Pro 150's limitations in its HyperTransport implementation, running a 600MHz LDT/HT clock for 1200MTs/sec, while its chief competitor, the K8T800, drove its implementation at 800MHz, offering theoretically better performance under certain situations. The nForce3 Pro 150 also chose to castrate the width of the downstream HyperTransport connection to 8-bits (just a single byte of data transfer), half the usual 16-bit width of the bus.
And while benchmarked performance doesn't really give any cause for concern, it's a tech-focussed checkbox that goes unticked by the enthusiast, allowing K8T800 to garner more positive press, and one has to therefore assume more sales.
Then there's the disadvantages to NVIDIA's single chip approach in its AMD64 core logic endeavours. With VIA favouring their well-proven V-MAP method of bridge production, keeping north and south bridges disparate for greater flexibility, NVIDIA dumped the whole shooting match onto a single piece of silicon. Clever engeering? Most certainly. Well judged in terms of the features they gave it? Hell no.
While current nForce2 boards for the ageing 32-bit Athlon XP platform have long made do with peripheral ASICs to augment absent features from the MCP southbridge, it was hoped that NVIDIA would hit the ground running with their single-chip nForce3, giving it an up-to-date feature set that would match what VIA could knock out, without the need for many peripheral chips hanging off whatever bus they could find. Remember, board costs for a brand new platform are rarely cheap, so having to add silicon to beef up a feature set is suboptimal in this stage of the game.
It wasn't to be so however, NVIDIA missing the boat in terms of features that everyone had come to expect, even more so with their decision to drop the major selling point of nForce2, the SoundStorm audio. The base processing hardware that NVIDIA supplied on the MCP for SoundStorm was able to do Dolby Digital hardware encode of audio, enabling connection to popular and prevalent decoding hardware and therefore decent speaker setups.
However it's not gone and forgotten at NVIDIA. It's true that they can't do all the APU functions, using the CPU, over the PCI bus, negating its use in off-bridge scenarios on discrete PCI soundcards. But think about upcoming new bus architectures. Bingo.
Therefore, while CPU level performance has shone on initial Athlon 64 and Opteron motherboards, other things have dulled the sheen that the CPU applies to everything AMD64.
So it's new chipset season then. K8T800 Pro is on the way, with 1GHz HyperTransport base frequency support (2GTs/sec, woo!) and apparent ability to lock the AGP and PCI clocks. Paired with VIA's already incredible southbridges and cheap peripheral ASICs, K8T800 Pro seems the core logic to have with second generation, enthusiast level AMD64 boards. But NVIDIA will argue that vehemently, and it's their new nForce3 product that I've been able to take a look at for this article.
It's still a single chip design, making it slightly suboptimal in many people's eyes (although SFF box makers will rejoice), but it seeks to put right many of the criticisms that were levelled at the original chip in terms of supported features and its HyperTransport implementation.
Let's take a closer look.