Has Nvidia dodged one puddle to fall flat in another?

BACK WHEN the GeForce4 MX was released Nvidia took a tremendous amount of fire over the card, mainly for two reasons. First was that the GF4 MX didn't incorporate any of the DX8 compatible features the GF3 or GF4 Ti lines had pioneered and extended, respectively, but second (and more important in the eyes of many) was that the card's status as a "GeForce4" part would fool buyers into thinking they were buying a lower-end GeForce 4 Ti, when what they were really getting was effectively a GeForce2 Ultra on Crack. Since the vast majority of video cards are sold into the budget market, it was feared that the GF4 MX would actually slow DirectX 8 deployment as game programmers would be reluctant to adopt features that a sizeable chunk of video cards couldn't use.

When Nvidia first unveiled the NV34 design it looked as though they'd greatly improved their budget card situation from the mess that was the GF4 MX. GeForce FX 5200 and GeForce FX 5200 Ultra were to pack full DX9 compatibility and support for all the new technology their bigger brothers could do, and, (in the case of the 5200 flavor) could run without active cooling.

The return of passive cooling to the video card market (something we've not seen since the days of the Voodoo3 and GF2 MX) is a positive and Nvidia deserves a bow for it, but unfortunately they seem to have dodged the pothole of full feature support in their low-end video cards only to fall smack into the pothole of abysmal performance.

The 5200's problem isn't that it lacks features but that it lacks the ability to use these features with any kind of speed. The problem is not unique to its DX9 abilities—even in games like Quake 3 the GF FX 5200 finds itself over-powered by the GeForce 4 MX according to benchmarks published in multiple reviews across the 'Net. This is not encouraging for a budget card that's supposed to supplant the earlier model.

This raises the question of exactly what sort of features we expect in an ultra-budget video card. The 5200, after all, is only meant to sell for about $100 per card, which doesn't allow much margin for extras or high performance technology.

Nvidia's first problem is one of positioning and core technology. The GeForce 4 MX is meant to be shoved into the ultra-ultra budget range (sub $80) which puts it 20% cheaper than the 5200 Ultra, yet outperforming it in a fair number of benchmarks. First rule of thumb when designing a product line-up? Your cheaper line of products shouldn't out-perform the more expensive ones. Second rule of thumb is, your card based on three year-old technology shouldn't be surpassing your card based on brand-new technology, pretty much no matter what. This last point might be less important if the respective speeds of the MX and FX 5200 were "really really fast" and "really fast" respectively, but unfortunately that's not the case. It's not as if the MX is trading its lack of a modern feature set for blazing speed. Instead you've got a choice between a card with a very mediocre feature set at a mediocre speed or a card with a great feature set at an even slower speed. That's not much of an option.

Rather than dropping Nvidia in a "damned if you do/damned if you don't" position when it comes to budget cards and feature sets we'd prefer to offer up some potential ideas on what might address the situation.

Re-evaluate, re-evaluate
It's a known fact that the emergence of a new Microsoft DX version and the appearance of software that uses it are events that take places at least a year to eighteen months apart. Given this, it might be a smart idea not to bother building the full features of a new DirectX version into a first-generation low-end part. Would anyone really complain if the GeForce4 Ti 4200 became Nvidia's sub-$100 part for the next six months? The GeForce4 MX caught heavy flak because by the time it was introduced Nvidia had already established a top-to-bottom line of GeForce3 cards that all had DX8 compatibility, only to intro a GF4 card that didn't.

By the time NV40 is introduced it'll be close to DX9's real debut in the open market and a fully DX9 / CineFX supporting part will make a lot more sense.

Another budgie line neded (sic)
While it took Intel some time to fine-tune the model, the Santa Clara giant has finally gotten a distinct Celeron (budget CPU) line and a Pentium 4 CPU line. The Celeron performs nearly as well as the P4 at many of your "budget" tasks—word processing, email, 'Net surfing, and basic movie-watching and such, but is much inferior to the P4 at gaming and heavy multimedia work. Assuming OEM's do their part and market the CPU's properly, this creates a stratified environment where you buy a P4 when you want to do P4-oriented tasks and a Celeron to do Celeron-oriented tasks.

The fact is that as GPUs grow ever more complicated keeping their high-end features intact requires an overall higher-powered card—and hence, a more expensive one. Having a budget card with a full feature set that's too slow to make use of it is ultimately just as useless as having no support for a feature set at all. It may be time for a bi-furcated strategy which would split 2D / business cards with perhaps only desktop-oriented 3D capabilities (Microsoft has suggested it wants DX9 support on the desktop for Longhorn) vs. higher-end gaming cards at higher prices. This might even improve the features of each—if 5200 was designed for a business environment the gaming / 3D sections of the core could be rebuilt and higher-end 2D components used in their place while maintaining a budget price.

We've offered two potential ideas for change here and while neither is perfect we feel a change is needed. The 5200, ultimately, looks just as un-desirable as the GF 4 MX did. It's too slow to use its feature set, and video cards from previous generations surpass it in modern gaming. Back to the drawing board, Nvidia—you've not quite figured out how to line up your budget offerings yet.