GeForce 3 - Overpriced?

We have seen the official unveiling of nVidia’s NV20, or GeForce 3. Sadly, the excellent potential offered by the GeForce 3’s new feature set has been overshadowed of late by talk amongst the hardware-buying community about the card’s astronomical price tag. Recently, x-bit labs ( http://www.xbitlabs.com ) posted a news story on their front page stating that the first batch of GF3 cards will cost $699 (translating to about £500 excluding VAT), and will include 64Mb of 3.8ns RAM running at 460MHz (230MHz DDR).

I’m willing to wager a significant proportion of my worldly assets that nVidia’s recommended price is going to be considerably below $699, and we will certainly not see any GF3 cards costing that much. I’m not sure where x-bit labs got their information from, but it is wholly inconsistent with everything else we’ve heard so far.

In fact, since that news post went up, it has been announced at MacWorld that the Apple-friendly incarnation will cost $600 BTO (built to order) for the PowerMac G4. On the basis of this, I think it’s safe to assume that nVidia’s recommended price will be an absolute maximum of $600 (quite possibly less, as Mac components have traditionally been more expensive than their PC brethren), and cards will be available off the shelf for less than that. All shall be revealed tomorrow, but I wouldn’t be too surprised if the basic, no-frills models cost in the $500-550 range.

Nevertheless, this is an undeniably hefty sum. I’d like to make it clear that I’m neither happy to see graphics cards costing several hundred pounds, nor do I feel it’s necessary or even justifiable. Despite this, I feel obliged to defend nVidia on several counts (and condemn them on several others), especially given the nature of some of the more vitriolic anti-nVidia rants I’ve seen on forums and hardware sites over the past few days. The most common accusation is that, in the absence of competition, nVidia are deliberately charging a high price for the GF3 to protect their profits. I can’t disagree that there is an element of truth to this, but there are some definite flaws in this argument.

Firstly, it is important to realise that two components comprise the vast majority of a graphics card’s cost – the memory and the core chip. Only one of these is manufactured and sold by nVidia, and thus produces a profit for them with every card sold – that’s the GPU, by far the less expensive of the two when we’re talking about cards with 64Mb of the fastest DDR RAM available. Thus it’s the OEMs – Guillemot (Hercules), Elsa, Creative, Leadtek etc. – that receive the bulk of the revenue.

However, while it is true that GPUs are relatively cheap to produce in comparison to high-spec memory, it’s not necessarily the case that they constitute a correspondingly small proportion of the final card’s cost. It’s entirely plausible that nVidia intend to make a huge profit on the first batch of GF3 GPUs and attempt to ‘hide’ the extra cost by using 460MHz DDR RAM - since the inclusion of such memory makes the card very costly regardless of the chip used, nVidia can happily charge an extra $50 for each GPU (taking an arbitrary figure) without it having too much of an impact on the total price of the card. This $50 per chip could well double nVidia’s profits, and probably won’t affect sales much as the only people interested in cards featuring such expensive memory will buy a GF3 as soon as possible no matter what the price is.

This could well explain the fact that each GF3 card is likely to cost about £50-100 more than a GF2 Ultra; ignoring the 3.8ns x-bit labs rumour, the memory on the two cards will probably be to all intents and purposes the same, so most of the extra cost is accounted for by the GPU being more expensive. I don’t doubt that the GF3 chip, with its 57 million transistors and numerous never-seen-before features, costs more to produce than the GF2, but it seems unlikely that production costs alone account for the price gap.

While it can correctly be argued that nVidia are exploiting the consumer with their pricing policy and are thus worthy of criticism, I think it’s safe to say that you’re quite a lot for your extra cash, even if most of it is going straight into nVidia’s pocket. The GF2U, being merely a heavily overclocked GF2 with exotic memory, had absolutely nothing new to offer in the way of features, yet at the time it was released we didn’t witness anywhere near as much unrest amongst hardware enthusiasts as now. Simply put, if you thought £350 or so was reasonable for a card based on a year-old architecture at the time of release, then you can’t complain about £400-450 for a card that is far in advance of anything currently available (and if you don’t believe me, I’m sure you’ll believe John Carmack – look at http://www.bluesnews.com/cgi-bin/finger.pl?id=1&time=20010222225435 ).

Also, don’t forget that while production is ramping up, costs are likely to be higher, so we can expect either the GPU price (and hence overall card price) to fall in a few months, or for nVidia’s profits per chip to increase further. So yes, nVidia are charging a premium for their latest and greatest, but at least you’re getting visual quality- and performance-enhancing features that the competition are presently unable to offer. It’s hardly a new concept in the computer industry for a new product to be sold at a considerable mark-up for the first few months (Intel have been doing it for years); I’m not saying it’s ‘right’, I’m just saying it happens and there’s nothing to make the GF3 any different from most other brand new hardware in this sense.

However, there is an important distinction to be made between the GF2U and the GF3 here: whereas the cost of the GF2U could be explained (if not justified) by it being a stop-gap product aimed at hardcore gamers who demand the highest performance available and are willing to pay for it, the only GF3 we’ll see for now is hardly within the usual price bracket for a mainstream product. This begs another question: why have nVidia only chosen to release a GF3 with the fastest available memory? Why not a more affordable version sporting RAM in the range of 166-200MHz DDR?

This is the key issue as far as I’m concerned. nVidia don’t (directly) make any money from including higher-spec RAM on their cards, so why do they do it? I can think of three possible reasons:

1. They wish to ‘hide’ the excessive profits they make on GPUs by including very expensive memory (as explained earlier)
2. They are not sufficiently confident about the chip’s HSR capabilities to release a version with comparatively ‘slow’ RAM – such a card might not actually perform any better than a GF2U in current games, which would be a bit of a PR disaster for nVidia (even though the card could potentially cost a lot less than a GF2U)
3. They intend to push multi-sampling anti-aliasing (which consumes the same bandwidth as previous anti-aliasing methods but no extra fill-rate) – hence the fastest memory available is required to maximise AA performance

Sadly, I suspect it’s a combination of the first two. The upshot of this is that, in a few months, we’ll see much more affordable cards using slower DDR memory, although if point 2 is true then they may not be stellar performers. By this time, production of GPUs will have ramped up fully, and their cost should be adjusted accordingly. If a GF3 with 166MHz DDR memory is released, there’s no reason why it should cost any more than the vanilla GF2 GTS does now.

In the absence of such cards, there is still a way of determining their potential performance – underclocking. Assuming it’s possible, I think it would be a very valuable exercise for a reviewer, or even an early GF3 buyer, to underclock the card’s memory to 333MHz and do a direct comparison against a GF2, which, apparently like the GF3, features four dual-texturing pixel pipelines running at 200MHz. This would serve two purposes – to test the efficiency of nVidia’s HSR implementation, and to determine how well future, cheaper GF3-based cards will perform. I strongly encourage anyone with a GF3 to try this and bring their results to the attention of the public (as long as you credit me for it).

There is one mystery that remains unanswered: why does top-spec RAM cost so much in the first place? System RAM prices have crashed in the past couple of months, so why haven’t we seen a similar reduction in the price of video card memory? This is something I know little about, but it is certainly worthy of further investigation. Granted, this 230MHz, 128-bit DDR stuff provides about 7.4Gb/s memory bandwidth compared to 2.1Gb/s with PC2100 (133MHz 64-bit DDR), but I remain unconvinced that the immense price premium is solely accounted for by production costs.

Ideally, of course, we’d like there to be a significant fall in the price of 5ns and faster DDR – then there’d really be no excuse for nVidia releasing a new product costing $600. Until then, though, the GF3, much like the GF2U, remains a card for the hardcore gamer who always needs to be at the cutting edge; the rest of us with a tighter grip on our purse strings will have to wait until a version with cheaper memory becomes available, whether it be due to a fall in RAM prices or simply the use of slower RAM.