The "halo effect" - is anyone really that dumb?

[Arun]

Do you have access to links or data for Q107?

Sadly, the datapoint for Q406 is exceptional: NVIDIA executives gave it as an answer to a question in the quarterly conference call, and nobody asked for a similar datapoint in Q107, so I don't think it's possible to get more information there, sadly.

Just how the "monster-chip" approach has been justified up until now is probably a quite complex question as well, but that doesn't mean that the industry can/will pursue that path indefinitely.

Yeah, everything does seem to be pointing at NVIDIA and AMD giving up on monster chips and adopting GX2-like solutions within the next 18 months, and probably sooner than that. We'll see, I think the consumer reaction to that will be very interesting to watch.

It's particularly tricky to make market predictions right now. The dismal uptake of Vista (and therefore slow adoption rate of DX10)

Well, that is very important for the AIB market, but it obviously doesn't really matter in the OEM market where the adoption is complete and instant no matter what, although with a few noticeable exceptions I'll admit, but nothing statistically significant.

The market will have changed in two years, but how? Are high-end monster chips going to be phased out in favour of something more modular, or are they going to be the last stronghold for the gfx-ASIC as the low and middle end is eaten up by the new integrated approaches?

It's interesting you bring up modularity, because it certainly seems like a good premise in theory. I don't think it's unbelievable that eventually the industry will move towards not having single-chip solutions that hit pricepoints above $149-199, and just have SKUs with 2-4 chips above that. And allow for traditional SLI between boards too using 2-4 PCI Express slots.

The problem with that, IMO, is that you waste way too much video memory in modern multi-board scaling approaches. Whoever figures out an elegant way to get rid of that problem (either through an exotic packaging solution or an innovative algorithm) certainly could change the industry a fair bit, but right now it doesn't make sense at all, I think. Furthermore, the $249 segment remains high-volume in the AIB market, so it might not make sense to have your 'base-chip' aim at anything lower than that.

It's hard to say really, but for the next 3 years or so, I would expect to see 3 chips in each IHV's generation of products, with a rare 4th when it makes sense. Who knows, however... Heck, I'm certainly not considering Intel here either, so it will be interesting to see what their lineup will look like too!

 

[3dilettante]

The problem with that, IMO, is that you waste way too much video memory in modern multi-board scaling approaches. Whoever figures out an elegant way to get rid of that problem (either through an exotic packaging solution or an innovative algorithm) certainly could change the industry a fair bit, but right now it doesn't make sense at all, I think. Furthermore, the $249 segment remains high-volume in the AIB market, so it might not make sense to have your 'base-chip' aim at anything lower than that.

There's shared memory with coherency or message passing, for starters.

I don't think the concept of multiple processors with local memory pools and non-uniform access latency is something GPU designers created only recently.

Virtual memory is about halfway there all by itself.

As for being cheap and having good performance, that's something else entirely.

 

[trinibwoy]

Er... that's another thread.

Where you were trying to make your point based on Steam numbers which now contradict exactly what you were saying. Fair game.

The Beyond3D forums is a place that attracts 3D enthusiasts and industry insiders alike, from all over the world. This allows some fairly extreme opinions and priorities to prosper. Periodic reality checks may be in order.

I think I've completely missed the point you are trying to make here. Nobody is trying to convince you that high-end cards are mass market products. The subject of the thread is the influence of the high-end on the perception of cheaper variants. Yet you continue to focus on the segment of the market that neither knows nor cares about high-end/low-end or anything in between.

 

[trinibwoy]

How can you assume that G80 cards will enjoy the same market share (or indeed remain high-end) for upcoming years? Does a factor of two make the slightest difference to the overall picture?

Not sure what you're talking about. I just showed you numbers. I wasn't predicting the future.

Why do you ignore the crossreferencing to market data which I referred to above, which shows that the total market share of $250+ cards is just over 2% for add-in boards. As I stated, (and as is obvious to anyone who actually checked the data,) this means that the Steam survey overestimates the impact of the G80 products.

I think we just disagree on the "overall picture" and the relevant market for these products. I'm curious about why you think we should concern ourselves with the vast numbers of IGP's and cheapo add-in boards out there. Who really cares besides the manufacturers? And that's from a financial standpoint. When discussing the technology and market impact they are irrefutably irrelevant.