[Dave Baumann]

Last year ATI and NVIDIA took critically diverging choices for their high end parts: ATI chose to push on with the the tired and trusted 150nm process, although exploring the boundries of the process for 3D use in terms of chip size, complexity and clockspeeds, while NVIDIA chose the route that has served them well in the past and opted to forge ahead with the then 'bleeding edge' 130nm process. The initial rmifications of this decision were that ATI managed to get Radeon 9700 to the market even before DirectX9's release, while NVIDIA's part was delayed unto '03 and then suffered production issues. Even now, when the 130nm process is more mature, NVIDIA has seen falling profit margins due to low yields (which should be resolved now) while ATI's margins have been on the up.

A report at Yahoo Business week uses this ATI / NVIDIA example to highlight that more and more semiconductor firms are playing it safer in terms of lithography process adoption. With increasing chip complexities, especially in the 3D market, yields are getting critical, an issue which is often exacerbated with new processes, and increased cycle times can result in longer periods of waiting before these issues can be resolved - all of which can have large financial ramifications.

The dynamics of the, still emerging, consumer 3D graphics arena are still a complex beast and not even with this current example we can't forsee if what will occur from now on, but will we begin to see more players in the 3D market taking a slightly more cautious approach to silicon process adoption?

[PatrickL]

Good read. Too bad that news is 2 weeks late to add spice to the long thread in Console forum

[Anonymous]

The article was a bit too broad to be very applicable to gfx-ASICs.
Their basic premise was that the adoption of finer lithography is due to improvements in yield (of chips per wafer presumably, though they don't clarify that).

That is hardly the only reason for gfx-chip manufacturers.

It's always been the case that the market has had a front line, but that behind that front line, there are production lines of differing age, maturity and cost structure. It's not surprising if the problematic transition to 0.13um makes some businesses more cautious than than they might have been, but the basic trade-offs remain the same.

The more fundamental bugaboo going forward would rather seem to be keeping leakage under control, and that problem may indeed lead to that the improvements in clock speed vs. power consumption that we have typically seen going to finer lithography may not be quite so pronounced.

For graphics however, I'd say even clock speed and power advantages are secondary and the major benefit to be realized from new processes is circuit density. So I can't see why ATI/nVidia wouldn't move to finer lithography as quickly as ever. The benefits remain very relevant to their business, and if there's competition realizing those benefits will be important.

Entropy

[Anonymous]

I can't help but think that once Moore's law slows down, we'll still see significant gains in performance. Engineers always have a way of getting more out of limited resources. If you told the ATI Radeon team that no more transistor shrinks will occur, and if you told them to get 50% more speed, I guarantee you they'd find a way.