Originally posted by: Cookie Monster
There is a difference between the first 55nm product being released by the major IHVs to the actual process technology being available. The 55nm GP process went "online" during March 07 timeframe, where such technology was offered to any companies requiring the need for such process technology, so roughly 21 months of service. With the new 40nm process going "online" late 2008, the argument that your making is kind of moot.
Its not a moot point, as 65nm and soon 55nm will be that much older and that much cheaper relative to the newest offered process, which in this case will be 40nm. So now that 40nm is being offered by TSMC, do you really think Nvidia and AMD will get them for the same price per wafer as 55nm?
Its a well known fact that nVIDIA is behind the process race ever since the NV30 fiasco compared to ATi/AMD which however is not a bad thing depending on the situation (theres no right or wrong method here, but in this case too much 65nm stock ).
Yep, but this was clearly by choice, erring on the side of caution with die shrinks and new architectures due largely to NV30's failure. AMD recently experienced similar with a shrink and new architecture with R600, they just quickly swept it under the rug and went with a 55nm shrink and RV670 within 5-6 months.
The significance of this latest roadmap is that it shows Nvidia is pushing the process envelope sooner so that they're no longer behind the 8-ball when it comes to process technology. In the last 8 months Nvidia has shifted from 65nm to 55nm and now 40nm when their typical progression would have been 12-18 months in the past.
With both IHVs being one of the biggest customers for TSMC, I doubt these transitional costs would have a large impact on the deals made between them and the other IHVs. But you are right that a newer process technology should cost higher than the older process technology, but its premature to say that newer process technologies should be dismissed because it costs more.
Who's saying it should be dismissed? I'm saying any cost analyses claiming wafer prices are the same between different processes are clearly flawed. Market pricing and financials have once again proven such claims to be false.
As for transitional cost not having a significant impact on process pricing, this blurb gives some insight to the significance of Nvidia pushing for newer processes sooner:
Doug Freedman, an analyst at American Technology Research, said in a research note that Nvidia was not getting the 55nm capacity it needs from the silicon foundry giant and that the problem was likely to get even once Nvidia starts to utilize more advanced fabrication processes, 45nm and 40nm. It is crucial for Nvidia to utilize the most advanced process technologies possible as this allows the company to build very powerful GPUs at lower costs.
Nvidia and AMD's GPU segment are certainly two of TSMC's larger partners, but Nvidia holds at least a 2:1 edge based on GPU shipments alone without considering their chipset business. Its obvious additional pressure for new process capacity will force TSMC to either transition more fabs to newer processes or limit production based on capacity. This is obviously less favorable for TSMC when Nvidia was placing large orders on older processes that did not require additional capital, and as a result, there's the resulting tension between TSMC and Nvidia.
Well this clearly indicates that the HD4800 series have had more traction than the GTX series (GTX280 being the worst of the bunch and anyone can guess why is it so). Im guessing most of it has to do with the introduction of the HD4850 and now the HD4830.
Certainly, the problem is the 4850 and 4830 don't compete with the GTX series, which is why the 9800 numbers were included. Once you properly factor in competing price segments its obvious 4800 didn't have nearly as much impact as recent generation GPUs, particularly the 8800 series and its commanding 25% of DX10 parts and 12% overall GPU share.
This last sentence characterizes AMDs business strategy in the GPU market ever since the RV670. nVIDIA on the other hand is still in the "single monolithic GPU" bandwagon, with GT212 and GT300 scheduled for 09 releases which both are single monolithic GPUs. Your theory above doesn't quite match with whats been happening this generation especially since the competition is providing a product that is much cheaper while providing ~90% of the performance.
Yep its AMD's little core strategy, Nvidia has shifted their strategy so that they can have their monolithic core and beat AMD's little core, but also double-up and beat AMD's multi-GPU solution at any given time as well. The only hurdle in the past has been being on a different process node, which shouldn't be an issue now that they're both pushing 40nm simultaneously.
Also core size alone doesn't tell the whole story, as Nvidia has shown twice now with both GX2 parts. Both times Nvidia's part has come in at a higher transistor count and larger die size, and both times their parts have drawn less power while outperforming AMD's parts. Obviously core frequency has a significant impact in this regard, as Nvidia has gone with larger chips and lower clocks while maintaining some overhead, whereas AMD has gone with smaller chips and fewer transistors with higher clocks and less headroom.
nVIDIA have obviously lowered the prices to increase demand for the cards yet the required demand is not there as indicated by the old 65nm inventory ala GTX260 (also suggests that the yield rate for full fledged GT200 chips were poor) and the lack of transition to 55nm for these performance parts unlike the GTX285.
Again, your assertions are not only contradictory but haven't been reflective of what's happened in the marketplace either. The abundance of GTX 260, Nvidia's ability to cut prices on them continuously, and their ability to seamlessly introduce variants with more cores (Core 216) shows yields were quite good. Not to mention GTX 260s clock as well or better than full-fledged GTX 280s from most reports. In reality Nvidia is simply self-neutering fully capable parts in order to meet market demand and pricing segments, which really wouldn't be any different than what they've done for years. Recently they've shown the ability to quickly burn off remaining inventory by dropping prices as shown with 65nm core 192, 216 and soon GTX 280.
Lower prices doesn't necessarily result in higher marketshare and sales volume (examples such as GTX280), same goes for higher margins and lower market share/sales volumes (G71 is a key example where it provided high margin, resulted in high sales volume which led to higher market share). Presuming RV770/GT200 has good margins at their rated MSRP (note that nVIDIA launched their cards first), you can already guess which IHV been hurt the most when the price war began.
Thinking about it, nVIDIAs G71 closely resembles the RV770 IMO.
Sure it does. Lower prices are always going to result in higher marketshare and sales volume. You don't think GTX 280 was selling better after its price drop from $650 to $450? It wasn't even statististically significant in Steam's Survey until November. Both vendor's take advantage of this basic economic principle in their product offerings, offering essentially the same parts with some performance limiters or enhancers with vastly different mark-ups. As for which IHV has been hurt the most, again, I'd say the one still posting a profit and maintaining 40% gross margins would be hurt the least. After the initial price cut forced by AMD, Nvidia has consistently undercut pricing with GTX 260 and G92-based parts.
Add the G7x generation also. Sure ATi fumbled with R600 (happens to companies all the time), but the RV670 generation did show what they were capable of, forcing nVIDIA to release its G92 based products earlier than expected which screwed up nVIDIA's lineup (same thing with the 9800GTX+ and the release of HD4850). I think your overestimating or just forgetting that nVIDIA has other numerous markets in which they compete in which can offset the potential losses (compared to estimates compiled by their financial teams) caused by the lack of sales of their GT200 based products.
Actually if you look at Note 16 of their
Q3 10-Q you'll see exactly how much of their business is still focused on the sale of GPUs:
GPU = 461M
PSB = 199M
MCP = 197M
CPB = 34M
Other = 5M
Total Revenue = 897M
PSB is their Professional Solutions Business, or Quadro and Tesla, which are still wholly based on their GPU architectures. From that you can see 75% of revenues and nearly 100% of their profit is generated from their GPU business. Their chipset and Other actually lost money and consumer products barely broke even.