Someone made a good argument to me that the BOM for Tegra2 is too high for the DS2. It needs too many support chips (high performance memory, etc. etc.) and that the overall silicon cost would be far too high for Nintendo to be comfortable.
Just lurking a bit and couldn't resist...
There's only one non-custom SoC today that would have a lower BOM for roughly that level of graphics performance, the Samsung S5PC110 (45nm 1GHz Cortex-A8, 200MHz SGX540 - it's the one in their first Bada phone BTW) and I'd be surprised if Nintendo is going with Samsung. There are two other chips in the same ballpark: OMAP4 and Snapdragon2 (+derivatives of the latter). The former has slightly greater external component requirements (64-bit LPDDR2, HDMI PHY, etc.) whereas the latter integrates a HSPA+ baseband Nintendo nearly certainly wouldn't want. Also the main external components on the Tegra2 reference board are netbook-specific (i.e. related to the multi-cell battery, LVDS, USB hub for >2 ports, etc.) and LPDDR2 isn't *that* expensive in the 2011 timeframe.
Oh, and before any fool takes this to mean the S5PC110 is what is in the iPad: it's amazing how clueless everyone is there. Come on, 'A4', how obvious can you get? It's their fourth custom chip: 90nm A1 in the iPhone 2G/iPod Touch 1G/iPhone 3G, 65nm A2 in the iPod Touch 2G, 65nm A3 in the iPhone 3G S/iPod Touch 3G, 45nm A4 in the iPad and 2010 iPhone. I don't know how many times I'll have to link this PDF, but I guess I'll try again:
http://www.samsung.com/us/business/semiconductor/news/downloads/FoundryBrochure0507.pdf (i.e. quite a bit of the IP is synthesized by Samsung, maybe less each gen, but each of those SoC is definitely custom, see
http://www.eetimes.com/showArticle....E1GHRSKH4ATMY32JVN?articleID=218100442&pgno=2 for the 3GS)
Anyway I do believe however that Nintendo is more likely to go with a custom solution at this point, whether from NV or from someone else. Tegra2 is a bigger chip (~50mm2) than most imagined, and it could also be an unannounced derivative on 40nm or even 28nm, but that seems unlikely. It is also perfectly possible that they have decided to go for a custom non-NV solution at a Japanese manufacturer, although there haven't been many leaks of what that design would be if so. BTW, NVIDIA has claimed publicly to the Linley Group that they will sample Tegra3 (afaik on 28nm with triple gate oxide) before the end of the year, aka Q4. This implies a tape-out schedule around TSMC's "initial risk production" timeframe of June 2010. That will nearly certainly be the first Tegra with a DX10-level/CUDA-capable GPU.
Oh, while I'm at it: there's a slight inaccuracy with the Linley Group's claim that Tegra2 does not support 1080p High Profile. It is supported AFAIK, but only up to ~10Mbps (so YouTube 1080p is fine), whereas Main Profile and Baseline are supported to substantially higher bitrates (so definitely no Blu-ray ISOs). Video Encode, on the other hand, remains 1080p Baseline. And I wasn't sure, but someone from NV confirmed on their forums that NEON is not supported on either core, fwiw.
And before anyone thinks I appear only for Tegra-related matters: I think the level of quality of App Processors throughout the industry has increased greatly in the last 2 years. Even Samsung and Marvell have extremely appealing solutions for several markets, and they seem to be delivering on the software front. In a way, the market is maturing (after all there are no mainstream standards beyond 1080p for example!) although certainly not yet a commodity. NV's greatest advantage here is velocity, and on 40nm it seems to have paid excellent dividends (85 tablets in flight, T1&T2 phones in Q3). For Tegra3, it will be interesting to see whether TSMC can deliver on 28nm SiON. It seems like a very incremental process to me, so they probably will, but I'm sure everyone is wary at this point.