In February, it was $500-700M depending on how you count. I assume it depends whether you count the PortalPlayer acquisition for example, and whether you count previous GoForce R&D (presumably if you didn't count it at all it'd be even a bit lower).1) Nvidia has spent $600M on tegra (I'm frankly astounded at that figure)
The fact of the matter is that the ARM11 MPCore used in the Tegra (and no other NV product before it) was licensed in May 2005. Also Rayfield joined NV in 2005. By that time, the GoForce 5500 was probably taped-out or nearly so; the only other two chips after that were the 5300 and 6100 - the former was just a cut-down 5500 with eDRAM on 65nm, the latter was taped-out by PortalPlayer before they bought them. So that's a ~500 people team for several years now...
This figure is misleading though: 35 of those are smartbooks, where NV has done a heck of a job reducing R&D expenses for ODMs through their module approach. That module they showed at Computex isn't just a devkit or a board support package; it's literally a production board that's used as-is in real products.2) 50 end-user products "in-flight". Media players 1st, then smartbooks, then phones.
The total number of Tegra smartphones in development was around 12 at last count, IIRC. And that probably includes ODM devices like the ones at MWC09. At this point it seems Tegra1 design cycles took too long to become part of any OEM/ODM's "platform"; i.e. it's probably used for one or two devices, maybe three in the best case, at a given manufacturer and that's it. Then they'll switch to Tegra2, and there their hope must be to penetrate full platforms of devices rather than just flagships. Combining velocity with SW compatibility can be quite powerful.
Don't count on it Mike's pretty damn good at making you think he answered or is hinting at something when he's really just dodging part of what you said.3) One segment they have design wins for is "media pad" when pressed he describes it as a 7-10" tablet type device. When the interviewer makes a passes remark about an Apple tablet, there is the mearest hint of a smile ?
Also, it's difficult for me to understand why Apple would do that when their iPhone 3G S SoC has a stronger CPU, a strong GPU that includes not just OGL ES but also DirectX9, *and* VXD (probably 370) for 40Mbps High Profile H.264 decoding. Meanwhile, Tegra1 is only capable of 20Mbps Baseline H.264 decoding, and they'd have to implement new drivers and so forth. Rather senseless, if you ask me. I'd be very confident both IMG and Infineon are still the big winners in that product. It's going to be pretty fun to see Apple surprise everyone who hasn't paid attention with 1080p playback on September 9th.
Can't say I disagree; he has had a tendency to underestimate Menlow a bit in our discussions, and it looks like Moorestown is on track to overdelivering on idle power compared to what both of us thought - but the fact of the matter is it's still a dehydrated PC architecture, and they simply aren't going to achieve traditional phone standby times before their single-chip 32nm SoC. And by then it might be too late in many ways. One problem is they'll have a significant active power *and* performance disadvantage; by the time that single-core in-order product is available, they'll be competing against quad-core A9s that will *still* beat them on price. Ouch! Does anyone really love x86 so much that they'd want to use that on anything else than Windows 7?4) Somewhat desparaging of Intel's attempts todate "dehydrating notebooks".
'His view'?! It's pretty sad that so many years after Apple has started making its own RTL code, nearly everyone still has no clue. Oh well...5) In answer a question, he indicates that it is his view that Apple designed the iphone processor, samsung manufactures it.
It's A9, but the key to understanding their performance claims is that there are at least two chips in the Tegra2 family: the handheld one and the netbook one. And then I'd guess there's also a lower-end handheld chip to be able to penetrate further down the tiers. I wonder what frequency they'll achieve; I guess that pretty much entirely depends on whether they chose 40LP or 40LPG though.6) Next year Tegras will hit x4 performance of current device, at same power (probably A9)
Yeah, that's on 28LPT and it should be it would be taping-out in 1H10. I assume that's 4xA9 and higher clock speeds. I wonder what we'll have on the GPU side... Seems a bit early for a DX11 arch derivative, but who knows whether it was developed in parallel.7) following iteration is a further x2.5 performance.