What does everyone think of the Tegra Road Map

SoC vendors have been making so many bombastic claims about GPU performance improvements that you have to take it with at least a little salt. I believe PowerVR has made similar claims about their longer term roadmap as well. The 2014 goals will be relying on at least 20nm, but quite likely ~15nm. If TSMC and/or the IBM/GloFo/Samsung alliance fail to deliver that by then then you can expect all of these SoC roadmaps to be pushed back. And of course with each new node there are significant new challenges, which isn't to say that they won't be overcome but inevitably things have to slow down.

One thing is that it's evident that nVidia is moving their ARM SoCs up - you can see this is Project Denver's deployment as a GPU control processor, but also in Tegra 3. At 40nm (same as Tegra 2) but with twice the cores and some significant increase in GPU capability it has to consume much more power, at least at peak. That isn't to say it's unsuitable for mobiles, since it'll be similar to what's in the NGP, but the typical current phone batteries might not cut it. A 75x power increase in only 3 years will also surely not come with the same power consumption of Tegra 2's GPU, although it should still be more efficient than the desktop models it'll be displacing in performance. It might still work out for tablets, which is going to be the main differentiating line.

One thing I don't really get is nVidia's preoccupation with pushing the performance of their video playback. I can see wanting to meet Bluray specs at 50mbit h.264 1080p so it won't get beaten out by CE4100 anymore but all of this marketing for 1440p and whatever seems silly. They should wait for industry demand for such a thing; personally it looks to me like a hard sell, since the subjective improvement from 1080p to 1440p is surely nothing compared to that of 480p to 760p to 1080p. nVidia is even talking about 2160p, and I just figure there has to be a point of diminishing returns. Doing all of this in an SoC targeting phones and tablets primarily is just needlessly showing off.

Speaking of roadmaps, this recently announced SoC looks very nice:


Should give OMAP5 a run for its money in the GPU department. Let's see if ST-Ericsson can actually release it in anything resembling the same timeline as OMAP5. nVidia's 1 year cadence with Tegra is impressive, but I doubt we'll see Tegra 4 products before the end of 2012 - with any luck the announced Cortex-A15 SoCs will beat it by a few months and look attractive compared to Tegra 3. But this probably requires a much faster turnaround than there has been with OMAP4 or U8500.