As I watched NVIDIA's (NVDA) CES press conference, I had expected something interesting but no show-stoppers. However, I was (as an NVIDIA shareholder) pleasantly mistaken. The company announced a very compelling 32-bit ARM (ARMH) based SoC with an insane amount of graphics performance and a respectable quad Cortex A15 for devices shipping in the first half.
However, what I thought was particularly compelling, was that NVIDIA announced that it would be shipping a version of this very graphically-powerful SoC coupled with two of its Project Denver CPU cores for the second half of 2014. This is a 64-bit core (and a very wide, very high performance core), and CEO Jen-Hsun Huang showed it running a 64-bit version of Android.
What Does This Have To Do With Intel?
One of the big head-starts that I believed Intel (INTC) would have, courtesy of the fact that its cores were already 64-bit capable, was in getting a 64-bit version of Android out. Now, I think Intel will still have somewhat of a head-start over the other ARM vendors (sans Apple), but with NVIDIA's announcement that it would be shipping its Denver-based Tegra K1 SoC for devices in the second half of 2014, that advantage is diminished significantly. The window shrinks from perhaps a year to about 6-9 months.
More chillingly for Intel, though, is that NVIDIA's Tegra K1 is likely to be substantially faster than Intel's current generation "Bay Trail" SoC in its 32-bit incarnation. The 64-bit version (which features two "big core" Denver processors) is likely to clean Bay Trail's clock on a per-thread basis in a manner very similar to what Apple's (AAPL) Cyclone cores did (except perhaps worse). It also goes without saying that NVIDIA's Tegra K1 (either 32-bit or 64-bit) will have dramatically better graphics performance than Bay Trail.
Cherry Trail Is Intel's Best Hope
At this point, it is unlikely that Bay Trail will be competitive with NVIDIA's Tegra K1, so the spotlight turns to "Cherry Trail" - the first 14 nanometer SoC which features four "Airmont" CPU cores and a substantially beefed up/redesigned GPU core (and four times as many as in Bay Trail).
This, too, is a "second half" of 2014 part as is the 64-bit version of Tegra K1, although its unclear just how late in the second half of 2014 we're talking. If it's as late as Bay Trail was, and if NVIDIA can execute well here, then this will be a disaster and Intel investors will have to wait for Broxton to really have a chance of a truly competitive part. However, even if Tegra K1 64 bit is late, K1 32 bit will still have a fairly monstrous GPU, dwarfing the Bay Trail GPU and likely putting the hurt on Qualcomm's Snapdragon 805, to boot.
Intel needs a better chip than Bay Trail out to compete with Tegra K1. Cherry Trail looks the best suited to do so although Moorefield (the mid-life kicker to Merrifield) could potentially put up a decent fight, too, depending on how aggressively Intel is pushing the graphics performance there (i.e. which GPU IP block it's using from Imagination Technologies). The bummer for Intel, though, is that even Moorefield is a Q3 2014 affair, so unless it's super-early Q3, it probably won't show up in time to even have a hope of fighting K1.
I await Brian Krzanich's opening CES keynote slated to be delivered at 6:30 PM (Pacific Time) on Monday. While I was pleasantly surprised by NVIDIA's keynote as a shareholder, the bar is now set exceptionally high for Intel, of which I am also a shareholder. I think that if Intel doesn't move more quickly than the roadmap it outlined, and if NVIDIA's Denver cores really are Core class rather than Atom class, then Intel's life just got a whole lot tougher.