At its financial analyst day last week, Nvidia spilled some details on upcoming projects that are worth looping back and catching up on. Nvidia is one of the rare tech companies openly cheerful about the last quarter – and announced that it would return $1 billion (£660 million) in cash to shareholders this year thanks to a mixture of stock buybacks and dividend payments. Mobile technology is one of the major reasons Nvidia is excited about the next 12 to 18 months, so let’s talk about the Tegra family.
Tegra 5 (Logan) in 2014, Kayla coming soon
Almost a year ago, I predicted that Kepler’s enormously improved performance-per-watt made it a good bet that we’d see the GPU debuting in Tegra 4. The idea was good, but the timeframe was off. Kepler DNA will first come to market with Nvidia’s Logan, Tegra 4′s successor, currently expected in 2014. Well, 2014-ish. Like most of its rivals, Nvidia’s ideas about when a part is actually launched are rather fuzzy. Widespread availability tends to lag behind the actual launch announcement by at least six months.
Nvidia is launching a new SoC development product, codenamed Kayla, that combines a Tegra 3 SoC with a PCI-Express slot and a Logan-class GPU that will debut in coming months. Developers that want an early look at the mobile Kepler’s capabilities will be able to buy the kits to do early prototype work. Logan’s CPU is unspecified, but the chip is almost certainly a refined version of the Cortex-A15 that Nvidia has already discussed for Tegra 4.
Logan’s successor, Parker, will be the first chip built on FinFET and the first iteration of Nvidia’s long-delayed Project Denver. It’s an ARMv8 64-bit core with an integrated GPU based on Nvidia’s Kepler refresh, Maxwell. The 2015 timeframe is ludicrously optimistic, given that TSMC’s CEO Morris Chang has said that he expects very small volumes of 16nm FinFET production in that time period. Barring a major change in TSMC’s roadmap, 2016 is a more likely timeframe.
Tegra has consumed hundreds of millions in R&D costs to date, but to hear Nvidia tell it, it’s been a vital investment. Jen-Hsun Huang, the company’s CEO, claims that Tegra puts Nvidia in a position to disrupt the PC market and secure huge revenue streams for itself while technologies like GPU Grid make server industry inroads. The company’s segment revenue breakdown between 2010 and 2013 makes it hard to argue with how important Tegra has been.
So what about CUDA?
CUDA: Fading star or phoenix?
One of the reasons I’m excited about a Kepler-derived mobile GPU is because I think Nvidia’s physics (PhysX) technology could drive some really amazing mobile experiences. With that said, it’s hard to take Jen-Hsun’s claim that CUDA is a “major driver of GPU growth” seriously. Nvidia chains the argument together as follows:
Because NV GPU CAGR (compound annual growth rate) is at 12 per cent at a time when AMD has only gained a fraction of that amount and the PC industry has been flat, therefore CUDA/Tesla have been enormous GPU growth drivers. That’d be a lot more believable if it weren’t for the graph I showed preceding this section. CUDA matters to the Tesla market, and Tesla is important to Nvidia, but CUDA doesn’t mean jack to the GeForce business.
Nvidia’s GPU-accelerated PhysX effects were implemented using CUDA, and the now-defunct Badaboom encoding software used CUDA. Team Green has never come out and said that PhysX is dead, but that term doesn’t appear once in any of the slides shown at Analyst Day. We know that the software version of PhysX – the middleware engine that can run on multiple game consoles – will continue to be developed and Nvidia has pledged PS4 support, but the number of new hardware-accelerated PhysX titles has shrunk to a trickle.
Frankly, I hope to see a mobile SDK that breathes new life into the standard, but CUDA as a brand isn’t driving GeForce adoption. Great performance in games has driven GeForce adoption.
As for Tesla, not much has changed since the last time we discussed the relative merits of Xeon Phi vs. Nvidia’s K20/K20X. This continues to be a two-way race; AMD is scarcely a shadow in the bid for Top 500 systems or other supercomputing endeavours.
Outside of specialised environments or a particular mobile implementation, CUDA isn’t going to win unique support. And that’s okay. Nvidia has always hedged its bets as far as CUDA was concerned; the GeForce cards are fully OpenCL-compatible.
Well-positioned against all comers
A few years ago, when AMD and Intel were talking up combined CPU + GPU products, Nvidia’s lack of an x86 license was seen as a serious liability that could wind up killing the company. That’s rather obviously not the case today. Nvidia still faces major challenges from Intel with Xeon Phi, the on-going integration of increasingly potent mobile GPUs, and AMD’s Radeon Fire Pro businesses. Qualcomm and Samsung are scarcely going to roll over and die, and Intel has sizable long-term plans for Atom.
Looking at Nvidia’s total business structure, the company is on a good footing. It has strong businesses in multiple segments, and the decision to branch into tablets and smartphones was clearly the right move. Tegra 2 wasn’t a fluke, Tegra 3′s 40nm design and quad-core A9 didn’t hurt it in the market place (even if quad-cores remain of questionable utility in smartphones), and the upcoming Tegra 4 looks like an impressive piece of silicon. If Nvidia continues to execute the way it has to date, and picks up a few high-profile tablet wins for Tegra 4, it’ll continue to carve out new business for itself in some of the hottest growth markets in technology.