ANALYSIS: nVidia has the IP to license but what about the ecosystem?
nVidia’s decision to license the IP cores for its GPU, visual computing and – possibly – LTE modem technologies is something of a surprise, even if you can see where the company’s coming from.
However, in his blog detailing the formidable technology nVidia is now releasing, EVP David Shannon doesn’t once mention the word ‘ecosystem’. Yet this may be the biggest challenge his company faces in becoming a fully-fledged IP as well as a silicon player.
To displace Imagination Technologies, the GPU IP leader right now, and/or ARM (either at the Mali GPU level or by arguing for GPU over CPU) is a tough one politically.
Imagination has both Intel and Apple as shareholders to begin with. Non-investor licensees include Samsung, Qualcomm, MediaTek, Renesas, Sharp, Texas Instruments, NXP, Freescale, Marvell and more.
ARM is simply ARM, the de facto standard CPU player for anything portable. Its aggression both in cutting power and moving up the performance scale toward 64bit are there for all to see.
But beyond that, both those companies have not only assiduously courted end-users but they have also built formidable networks of IP, tools and foundry partnerships.
ARM’s is the most striking. It’s almost easier to name the vendors who are not part of its Connected Community than those who are.
But Imagination too has been extremely active. In Synopsys’ just-launched DesignWare HPC Design Kit, Imagination is the star partner. It is the company in the featured case study as to how the tool optimizes across power, performance and area.
nVidia’s chicken-and-egg conundrum
The leading tool vendors all know nVidia well. As suppliers to the company’s design teams, they are familiar with implementations of its IP up to and including the Kepler architecture that will be the first nVidia is to license. nVidia also has a long history with TSMC as a foundry (though it got somewhat fraught at 28nm and its TSMC-only policy to date will raise questions about implementation on other foundry platforms).
However, the numbers that nVidia is targeting – its self-defined benchmarks are the money it made off a license for PlayStation 3 and $250m a year in visual licenses from Intel – now mean that it must publicly assemble its own comprehensive ecosystem.
Even the big customers expect to see those in place today. Consider, Sony, Toshiba and IBM made Cell a money-no-issue ground-up project, but for the PlayStation 4, Sony has retreated to the integration of off-the-shelf (if still bleeding edge) technologies.
Beyond those Tier Ones, there is also big money in platform integration, particularly in the Asian and sub-continental markets. Here again, customers expect to be able to work within pre-defined multi-vendor flows that effectively allow them to concentrate on their own packets of differentiation and keeping the cost down.
We can be pretty sure that nVidia has flagged its plan to the leading tool vendors and foundries – and given nVidia’s clout they’ll listen more closely than they will do for a start-up. But even then, there’s a chicken-and-egg element. These suppliers optimize processes for particular IP blocks when they see market traction for them – however, for the reasons above, it’s hard to get that traction if you don’t have an ecosystem.
nVidia’s strength here is that, for now, its core silicon business may face challenges – rightly identified in the PC space and elsewhere by David Shannon’s blog post – but it does have the revenues to play a long game as the market shifts toward IP.
Also, the company does understand the need for ecosystems. For example, it has built a substantial one to support its CUDA parallel computing architecture. It just has to do the same now to support mainstream chip design (and it may have to develop some of the tools there internally, but, again, it has done that for CUDA).
Still, the GPU space is highly competitive and hugely cost-sensitive. This is as much a gamble as a recognition of technology trends.
Which, in closing, makes nVidia’s hints about its LTE modem technology and the possible licensing of that very interesting indeed. That is arguably more of a ‘greenfield’ play through which it could build partnerships and alliances that extend back upstream into GPU… perhaps.
Certainly, having shown its hand – and rather suprisingly around E3 rather than, say, at DAC – nVidia’s stock market analysts will track not just the roll-out of the IP itself but also the support for it. And they are an impatient bunch.
