CHIP DESIGNER Nvidia's decision to license its Kepler GPU architecture to other companies is remarkable, given the firm's historical stance on licensing, and it highlights how hard it is for companies to break into parts of the semiconductor industry.
In March, Nvidia CEO Jen-Hsun Huang told financial analysts and journalists that the firm would spend $600m of its research and development budget on Tegra in 2013, because that is the product that serves the fastest growing market. Huang's comments made business sense but what he didn't say was that Nvidia's own products won't be enough to offset the years of investment needed to compete against entrenched rivals.
Nvidia's Tegra chip got off to what can only be described as a disastrous start, perhaps best signified by the fact that its most visible design win was Microsoft's Zune media player, which itself was hardly a beacon of sales and marketing success. The firm deserves some credit for pushing Tegra 2 and Tegra 3 to break boundaries such as being the first dual-core and quad-core system on chip (SoC) designs for smartphones and tablets, but aside from Google's Nexus 7 there have been relatively few high volume design wins for the firm.
Although Nvidia would have made the decision internally to consider licensing its architectures many months ago, the advantages of its announcement are perhaps best illustrated through its Shield handheld device.
With Shield, Nvidia thought it would be a good idea to build a mobile device that could showcase the graphics capabilities of Tegra 4 and bring forward what is arguably the firm's unique selling point when it comes to embedded processors, the ability to stream games from one device to another, which in the case of Shield is a PC running a Kepler based graphics card.
Nvidia, however, has made a dog's dinner of the Shield development process, which could come to overshadow what is decent product and a competitive processor.
Nvidia's decision to delay Shield shipments a day before the announced release date smacks of a company that is grappling with supply chain issues. The problem, as Nvidia is now finding out, is that with a supply chain that is largely handled by board vendors such as Asus, Gigabyte, MSI and EVGA, Nvidia is insulated, to a large degree, if there are delays in getting graphics cards onto retail shelves.
However with Shield, Nvidia is overseeing everything, from chip design, integration, testing and final packaging and that means any negative development, such as the delay, hits the brand full in the face. The other problem Nvidia faces is that Shield, while a nice bit of kit, is overpriced for what it is, even with the $50 drop in price.
Microsoft had to face significant criticism when the Xbox 360 red ring of death debacle unfolded, but unlike Nvidia, Microsoft knew that getting into the games console business would be a decade long money pit. And unlike Nvidia, Microsoft had billions to throw at a problem for a decade and thanks to the way it engineered Windows and Office sales it wouldn't fundamentally affect the company's day-to-day business.
Nvidia on the other hand has now felt the pain of trying to break into the smartphone and tablet market even when it pushes new technology ahead of its competitors. The firm is also feeling the pain of trying to build and sell devices itself, a plan that even before its latest setback sounded like a high risk manoeuvre for something that simply isn't going to be Nvidia's bread and butter in the long term.
The problem with Nvidia's decision to license its graphics technology to mobile device makers is that for all its discrete GPU technology, the firm hasn't truly proven itself in the mobile market and this is evident in the total number of design wins from the three previous generations of Tegra SoCs. And unlike discrete GPUs, Nvidia is going up against entrenched GPU technology firms such as Imagination, ARM and more recently Qualcomm.
Imagination in particular is the GPU vendor of choice for a number of big name chip designers including Apple and Intel, while ARM is pushing its Mali GPU technology very hard. Qualcomm's GPU business, which it largely bought from AMD, isn't licensed but given the market presence of Qualcomm, it is a major competitor to any GPU vendor wanting to crack the mobile market.
Nvidia's decision to license its GPU technology has, with the benefit of hindsight, come years too late. This could be because the firm didn't have an architecture that it felt could scale down to the power requirements of mobile processors, or it could be that the firm believed it could work with smartphone and tablet makers in the same way it deals with graphics cards partners.
The problem Nvidia will have is that firms are increasingly looking for a single vendor for both CPU and graphics designs and how much Nvidia is willing to license will have a big bearing on the number of firms that decide to use its technology. ARM and Imagination go to considerable lengths to not just design their CPU and GPU architectures but to make sure their partners can interconnect the two units and make them easy to fabricate, something that Nvidia hasn't had to do in the past because it handled everything up to the fabrication of its GPUs and SoCs with TSMC.
All of this doesn't mean that Nvidia cannot compete with Kepler and future architectures in the mobile market, and it is simply wrong to judge final performance by simply scaling down from discrete Kepler GPUs using nominal performance per watt figures.
However Nvidia's success in licensing its GPU technologies will be more than merely making Kepler available to those that sign a nondisclosure agreement, the firm will have to put considerable effort in building the support structures around the architecture so potential customers can design and produce chips that are cost effective as well as perform well against the formidable competition. µ