Over a year ago, Nvidia sued both Qualcomm and Samsung, alleging that they had infringed on its GPU patents and requesting that the International Trade Commission (ITC) ban all imports of devices containing ARM Mali, Adreno, or PowerVR processors. On Friday, ITC administrative judge Thomas Pender handed down a ruling slapping down Nvidia’s claims of patent infringement. Of the three patents that Nvidia claims Qualcomm and Samsung violated, two were ruled non-infringing, while the third was ruled invalid under the so-called “obviousness” test.
This marks a significant setback for Nvidia’s legal efforts, though the company noted in a blog post that it intends to ask the full commission to review the decision and not to throw out the third patent (the 6,690,372 patent). That patent covers “shadow mapping while rendering a primitive in a graphics pipeline.” Pender’s decision (Google cache) found that while Adreno, PowerVR, and Mali GPUs violate claims 23 and 24 of the ‘372 patent, the patent is invalid because the method described in Claims 23 and 24 isn’t novel. Those two claims describe a system in which a texture module is connected to a shader module and retrieves data using texture coordinates.
Pender’s ruling suggests that Nvidia’s construed patent claims were too broad to be valid, and while the company can appeal the decision to the full ITC, that body’s rulings typically reflect the determination of the original presiding judge.
The fundamental problem with Nvidia’s patent assertions — the company claims to have invented the GPU, programmable shading, unified shaders, and multi-threaded parallel processing in GPUs — is that all of these can be credibly claimed to have existed long before Nvidia came on the scene. It’s true that Nvidia implemented the first hardware T&L engine in a consumer GPU, for example, but that doesn’t mean Nvidia is legally entitled to financial compensation for it. I’m not a patent lawyer and don’t pretend to be, but I suspect that patents filed by companies like SGI could be used to argue against such broad application of Nvidia’s patent portfolio.
As for the reason why Nvidia is aggressively hunting for new licensing revenue, well, that’s thanks to Intel. Back in 2011, Intel and Nvidia signed a multi-year patent agreement in which Chipzilla would pay Team Green $1.5 billion over five years, beginning in 2011. That agreement has tacked a healthy $66 million per quarter on to Nvidia’s bottom line, and it’s cash that the company wants to replace with new licensing agreements. Unfortunately, those agreements don’t seem to have materialized. Intel has announced that it will support Adaptive Sync rather than licensing Nvidia’s G-Sync, while this new decision will hamper any chance of coming to terms with Samsung and Qualcomm.
The $66 million per quarter that Nvidia receives from Intel isn’t a make-or-break for the company, but the chart above shows just how nice a feather it’s been. At ~$264 million per year, Nvidia’s net income has been significantly larger than it would’ve been otherwise. Given the hammering that Tegra has taken over the same period, one could argue that patent licensing cushioned the impact of the automotive and deep learning pivots and away from the consumer smartphone and tablet markets.
Samsung is still bringing suit against Nvidia for its own alleged infringements of Samsung patents, but we expect both of these cases to likely vanish into the ether. Samsung’s lawsuit was a response to Nvidia’s initial, and Nvidia is fishing for licensing revenue — not seriously defending 11 year-old patents from supposed infringement by a company that uses third-party GPU IP rather than developing its own. It may have made sense for NV to bring suit against Qualcomm, since Qualcomm develops its own graphics hardware, but Samsung licenses other solutions and doesn’t build its own architectures.