The idea that AMD might license its graphics IP to Intel for inclusion in upcoming Intel CPUs sounds like the kind of headline we might write as an April Fool’s joke — but there’s a rumor that this may, in fact, be in the works. The question is, what kind of product would Intel wind up building, and when might it hit market?
This rumor comes courtesy of [H]ardOCP’s Kyle Bennet, who writes that “The licensing deal between AMD and Intel is signed and done for putting AMD GPU tech into Intel’s iGPU.” So why do we expect this could well be true, when it’s such a shift from the status quo? Several reasons. First, the $ 1.5 billion license agreement between Intel and Nvidia is drawing to a close, and there’s no sign that Intel is going to renew its licensing with Team Green. Intel took a baby step towards AMD’s side of the fence when it announced that future Intel GPUs would be FreeSync-compatible, as opposed to licensing G-Sync technology from Nvidia. Second, there’s the fact that Intel probably needs an IP license from either AMD or Nvidia, given how much collective IP the two companies have in this space.
There are several ways this could play out in-market. One option is that things continue to be business as usual: Intel signs a license and continues to develop its own graphics hardware. This is probably the simplest option with the smallest upside for AMD, but it also doesn’t give Intel much room to maneuver. Intel has included on-board graphics IP on its motherboards and CPUs for 17 years, but for most of that time its performance and compatibility ranged from “nonexistent” to “utterly terrible.” It wasn’t until the launch of Sandy Bridge, in 2011, that Intel started taking graphics more seriously. Even now, Intel’s GPU hardware and driver support lag behind both AMD and Nvidia, though the situation has improved markedly over the past few years, especially in mobile. An IP license could give Intel a leg up in certain areas, depending on what kind of chip the company wants to build. But it doesn’t really change anything about Intel’s relative position in the market.
Would AMD help Intel build a custom GPU?
The more intriguing option is that AMD could build a custom GPU for Intel. This would truly be uncharted waters for both companies — AMD and Intel have never cooperated on an effort of this sort, and it would have risks for both. Intel would be betting that AMD would design a better chip than they themselves could field, while AMD would be betting that Zen is strong enough that it no longer needs to lean so heavily on its GPU strengths in the market.
Part of what makes this more likely, at least from the AMD side of things, is that leaning on the strength of their integrated GPUs clearly hasn’t helped AMD much to date. AMD’s total APU business has collapsed compared with where it was five years ago. In Q3 2016, AMD recorded CPU, APU, and GPU sales of $ 472 million. In Q3 2011, AMD sold $ 1.286 billion worth of CPUs and APUs, while graphics alone accounted for $ 403 million. Losing more than three-quarters of your market revenue in five years is truly disastrous, and console revenue is the only thing keeping AMD alive right now.
That fact, more than anything, may account for AMD’s willingness to sign a deal with Intel. A simple licensing deal could still be worth $ 150-$ 200 million per year, while the royalties from a major GPU development project could be worth significantly more. AMD needs that revenue if it wants to compete with Nvidia, launch itself into the HPC space, or continue to advance its Zen CPU designs. Intel, meanwhile, may see an opportunity to appeal to a larger market segment and make processor graphics a more appealing upgrade opportunity. It’s also got die size to spare — Intel already dedicates more of its die space on desktop processors to graphics than to x86 performance, and Skylake’s quad-core die size is a svelte 122.4mm sq.
If AMD has inked a deal like this, we expect to hear about it sooner rather than later, though it could take a year or more before we see shipping silicon. The nature of this type of rumor is that it raises more questions than answers — there’s no word on how Intel’s GPU would evolve over time compared with AMD’s technology, for example. AMD’s decision to structure its Radeon Technology Group as a subsidiary could make it easier to line up this kind of deal, since it’s not as tightly affiliated with AMD. But we don’t know if the deal would cover older GCN technology, extend to AMD’s newer Vega architecture, or even call for the development of a new GPU core built explicitly to Intel’s specifications.