A new report today indicates AMD may be teaming up with MediaTek to bring its Radeon graphics solution to tablets and smartphones. If true, it could open new markets for AMD’s hardware and fix what many have believed was a significant mistake.
Back in 2009, AMD negotiated the sale of its Imageon division to Qualcomm for $65 million. Qualcomm later took the Imageon line and rebranded it as Adreno (Adreno being an anagram of Radeon), and turned the segment into the cornerstone of its custom SoC division. AMD has taken a great deal of heat over the years for selling the business segment, but I’ve never been personally convinced it was a bad move. While it’s possible that the company could’ve earned better returns from continuing to license its IP to Qualcomm, it’s also possible that Qualcomm wasn’t interested in continuing that arrangement, or that AMD simply didn’t have cash to spare to drop into mobile graphics. Selling a business unit at a modest profit is better than hanging on to a segment you can’t afford to develop until the IP is outdated and useless.
Regardless of what the company might have done differently six years ago, this new deal could mean AMD is preparing to throw its hat back into the smartphone and tablet ring. Now, if you’re familiar with Radeon’s latest desktop incarnations, that might seem a bit odd — the company’s hardware is notably less power efficient than Nvidia’s Maxwell, after all.
AMD’s Mullins hardware (that’s the second-generation Kabini) have TDPs of ~4.5W and are meant for tablets to start with. That 4.5W TDP includes both the CPU and GPU, with 128 cores, 8 texture mapping units, and 4 render outputs. It should be theoretically simple for MediaTek to pair a Radeon mobile GPU with an ARM core of their own design.
If AMD’s Project Skybridge has advanced to any degree, it seems likely that the company has experience with pairing ARM Cortex cores and its own Radeon GPUs as well. The entire point of Skybridge was to build a swappable platform that x86 and ARM cores could both interface with — and that process is simplified if both SoCs have the same graphics engine. Toss in the fact that AMD has further refined its SoC implementation of its own GCN architecture by pulling down overall power consumption, and you’ve got a plausible case for how AMD would drop a mobile Radeon core into an ARM SoC and license that architecture to a third party.
The big question about how AMD would license its hardware to third parties actually doesn’t revolve around the GCN architecture at all, nor its implementation. The sticking point would be drivers — and whether AMD could provide a top-notch experience under Android.
Linux performance and compatibility has always been dodgy for AMD when compared with Nvidia or Intel. AMD, to its credit, has apparently been laying plans to improve its overall state of Linux support, but that’s a long-term initiative. Linux, of course, isn’t identical to Android, but the ability to write good driver code in a non-Windows platform is still critical to any licensing initiative. This is an issue that’s tripped up larger companies; Intel’s Bay Trail for Android was reportedly months late due to issues with the Android GPU drivers.
If AMD can provide the software, however, a MediaTek partnership could make a good deal of sense, particularly if that arrangement included support for features like HSA or helped push APIs like Vulkan into the mainstream. For now, it’s just a rumor, but with Nvidia largely pulling out of mobile, AMD may feel there’s an opening it can step into. Lisa Su, AMD’s CEO, has talked about continuing to pursue new market opportunities for embedded products — this could be one of the business segments she had in mind.
Source: Extremetech
