AMD headquarters
A pair of new rumors around AMD have popped up in recent days, both with significant implications for the future of the company. AMD has been in the headlines already this week for launching its new R9 Nano GPU and reorganizing its entire graphics division. Nano has had good reviews and promoting Raja Koduri to head AMD’s entire graphics division is, in our opinion, a great decision. Unfortunately, the CPU side of the business might not be so rosy.
According to DigiTimes, AMD’s new Zen CPUs will ship in Q4 2016 “at the earliest” due to problems with GlobalFoundries yield and ramp-up rates of their 14nm process. On the one hand, “DigiTimes” is not a synonym for “accuracy” in any dictionary with which we’re familiar. On the other, however, we know that GlobalFoundries has always struggled with cutting-edge nodes. Relations between AMD and its former manufacturing division were downright frosty by the time Llano launched, and while the situation has improved since then, AMD has never moved its graphics card production to GF, despite having agreed to do so when its manufacturing arm became an independent company. Last year, GlobalFoundries announced that it was canceling its own 14nnm technology (14XM) and would instead license Samsung’s 14LPE and 14LPP.
We reached out to GlobalFoundries on this topic and were told that while the company can’t comment on any customer’s products or roadmap, it could give us a general update on the state of 14nm technology. According to a GlobalFoundries spokesperson, 14nm LPE is already qualified and on its way to volume production, having met yield targets on “lead customer targets.” The second-generation 14nm node, 14LPE, is expected to qualify in the back half of this year and begin volume ramping in early 2016.
One thing GlobalFoundries did tell us, however, is that “we no longer offer 14XM.” AMD, meanwhile, showed this slide at its Analyst Day last May:
AMD-Invest
Slide from AMD Analyst Day
There’s some wiggle room in “customized process technology,” but not that much. It implies that while AMD might be willing to pay GF to make some modifications to an existing Samsung node in order to optimize it for Zen’s target power and frequency envelopes, it’s not going to deploy specialized technologies like SOI. Assuming that AMD hasn’t shifted its manufacturing to TSMC (and while we’ve seen this happen before, there’s no evidence of it currently), it means Zen will be the first high-power chip built on a Samsung process technology. It doesn’t strain credulity to think that AMD and GF might have hit some snags in this process — though, as we’ve said, anything DigiTimes prints must be taken with a heaping handful of salt.

The impact of delay

The good news for AMD is that a Zen slip wouldn’t really harm the company’s competitive position. Intel has already stated that it won’t move to 10nm until 2017, and the Kaby Lake refresh isn’t likely to dramatically change Skylake’s performance characteristics. Even Cannonlake, when it debuts, will focus on power consumption. Intel’s “ticks” typically only improve performance by 1-3%. If Zen hits its 40% IPC improvement, the impact of a delay against what Intel would be doing in the market at the same time would be fairly minimal.
Carrizo-1
The FX-8700P’s measured GPU performance. Not bad for a 15W processor
The bad news of such a delay, however, would mean that AMD would be stuck with its same APU complement for another full year. I haven’t yet had the chance to play with Carrizo, though the early reviews are positive as far as performance and power efficiency and generally on-par with lower-end Broadwell-based Core i3’s. Whether or not these results are enough to win AMD fresh designs is a very open question, and its Q2 performance didn’t exactly inspire performance in the future of Bulldozer-derived APUs. AMD might not be competitively injured by a six-month delay, but it badly needs the revenue new products could bring.
While AMD does expect to begin recognizing additional revenue from new semi-custom wins in 2016, Lisa Su has previously said that this revenue would be between $250 – $300M for the lifetime of each product. Assuming a five-year lifespan for each and the maximum estimated value, that’s still just $30 million in revenue per quarter from the new semi-custom wins. That’s a great residual income, but it can’t replace the value of lost CPU market share. Even if it’s true that AMD captured the Nintendo NX, it won’t start seeing revenue from that business until the end of 2016 at the earliest.

Another day, another buyout rumor

Meanwhile, in other, other news, Fudzilla is reporting that Microsoft is still nosing around AMD. Given that the same site reported that Silver Lake Partners bought 20% of AMD just a few days ago, it’s almost comical to see this kind of rumor circulating. That’s not to say that Microsoft couldn’t make its own investment, but it shows how fragmented the rumor mill can be. At the same time, however, it makes a certain amount of sense. Earlier this year, we reported that Lisa Su had met with investment firms to discuss possible options for AMD, including restructuring or spinning off parts of the company. The recent GPU restructuring might or might not be part of a deal,
At this point, we recommend taking all buyout / investment rumors as fan theorizing as opposed to established fact.