Pages: 1 2
The biggest problem for AMD is that it is exceptionally unlikely that they can compete in the smartphone market, even with an ARM license. Simply put, do not have some of the essential building blocks to create a compelling SoC. Critically, AMD has no baseband processor, and are missing some of the other platform elements such as DSPs, wifi and power management.
To put this in context, Nvidia spent $367M to buy Icera, a startup with a nice programmable baseband processor, but nearly no presence in phones. Intel paid $1.4B for Infineon’s successful wireless division, a bargain available only because the parent company was restructuring. AMD is still $2B in debt, and is unlikely to be able to raise the capital necessary to purchase a 4G baseband company.
Moreover, AMD’s current CPUs and GPUs are far higher power consumption than appropriate for mobile phones. It is unlikely that they could have a plausible offering before the 20nm node and it might take another year beyond that. Given that TSMC is just ramping 28nm at the start of 2012, 20nm production should be in early 2014.
Using an ARM-based SoC only makes the problem worse. From inception to production, an SoC typically takes 3-5 years, plus more time for FCC certification and carrier qualification. So even if AMD’s strategy was to adopt ARM, they would not have revenue for 4 years or more. Even then, the first generation products to enter any market are exceptionally unlikely to turn a profit. Take Nvidia for example, the first generation of Tegra was basically an opportunity to engage with customers and had no immediate gains for the company.
Last, entering the smartphone market would require a huge investment in software and be a significant distraction from the core PC business. It would be an even larger challenge if they were to shift from x86 to ARM, a software ecosystem that AMD has little experience with. Given the gap in AMD’s capabilities, the implied product timeline and the overall risk, smartphones seem beyond reach for the moment.
Unlike smartphones, tablets are favorable from both a software and hardware standpoint for AMD. The market is rather new and fluid. Apple’s iOS is the dominant software platform, but their SoCs are all proprietary ARM designs. It is an open question whether Android or Windows 8 will become the commodity OS for tablets.
Since Android now supports x86, the software ecosystem favors x86, if anything. While Intel does much of the work ensuring Android/x86 works smoothly, AMD can reap some of the benefits. AMD is a longstanding Microsoft partner, and has a deep understanding of the Windows ecosystem. For instance, AMD’s experience with DirectX is a distinct advantage compared to vendors like Qualcomm, Samsung or TI. Moreover, Windows 8 on x86 will have much stronger compatibility with existing applications, another benefit for AMD.
On the hardware side, AMD’s 40nm Bobcat SoCs are great for low-end notebooks, but consume far too much power to be viable in most tablets. The SoC itself is 6W and the I/O hub is another 1-2W, and also includes interfaces that are really optimized for PCs, rather than tablets. An upcoming 40nm variant that has been targeted for low-power will reportedly achieve about 5.5W TDP for the SoC and I/O hub combined.
Since TSMC’s 28nm process will use high-k/metal gates, the static and dynamic power consumption should drop substantially by virtue of manufacturing alone. If AMD were to integrate everything into a single chip, and further optimize for power, it’s entirely conceivable that they could reach 2-3W on 28nm. That puts AMD in the right position for tablets. Previously, AMD’s roadmap hinted at such a design in 2013, but this could be pulled in earlier.
The last point about x86 in tablets is that AMD’s growth can be entirely organic, evolutionary and synergistic with the core PC business. There is no large R&D expenditures required to master a new instruction set and new CPU designs. Nor will they be faced with the challenge of investing in a product line with minimal revenue as it grows into a new market. Much of the effort that goes into an x86 tablet can be re-used for low-power notebooks.
Looking at AMD’s situation objectively makes it clear that adopting ARM is exceptionally unlikely.
AMD is not in a position to enter the smart phone market, so realistically ARM would only help in tablets. However, AMD already has an x86 tablet solution that should be quite competitive in a generation or two. AMD’s CPU designs are quite reasonable and their graphics are excellent, but the most unique aspect of the company is their deep competency in x86. Given the limited potential benefits, there is no reason to sacrifice such unique expertise. Moreover, AMD’s competitive position would be weaker in the ARM ecosystem and such a shift would require an unaffordable investment and significant risk to their core business that focuses on the x86 market.
A much more realistic scenario is that AMD stays the course with x86 and puts greater emphasis on low-power SoCs for tablets. Based on recent rumors, some of AMD’s 2012 low-power SoCs have been cancelled, and the most likely explanation is that they opted to put extra resources into products originally slated for 2013, codenamed Samara. The goal is presumably a 28nm SoC that includes tablet-optimized I/O. The number of cores is likely to scale from 1-4, with graphics also scaling to address different parts of the market. The extra resources may be used to shift Samara to an earlier schedule (perhaps late 2012), to more optimize the products for lower power or to target a wider swath of the market. In all likelihood, it is a combination of the three.
Discuss (338 comments)