Business, Graphics, Hardware

BSN Exclusive: AMD Working on ARM SoCs for Consumer Products

This may not come to many of you as much of a surprise, but AMD is already working on developing ARM processors for consumers. These chips will utilize ARM?s CPU IP and AMD?s GPU IP to create a mobile SoC for AMD?s consumer devices, likely starting with tablets, AMD?s most difficult market. There are teams of people at AMD working on making these consumer ARM CPU-based chips a possibility. If you look at all of the supporting evidence behind such a strategy, it only makes sense, so let?s do that.

First and foremost, AMD is already an ARM licensee on multiple levels, so their relationship with ARM, while new, is already pretty deep. Personally, I?ve seen them at almost every ARM event so far and I with this development, I don?t expect that to go away any time soon. Currently, AMD is a licensee of ARM?s Cortex A57 cores for embedded/server as well as the Cortex A53, for what I assume is a big.LITTLE implementation or for a cost-down solution. In addition to these Cortex A5X series chips, AMD is also a public licensee of the Cortex-A5, which they plan to use in their x86 mobile SoCs for the TrustZone security implementation.

This move is both a good one and a bad one for AMD depending on your opinion of the company. They have been an x86 licensee for decades now, and the idea of them going away from x86 could frighten some investors. The truth is that ARM and Intel are constantly having battles between each other in terms of performance and power efficiency and a real winner remains to be decided. However, it is clear that ARM?s low-power approach has won them many licensees and an expansive market of billions of chips being shipped. Even though Intel is steadfast in their x86 CPUs, they have found ways through process technology to make themselves more competitive with ARM than they have ever been before.

However, if you ask around, there is a general consensus that AMD?s CPUs are their weakness and if they want to be competitive in tablets, they need a better lower power CPU. Currently, they are preparing to release their Beema and Mullins SoCs, which feature AMD x86 cores and AMD Radeon GPU cores. These are replacements for their Kabini and Temash APUs which had moderate to poor pull within the consumer market. These x86 SoCs remain 28nm but are updated with marginally faster Puma CPU cores and an AMD Security Processor (The Cortex-A5 TrustZone). I would expect AMD?s ARM CPU-based solutions to be replacements for their Mullins APU SoC as they currently claim an SDP of ~2watts but likely at a much lower performance level. With lower power ARM CPU cores, AMD could give themselves more room on GPU performance and still maintain relatively good CPU performance.

Also, keep in mind that in 2012 AMD hired a certain Mr. Jim Keller onboard to be their chief of their processor group. As many already know, Jim Keller was originally an AMD employee in the late 90?s and helped co-author the HyperTransport spec as well as the x86-64 processor instruction set and the CPUs that utilized it. He then ended up working for PA Semiconductor which eventually got acquired by Apple and worked on Apple?s A4 and A5 SoCs. He clearly has a lot of experience in building ARM SoCs and I wouldn?t question for a second that they are using his knowledge and expertise to build their own ARM CPUs, most likely utilizing ARM?s v8-A architecture. The reason for that is because AMD is already working with ARM v8-A in the Cortex-A53 and Cortex-A57 and it would make sense to start with the same architecture across all of AMD?s ARM SoCs. That is, if they are actually building their own custom ARM SoC.

Now that we?ve looked inside AMD, let?s see what kind of competition they?ve got. In the high-end, you?ve got companies like Qualcomm, Apple, Nvidia and Samsung all of whom have absolutely gigantic presence in the market and have done so for quite some time. Then you?ve got up and coming companies like MediaTek, Rockchip and AllWinner that deliver the value proposition and will undercut your pricing like no other. The convenient thing for AMD, though, is that they already have pretty prominent positions within both TSMC and GlobalFoundaries and have the ability to utilize some of their allocated capacity to make some pretty good SoCs. You also have companies like Marvell and Broadcom that have kind-of sat out of the consumer ARM SoC segments, but clearly still want to play ball. It clearly won’t be easy for AMD, but with the people and graphics IP they’ve got, it’s entirely possible. Especially if they implement HSA as one of the features of their ARM based SoCs.

Also, keep in mind that mobile data connectivity is playing an evermore important factor in smartphone and tablet design wins. While connectivity is currently more important in smartphones, it will be a matter of time until that becomes true with tablets as well. Currently, Qualcomm has a stranglehold on this field, but Intel, Samsung, Nvidia, Broadcom, MediaTek and Marvell have already developed or purchased their own IP for modem technology. All of the modem technology companies that once sold just modems (without an SoC) have all been acquired in the past few years and there simply isn’t a company that AMD can acquire to gain this functionality. Lately, Qualcomm’s messaging has been all about their wireless technology and how it sets them apart from the competition, so I feel that they are trying to drive the entire market discussion towards their strength as SoC performance reaches a level of parity.

Another concern is the fact that Microsoft’s Windows 8 RT is not getting very much pull and if they are targeting an ARM SoC for mobile/tablets, they are going to have a very hard time selling very many chips. Nvidia learned this lesson the hard way with the first batch of Windows 8 RT tablets, as did Qualcomm, although they were more cautious than Nvidia was. Hopefully by the time AMD’s consumer ARM SoCs reach the market, Microsoft will have a better operating system with a much more unified kernel across all CPU architectures.

Looking at the way things are right now, the x86 field is far less competitive for AMD when compared to the ARM field of competitors in mobile. AMD has struggled against Intel in terms of x86 absolute performance and has failed to compete with Intel and ARM companies in the tablet space. While I don?t entirely see AMD?s x86 mobile SoCs disappearing it certainly seems possible to consider that they will become a more ARM-heavy company in mobile and possibly even servers. Also keep in mind that I wouldn’t expect AMD to come to market with any consumer products utilizing an ARM chip until at least late 2014 or early 2015. This would theoretically also put them in a good place for a new Windows operating system launch as well, if Microsoft doesn’t botch it like they did Windows 8.