This is exactly what AMD has announced yesterday. They will begin designing x64 ARM based SoCs for the datacenter. This is an important step for AMD, but might not yield the results they are looking for. Undoubtedly AMD has a large amount of expertise in designing x64 hardware (if any of those engineers survived the recent purges). It was AMD that launched the x86-64 architecture that we still use today in some form or another. They saw that the market was not ready to make the leap to true 64-bit computing and working with Microsoft they came up with an instruction set that would give us the best of both worlds (Intel later followed with EM64T for their CPUs). Unfortunately a lot has changed since those days and AMD seems to have stumbled after making that great leap.
Since that time AMD has made a few moves to bring them back to attention in the market. Their APUs have been a hit with the cost conscious crowd. This group is looking for solid performance at the best price and although they are aware that their new APU is not as “powerful” as what Intel has to offer they are not willing to pay the extra to get Intel inside their new device. They are also often very aware that AMD’s APUs have significantly better graphical performance at the same resolutions. It is an interesting balance of performance that allows AMD to sell them for far less than what Intel can at the moment.
In fact AMD’s APU foray has been so successful that they are moving out of the mobile and All-in-One space and into the server and workstation space with their FirePro CPUs. Now they hope to combine what they have learned from building their APUs with what they know about x64 and bundle it into an ARM based offering for the market. As mentioned they will kick things off in the data center where they can find the best balance of performance and energy efficiency before pushing these out to the consumer market where they will compete head to head with the likes of nVidia, Samsung, Qualcomm and Texas Instrument. Some of these are also part of AMD’s HAS (Heterogeneous System Architecture) Foundation which means that AMD will have a lot of help in getting the ARM side of things right.
Now the question is; can AMD find the right balance between high-density x64 ARM CPUs and x86 processors in their Seamicro systems to be attractive to future customers or will Intel and Cray still hold sway in the datacenter especially as they now have Phi (their answer to nVidia’s Tesla) to offer? AMD and their partners certainly think they can. Already there are quote from Dell, HP, and Redhat about the move that show interest in AMD’s new direction. For what it is worth we see this move in the same way that we saw the SSD. The SSD make sense in the enterprise when you use them correctly. Just dropping in a SAN (Storage Attached Network) full of SSDs is not the answer to all of your problems even though they can be faster and more energy efficient, they still have their own problems. You typically end up having a "tray" of SSDs for cahcing and HDDs for actual storage for the best balance of performance and efficiency. The same can be said of ARM in the datacenter. You will always need the computational power that an x86-64 (or in some cases true 64) CPU brings. You just cannot get around that. Even the new Super Computers running on nVidia’s Tesla need CPU power to run at their best. The same will be the case with ARM, they will be for specific tasks, but for the near term they will not be able to match the power and performance that an x86 CPU can, they will be there for less demanding tasks so you can balance the overall energy efficiency of your data center. We are guessing that AMD knows this and will be working with the Seamicro team and their other partners to find that magic balance and offer a product that has the “best of both worlds” at a good price.