Techno Musings

There seems to be a general move afoot for the entire IT sector to standardize on the x86-64 processor architecture.  Examples include Cray and SGI who use x86-64 from AMD and Intel respectively.  In essence these two companies that used to make their own processors have stopped.  (Further HP no longer makes the PA-RISC processor, but relies on Intel Itanium. While not an x86-64 based processor it does follow a parallel direction of outsourcing microprocessor design and fabrication.)  Largely this leaves IBM as the only remaining successful independent general purpose CPU manufacturer. (Well SOracle is there too, but even they are somewhat reliant on Intel, AMD and Fujitsu for their processors.  Further it is unclear if Oracle will continue investing in SPARC processor design or not.)  Well this got me to thinking on what the impact of this trend would be if  we projected that x86-64 consumed 100% of the microprocessor space.

In the embedded space Intel is facing stiff competition from the ARM consortium.  Some signs of this are in the smartphone space and the emerging tablet market, where both appear to be dominated by the ARM architecture due to the better mix of price/performance and more importantly power consumption.  For this I have no opinion one way or the other, but I do find it interesting that Apple has designed their own processor, the A4, bypassing Intel Atom for the iPad.  Instead I’m going to stick to some discussions about hybrid or many-core computing to make my case. (Note I’ve talked a lot about this in the past and for a good smattering you can issue the following search on my blog: hybrid computing.)   I predict that the dominance of x86-64 will spur on a renaissance in development of special purpose silicon either in ASIC or FPGA form coupled to x86-64 — meaning Hybrid Computing.  The simple question that VCs or other investors within companies will face is with x86-64 being good enough and even as the basis for specialized SoCs (think Jasper Forrest or AMD’s Fusion in the future) how is it possible to beat out either AMD or Intel?  I believe that the answer will be to not compete at all, but instead take advantage of AMD opening up Hyper-transport and their socket designs to 3rd parties or Intel opening up their Front Side Bus to 3rd parties.  This allows for specialized processing elements to be created and better interfaced with the x86-64 platform.  Already Xtremedata offers an FPGA package that either plugs into the FSB on Intel platform or the socket on the AMD platform.  Thus allowing companies to create their own “software” processors and interface to the operating system of their choice through some kind of software — most likely some kind of a device driver and user space program combination.  Today there are toolkits and projects that seek to ease the creation of customized processors, accelerators, etc.  One such project I’ve talked about in the past is the Liquid Metal project from IBM.  There is also a cool project over at the University of Michigan called CCCP — and when you go there take a look at their project logo it is a bit of a trip.  So another portion of the story is coming together namely creating the compliers, toolkits and IDEs needed to make building your own custom silicon “relatively easy.”  So perhaps your next job as a programmer will be to create some customized hardware.