When Intel acquired FPGA (Field Programmable Gate Array) chip designer Altera last year, the company set in motion a plan to further leverage the configurable logic technology in its primary cash cow data center server business. Not only were FPGAs making significant headway already in the data center, but the move established a beachhead for Intel that will likely help stave off future threats from competitors like ARM, in their core enterprise businesses.
Field Programmable Gate Arrays are a programmable chip technology that has been around for decades. Historically, FPGAs were used mainly as pre-production vehicles, allowing system designers to get to market quickly with new design ideas needing custom solutions. Companies would eventually spin that FPGA to a full custom ASIC (Application Specific Integrated Circuit) when the time came to ramp to full volume production and drive cost out. I actually helped build some of the very first cable modems with Motorola, back in the day, with FGPAs. Mot was stuffing these $200 chips into tens of thousands of pre-production units, just to get to market first with the technology. We knew full well that a box, that would be leased by the customer for a few dollars a month, eventually would need a custom ASIC at a fraction of that cost. So yes, compared to ASICs, FPGAs are costly; but the modern FPGA has grown-up and literally evolved in a big way.
These days, FGPA technologies have advanced to the point that engineers can build much larger, more complex chips with pre-built modules for anything from processor cores to custom algorithm accelerators. Couple that with the on-the-fly reconfigurable nature of modern FPGAs and you can imagine how powerful the technology can be in the age of machine learning. With FPGA technology support, not only can the processor reconfigure the software algorithm, tuning it for what it has learned and its changing workload, but it can also reconfigure its own hardware – and quite literally adapt its brain, if you will, over time.
FPGA technology is so powerful that Microsoft has deployed it in a program called Project Catapult, and Catapult FPGA boards are in nearly every server in the company's data centers currently. Microsoft Project Catapult FPGA accelerator boards are configured in a way that they can communicate between both blades and racks, such that they create a configurable “acceleration fabric.” The company notes this mesh of boards and servers can become pieces of a larger “brain of interconnected machine learning network of servers.” Microsoft is claims to be building the world’s first “hyperscale AI supercomputer” with Project Catapult. Fixed-function compute engines like CPUs and even GPU accelerators only get you so far in machine learning applications like speech recognition and translation and real-time network threat analysis. The applications are myriad and throwing more standard compute cycles at the problem eventually reaches a point of diminishing returns. Both Microsoft and Intel, as well as other players in this space, are faced with addressing this new processing paradigm and FGPAs are a very well-suited solution.
Intel’s FPGA technology also sits along side NVIDIA’s GPU technology in the Audi ADAS (Advanced Driver Assistance System) for piloted driving technology in the A7 series (with integrated ARM cores no less), and the application reach for the technology goes far beyond the data center as you might imagine. The company has also announced even tighter integration of the technology with its core processor offering, debuting Xeon processors with on board integrated FPGAs in the same socket.
In the high performance AI processing world, it’s quite literally evolve or die. Intel appears to be geared-up rather well with evolvable, reconfigurable technology that will allow the company to continue to innovate in the core markets it serves, as well as address new markets moving forward.
