21 June 2014

Intel unveils new Xeon chip with integrated FPGA, touts 20x performance boost

22nm silicon die and wafer (Intel, Knights Ferry)
Late yesterday, Intel quietly announced one of the biggest ever changes to its chip lineup: It will soon offer a new type of Xeon CPU with an integrated FPGA. This new Xeon+FPGA chip will fit in the standard E5 LGA2011 socket, but the integrated FPGA will allow each chip to be customized to specific workloads. This move is almost certainly intended to make Intel-x86 a better all-round platform for a wider variety of workloads in enterprise and data center settings, and to dissuade customers from switching to GPGPU accelerators from the likes of Nvidia.

The Xeon+FPGA also raises the question of whether Intel would ever consider integrating an FPGA into its consumer Core line of chips it’s exceedingly unlikely, but it it’s hard to deny how awesome it would be if next-gen games and apps had access to an FPGA to speed up core processes. But more on that at the end of the story.

What is an FPGA?

First things first, let’s talk field-programmable gate arrays (FPGA). As the name implies, an FPGA is essentially a blank chip that can be repeatedly reprogrammed after manufacturing. With very few exceptions, every chip inside your computer is hard-coded (at the time of manufacturing) to perform just one set of functions. Your CPU can only do exactly what Intel or AMD designed it to do. You can’t take your CPU and turn it into a GPU. But you can take an FPGA, program it to perform one set of functions (say, graphics), and then reprogram it to handle another type of workload (say, sorting through databases).

Altera StratixIVGX FPGA
An Altera FPGA

The main advantage of an FPGA, other than its customizability, is that it has monstrously high performance. In much the same way that an ASIC is by far the fastest and most efficient way of processing a specific workload (and thus why they’re used for Bitcoin farming), an FPGA is also very fast and efficient. They’re not quite as fast or efficient as ASICs, but what you lose in speed you gain in reprogrammability (again, ASICs are set in stone at manufacturing time).
Why is Intel pairing its Xeon CPU with an FPGA?

Over the last few years, as more and more processing has moved to the cloud, Intel’s dominance of the server market has been attacked on a few fronts. Xeon chips are good all-round chips but there are plenty of cases where another, more-workload-specific chip might make sense. This is why people are starting to look at cheaper, low-power chips for web servers (which don’t need expensive, hot Xeon chips), and GPGPU accelerators for highly parallel processing (as in supercomputers). For these more specific workloads, Intel is pitching Atom for lightweight stuff, the 50-core Xeon Phi for supercomputers, and now the Xeon+FPGA (it doesn’t seem to have an official name yet).

Xeon Phi Coprocessor
Intel has tried to fight back against GPGPU accelerators with its own massively parallel chips, such as the Xeon Phi

Intel also notes that, in the last year alone, it has delivered “15 custom products” for big customers like Facebook and Ebay (probably Xeons with more/less cache, more/less cores, etc.) and it will do over 30 custom designs this year.

What’s the purpose of this new Xeon+FPGA product? In the words of Intel: “The FPGA provides our customers a programmable, high performance coherent acceleration capability to turbo-charge their critical algorithms.” Intel estimates that the Xeon+FPGA will see massive performance boosts in the 20x range (for code executed on the FPGA instead of a conventional x86 CPU but obviously there will be big overall speedups as bottlenecks are removed. The other advantage is that workloads change so if your critical algorithms change, or your whole company pivots, the FPGA can be repurposed without having to buy lots of new hardware.

Xeon E5
An eight-core Xeon CPU, with cache down the middle and various controllers/IO along the left and right edges.

As far as tech specs go, we can sadly only guess at this point. The FPGA is probably manufactured by Altera, which Intel has a close working relationship (it was one of the first companies that was allowed to use Intel’s closely guarded chip fabs). FPGAs in general are very bulky (that’s the trade-off of using programmable gates), and there isn’t a whole lot of spare space on an LGA2011 Xeon E5 chip package, so we’re probably not looking at a very large FPGA. Though if the FPGA can leverage the Xeon’s cache, and perhaps some other low-level resources, some space-saving efficiencies might be realized. (Intel says there’s a “low-latency coherent interface” between the Xeon and the FPGA, but that’s about it.)

There’s no word on pricing or availability but obviously it won’t be cheap (the top Xeons already cost thousands of dollars). Intel is hoping that the Xeon+FPGA package is enticing enough to convince enterprises to stick with x86, rather than moving over to a competing architecture. It is probably a lot easier to rewrite a small segment of critical code to run on an FPGA than to rewrite the entire application in OpenCL.

Finally, a fun thought experiment: If this little Xeon+FPGA experiment is a success, might we one day see a Core+FPGA chip? Or a GPU+FPGA setup? Imagine if you could load a core part of a game engine into the FPGA; the physics engine perhaps, or the enemy AI. Or a core part of your PC’s operating system. That could be very cool indeed.

Courtesy Extremetech


Post a Comment

Get every new post delivered to your Inbox.


Copyright © 2015 Tracktec. All rights reserved.

Back to Top