GPUs aren't the only kinds of compute accelerators finding their way into servers these days. Field-programmable gate arrays, or FPGAs, help bridge the gap between general-purpose processors and application-specific integrated circuits. Intel announced today that major server vendors will be offering its Programmable Acceleration Cards (PACs) as configuration options for their products. The blue team says Dell EMC is ready to deploy PACs "in volume" in its PowerEdge R640, R740 and R740XD servers and that Fujitsu will offer PACs as an option in its Primergy servers.
The PAC puts one of Intel's Arria 10 GX FPGAs on a half-height, half-length PCI Express Gen3 x8 adapter card. The Arria 10 GX chip gets 8 GB of DDR4 memory to play with and 128 MB of flash for non-volatile storage, and it operates in a 60 W thermal envelope. An FPGA is nothing if one can't program it, and Intel claims it's ready to help developers fire up their workloads on those servers with its Acceleration Stack for Xeon CPUs with FPGAs. Intel says the Open Programmable Acceleration Engine, or OPAE, available as part of this stack makes it faster and easier for developers to get their ideas onto programmable logic without the headaches of low-level FPGA wrangling.
For an idea of the performance benefits available to developers looking at FPGA acceleration, Intel notes that Levyx, one of its development partners, was able to achieve "an eight-fold improvement in algorithm execution and twice the speed in options calculation compared to traditional [Apache] Spark implementations" for financial modeling workloads by using servers with PACs. Folks with lots of data to analyze and workloads amenable to FPGA acceleration will likely want to call up their Dell or Fujitsu reps for more information about servers with Intel PACs inside.