Ever since Intel moved to acquire Altera in 2015, it was only a matter of time before field programmable gate arrays (FPGAs) would show up in data centers in large numbers. Intel this week made good on that promise by unveiling Intel Programmable Acceleration Cards and associated software that enable FPGAs to be employed to offload the processing of some classes of workloads from Intel Xeon processors.
Dell EMC also announced it would become one of the first systems vendors to make use of the Intel FPGA cards in a system.https://o1.qnsr.com/log/p.gif?;n=203;c=204663295;s=11915;x=7936;f=201904081034270;u=j;z=TIMESTAMP;a=20410779;e=i
FPGAs are made up of logic blocks that can be programmatically reconfigured to process data at high speeds across bidirectional data busses and high-speed I/O interfaces. Intel estimates this first generation of FPGA cards could accelerate the performance of real-time analytics by a factor of 10 or more.
Bernhard Friebe, senior director of FPGA software solutions marketing at Intel, says Intel and its partners have developed a suite of software libraries that make it possible for FPGA-based cards to be transparently invoked by an Intel Xeon-based system to process analytics, video transcoding, machine and deep learning algorithms, certain classes of transaction, genomics and even I/O requests made over an NMVe backplane.
“You will be able to dynamically change the type of workloads running on the FPGA cards,” says Friebe.
In many instances, Friebe says Intel is betting that FPGAs will prove to be a more flexible approach to running these types of workloads than existing graphical processing units (GPUs) that have to be optimized to run a specific type of workload.
The result should be much higher utilization rates for systems based on Intel processors thanks to the software libraries developed by Intel, says Friebe.
The Intel Programmable Acceleration Cards are expected to be generally available in the first half of 2018. Intel will follow up with the launch of an Intel Xeon Scalable Platform with Integrated FPGA in the second half of 2018, which will enable FPGAs and Intel Xeon processors to share the same motherboard.
The impact of FPGAs is likely to be profound. Not only will the amount of IT infrastructure required to run complex existing applications be substantially reduced, new classes of applications that were not practical to develop should emerge in time. The time for developers to start imagining what those applications might be, however, is now.