Intel starts baking speedy FPGAs into chips

Intel is packing Altera Arria 10 FPGAs with Xeon E5-2600 v4 processors in a multichip module

With rivals Nvidia and AMD both offering graphics processors, Intel is now deploying screaming co-processors of its own in the form of FPGAs.

FPGAs (field programmable gate arrays) are extremely fast chips that can be reprogrammed to do specific tasks. Intel last year acquired Altera for $16.7 billion as it started thinking beyond CPUs and stressing co-processors for demanding computing tasks.

Intel recently started shipping server chips paired with FPGAs as part of a pilot program. The company is packing Altera Arria 10 FPGAs along with its Xeon E5-2600 v4 processors, code-named Broadwell-EP, in a multichip module. The Xeon E5 chips were introduced last month.

Over time the FPGA technology will be integrated in the "same piece of silicon die as the CPU," an Intel spokesman said.

The shipment announcement was made at the ongoing Intel Developer Forum in Shenzhen, China.

FPGAs are being used by Microsoft to deliver faster Bing results and by Baidu for image search. FPGAs are less flexible than GPUs and execute tasks based on functionality programmed into a chip. FPGAs can be faster than GPUs on specific tasks, but are also very power hungry.

Intel plans to put FPGAs in cars, robots, servers, supercomputers and IoT devices. The chip maker has provided examples of how FPGAs could be programmed for genomics, or how the chips could tackle specific functions in a car. Integration of the FPGA into a chip will bring down power consumption and provide a direct path of communication with the CPU.

FPGAs are also being used in communications equipment, a market that Intel is chasing as 5G deployments are expected to grow exponentially. Intel's components and equipment could also be the backbone of many IoT installations, which already use FPGAs to connect devices with cloud services.

Join the TechWorld newsletter!

Error: Please check your email address.

More about AlteraAMDIntelMicrosoftNvidia

Show Comments
[]