Intel’s first processor with performance-boosting FPGA to ship early next year
The Xeon processor will be aimed initially at hyperscale cloud providers like Microsoft and Google
By James Niccolai
Intel will ship its first Xeon server chip with a programmable FPGA from Altera in the first quarter next year, some 18 months after announcing work on the product.
FPGAs, or field programmable gate arrays, are specialty chips used to accelerate specific workloads. With Moore’s Law starting to slow, some see the programmable chips as an important way to accelerate tasks like machine learning and data analytics in an energy-efficient way.
Intel announced in June last year that it was working on a Xeon part with an integrated FPGA, and on Wednesday the head of its data center group, Diane Bryant, said the chip will arrive in the first quarter next year.
“We’ll be shipping it to the largest cloud service providers in Q1 so they can begin tuning their algorithms,” she said during an on-stage interview at the Structure conference in San Francisco.
It’s taken a while to get the chip to market, but the pace of work will probably increase now that Intel is buying its FPGA partner, Altera, for almost $17 billion. It announced the deal in June and said it would close in six to nine months.
Bryant didn’t provide any details about the chip except to say the “footprint” will be compatible with current Xeon processors, meaning presumably that it can be a drop-in replacement for existing chips.
Programmed to run a particular algorithm, FPGAs can be more efficient than general-purpose CPUs on a performance-per-watt basis.
Microsoft has described successfully using FPGAs to speed performance of its Bing search engine, but they can be used for many types of workloads.
Using FPGAs to accelerate tasks isn’t new, but they’re usually discrete components on a motherboard, linked to the processor via PCIe. Integrating them in a dual-chip package using Intel’s QPI interconnect allows direct access to the Xeon’s on-chip cache and main memory, reducing latency.
Intel has said an integrated, dual-chip package can double the performance gain from an FPGA compared with using it as a discrete component.
It’s not the only chip maker eyeing FPGAs. Qualcomm, which is developing an ARM server processor, and IBM, which makes the Power processor, have both partnered with Altera’s main rival, Xilinx, to combine FPGAs with those chip architectures.
Note: When you purchase something after clicking links in our articles, we may earn a small commission. Read ouraffiliate link policyfor more details.