Advanced Micro Devices officials reportedly are looking into using FPGA accelerators with their Opteron server processors, a move that would advance the company’s efforts to gain traction in a data center market dominated by larger rival Intel.
Citing unnamed sources, Bits and Chips, an Italian technology Website, is reporting that AMD is studying integrating field-programmable gate arrays (FPGAs) with its x86 Opteron CPU to create a customized chip for a particular but unidentified customer. The custom chip also would leverage the capabilities in AMD’s upcoming second-generation High-Bandwidth Memory (HBM2) technology, which would help create high-speed communications between the CPU, FGPA and memory.
Such a move would help AMD in its competition with Intel, which is in the process of buying FPGA-maker Altera for $16.7 billion. Intel has used Altera FPGAs in its Xeon server chips in the past, but buying the company will enable Intel to more tightly integrate the technology into its roadmap. Intel is looking to a number of technologies to help accelerate the performance of its Xeon chips, from its own x86 Xeon Phi co-processors to the Altera FPGAs to its partnership with eASIC to bring application-specific integrated circuits (ASICs) to custom Xeons that can be used in data centers and cloud environments.
It also would help AMD address the rapidly changing workload demands found in such areas as the Internet of things (IoT) and cloud environments. Chip makers increasingly are looking for accelerators to boost the performance of their processors and the systems that run on them while holding down power consumption, important concerns for cloud providers and scale-out environments. AMD already offers GPU accelerators.
AMD officials declined to comment on the speculation regarding the FPGAs.
While AMD has struggled in recent quarters as the global PC market continues to stagnate, its semi-custom chip business has been a high point. The company was able to place its chips in the latest gaming consoles from Microsoft, Sony and Ninetendo, and sales of the systems have been strong. In a conference call with analysts and journalists to talk about second-quarter financial numbers, President and CEO Lisa Su again noted the strength of the semi-custom business, and said the company was working on a new chip for a new unnamed customer.
“We also began development of a new semi-custom design” during the quarter, Su said. “Like our other semi-custom designs, the details are customer confidential, but we’re pleased with our progress continuing to expand our customer base in this important part of our business.”
An FPGA would give AMD another capability to bring to its semi-custom chip business. Unlike traditional CPUs, FPGAs can be programmed through software, which make them increasingly important accelerators for cloud and Web-scale environments because they can be reprogrammed to server particular workloads. Microsoft last year announced Project Catapult, an effort to use FPGAs in servers running Intel Xeon chips to speed up Bing search results.
At a meeting in May with financial analysts, Su and other AMD executives said that the company—which has seen its share of the x86 server market dwindle in recent years—is aiming to reinvigorate its data center business as one of several avenues for returning to sustainable profitability. The chip maker is banking on such new technologies as its upcoming “Zen” core design (which will first appear in PC processors), accelerated processing units (APUs) for servers that will integrate the CPU and GPU on the same die, and HBM to help it gain traction in the data center. Both Su and Forrest Norrod, senior vice president and general manager of the company’s Enterprise, Embedded and Semi-Custom Unit, said they want to start seeing gains in the data center space next year and ramping into 2017.
