Updated Intel has expanded its chip customization business to help it take on the hazy threat posed by some of the world's biggest clouds adopting low-power ARM processors.
The chip company announced on Wednesday at GigaOm Structure in San Francisco that it is preparing to sell a Xeon E5-FPGA hybrid chip to some of its largest customers.
By doing this, Intel hopes to persuade some of the world's largest data center operators that they can source all their compute needs from Intel, rather than from rivals like AMD or ARM-based chip vendors, or build their own.
"That allows end users that have applications that can benefit from acceleration to load their IP and accelerate that algorithm on that FPGA as an offload," explained the general manager of Intel's data center group, Diane Bryant, in a speech at GigaOm Structure. "The FPGA has direct access to the Xeon cache hierachy and system memory," she explained.
This gives large customers – such as Facebook and eBay, – the ability to pair a reconfigurable chip with a typical x86. This could be useful for speeding up known tasks such as compression or decompresssion, certain search or machine learning algorithms, encryption and decryption, and more.
"It's not the prettiest," admitted Intel's director of cloud computing Raejeanne Skillern in a chat with El Reg, "it may not be the most aesthetic looking thing but it's functional and it fits."
By sticking an FPGA on top of a Xeon and linking it via Quick Path Interconnect tech, Intel reckons it has a compelling product for large customers.
There's evidence the company may be right. Earlier this week Microsoft announced a scheme named "Catapult". With this system, the company added FPGAs to over 1,600 servers used by its Bing search engine and, in doing so, had almost doubled throughput while only increasing power consumption by ten per cent.
Microsoft slotted these FPGAs in via the PCIe slot because it didn't want to sacrifice one of its sockets for an FPGA instead of a Xeon. With Intel's new product, maybe Microsoft won't have to make such a sacrifice.
Intel announced last year that it was getting into the chip customization business to help it take on the lurking threat of low-power ARM chips. At the time the main question was on how far Intel would go in customizing chips for its customers. Later in the year the company explained that it was mostly focusing on power management tweaks for customers, but had also etched some new instructions in.
"We have done discrete FPGA solutions. We've done software development kits," she said. "This is the first time we're talking about coherently packaging [FPGAs and Xeons]."
Using FPGAs in a data center "does require some level of sophistication," Skillern said, but noted that some of these large providers have been busily hiring their own chip engineers. Google, for instance, recently hired one of the big brains behind failed ARM startup Calxeda, and Amazon has also hired a bunch of semiconductor engineers.
However, if someone chooses to buy a bunch of FPGA-Xeon frankenchips, they will be wedded to that hardware for anywhere from two to five years, depending on refresh rates. This means that if the tasks being offloaded to the FPGA change dramatically enough, the FPGA's logic may no longer be quite right even after in-the-field reconfiguration. It's a big bet for customers to have to make.
That said, Intel says FPGAs can demonstrate a speedup of 10X over traditional chips for certain tasks, and that adding an FPGA to a chip via QPI give its a 2X speedup again.
Intel is not disclosing the manufacturer of this particular FPGA, though we would point out the company has an agreement with chipmaker Altera that sees the FPGA-specialist build its chips within Intel's own data centers. Make of that what you will.
The original version of this story said that Intel had begun selling the chips. An Intel spokesperson got in touch to tell us that the company is in early discussions with some of its mega-customers, and hasn't started selling the frankenchips en masse – yet. ®