For years, we've heard rumors that Intel was building custom chips for Google
, but these deals have always been assumed to work with standard hardware. Intel might offer a different product SKU with non-standard core counts, or a specific TDP target, or a particular amount of cache -- but at the end of the day, these were standard Xeon processors. Today, it looks like that's changing for the first time -- Intel is going to start embedding custom FPGAs into its own CPU
The new FPGA-equipped Xeons will occupy precisely the same socket and platform as the standard, non-FPGA Xeons. Nothing will change on the customer front (BIOS updates may be required), but the chips should be drop-in compatible.
What's An FPGA?
To understand the significance of this arrangement, we need to explain what an FPGA is. Computing in the modern world generally happens on one of four devices: A CPU, a GPU, an FPGA, or an ASIC. A GPU is a graphics card with significant compute capability added on (sometimes called GPGPU). A CPU is a general purpose compute engine that can handle a vast ecosystem of tasks but has a great deal of hardware designed to translate those tasks into its own particular instruction set and execute them accordingly.
An FPGA (Field Programmable Gate Array) is an array of logic gates that can be programmed to do a great many tasks by manipulating those gates and what tasks they perform. It differs from an ASIC (Application Specific Integrated Circuit) in that an ASIC chip is a hardware processor that's designed for one specific task. A suitably complex FPGA can theoretically perform any task an ASIC
can perform, though there may be marked differences in power consumption and performance.
The easiest way to think of an FPGA is as a sort-of software-hardware hybrid solution. FPGAs can be reconfigured to run updated software packages more efficiently or when different tasks need to be performed. They can also be used to accelerate specific tasks; Intel alludes to this when it writes:
The FPGA provides our customers a programmable, high performance coherent acceleration capability to turbo-charge their critical algorithms. And with down-the-wire reprogramability, the algorithms can be changed as new workloads emerge and compute demands fluctuate. Based on industry benchmarks FPGA-based accelerators can deliver >10X performance gains. By integrating the FPGA with the Xeon processor, we estimate that customers will see an additional 2X in performance thanks to the low latency, coherent interface.
The company has not stated who provided its integrated FPGA design, but Altera
is a safe bet. The two companies have worked together on multiple designs and Altera (which builds FPGAs) is using Intel for its manufacturing.
This move should allow Intel to market highly specialized performance hardware to customers willing to pay for it. By using FPGAs to accelerate certain specific types of workloads, Intel Xeon customers can reap higher performance for critical functions without translating the majority of their code to OpenCL or bothering to update it for GPGPU