News Categories

The Stratix 10 chip is the result of Intel's acquisition of Altera

By Bryan Chan & Koh Wanzi - on 11 Oct 2016, 6:20pm

The Stratix 10 chip is the result of Intel's acquisition of Altera

Image Source: Intel

Something is starting to come out of Intel’s recent Altera acquisition. The Stratix 10 (first seen at IDF 2016) is a new FPGA product that features an ARM design for the microprocessor. That said, it is actually based on Intel’s 14nm Tri-Gate process, and combines a quad-core Cortex-A53 ARM processor, four stacks of HBM2 memory, and a 5.5 million logic element Altera-designed FPGA in a single package.

This is no consumer product, and the Stratix 10 is targeted at networking, data center, and radar or imaging customers. Intel says it has been fundamentally reworked to offer performance that is significantly better than competing solutions on the market.

Compared to the Stratix V, the Stratix 10 offers double the core performance, up to 70 percent lower power, and up to 1TBps of memory bandwidth thanks to the use of HBM2 memory. Single-precision floating point performance is pegged at up to 10 TFLOPS.

Image Source: Intel

In a nutshell, an FPGA is an integrated circuit that can be rewired in the field. They are good with high parallel workloads and can efficiently handle huge volumes of inputs, processing the data in parallel through custom-programmed logic gates. Ultimately, their biggest advantage is their flexibility as they can be reconfigured in milliseconds to accelerate specific tasks, although the complexity of programming them adds to their cost.

The Stratix 10’s claimed 10 TFLOPS of single-precision floating point performance is significant because this is reportedly the most important factor in training neural networks.

As it turns out, Microsoft recently deployed FPGAs on its Azure cloud computing platform, and had previously deployed Stratix V FPGAs to most of its Azure data centers. Microsoft intends to build the world’s fastest AI supercomputer, and this programmable silicon will initially be used to accelerate its AI research efforts, among other things.

Intel spent US$16.7 billion to buy Altera, and its new FPGA efforts will be followed closely to see if it can make farther inroads into the data center and Internet of Things market segments. Today's networks require ever higher bandwidth and lower latencies, data centers have to deal with frequently changing workloads, and power efficiency is more important than ever. It is these three factors that will drive FPGA adoption. 

Source: Intel via ExtremeTech