AMD unveils the Radeon Instinct brand with the Vega architecture for machine learning

Machine Learning or Deep Learning is something Nvidia has been pushing hard for with both its graphics cards and the Tegra system circuits, which are used to accelerate this type of business. Now AMD chooses to jump on the bandwagon and do so with not only a new brand, but also the architecture Vega.

Radeon Instinct is set to dramatically advance the pace of machine intelligence through an approach built on high-performance GPU accelerators, and free, open-source software in MIOpen and ROCm. With the combination of our high-performance compute and graphics capabilities and the strength of our multi-generational roadmap, we are the only company with the GPU and x86 silicon expertise to address the broad needs of the datacenter and help advance the proliferation of machine intelligence, berättar AMD:s VD, Dr. Lisa Su.

The brand is Radeon Instinct and the name of each plug-in card in the series begins with the letters MI (Machine Intelligence) followed by a number, which roughly gives an indication of performance (TFLOPS) in the type of applications Radeon Instinct is intended for. In terms of appearance, it is a yellow-black color theme and all models have passive cooling, where the idea is instead that fans in a server rack should be responsible for heat dissipation.

The first two models in the series will be the Radeon Instinct MI6 and MI8, where the first is based on the Polaris 10 graphics circuit with computing power of up to 5.7 TFLOPS and an unknown amount of GDDR5 memory with a memory bandwidth of 224 GB / s. The power consumption is stated to be below 150 W.

The Radeon Instinct MI8 is very similar to at least the Radeon Nano. The graphics circuit is Fiji XT at up to 1,000 MHz for 8.2 TFLOPS in computing power. This is accompanied by 4 GB HBM for a bandwidth of 512 GB / s, a shrinking form factor and power consumption of less than 175 W.

The flagship of the series is the Radeon Instinct MI25, which means that it is capable of delivering up to 25 TFLOPS in performance. This is made possible by the upcoming Vega architecture with the sequel (NCU) to Graphics Core Next (GCN) and a new type of cache and graphics memory controller. Despite a real jump in performance, the power consumption is stated to be below 300 W.

In addition to the Radeon Instinct, Radeon CEO Raja Koduri is taking the opportunity to show an early prototype of a “cube” that will be able to deliver 100 Teraops or a trillion calculations per second. This is possible with no less than four graphics circuits from the Vega architecture.

The launch of the Radeon Instinct will take place during the first half of 2017. First out are MI6 and MI8, which are based on existing graphics circuits, while MI25, which is based on Vega, will arrive later in the specified time frame.