American multinational semiconductor company Advanced Micro Devices, Inc. has just revealed its first step into the field of artificial intelligence. The California-based company officially announced Radeon Instinct and the release of three new accelerators at the Radeon Summit held this week. These three new GPUs are indicators of AMDs latest agenda: machine learning applications and deeper exploration of artificial intelligence.

The three AMD Radeon Instinct GPUs are the MI6, the MI8, and the MI25 Vega with NCU. While all are vastly different accelerators with unique specs, all three have been optimized to accelerate popular deep learning frameworks --- at least, according to a report by iTechPost. There is no exact ship date for these products yet, but most speculations can agree on it being released some time in 2017.

The first accelerator, the MI6, is supposedly derived from AMD's Polaris 10 chip. It has 16GB of total onboard RAM and a memory bandwidth of 224 GB/s. It features 5.7 TFLOPs performance, runs on a somewhat lower clock "than the boost frequencies on consumer parts", and draws less than 150W.

The second accelerator is a smaller GPU with a cap of 4GB RAM and 512 GB/s of memory bandwidth. The MI8 was based on AMD's Fiji architecture and built around AMD's R9 Nano. It features 8.2 TFLOPs and draws less than 175W.

Being an AMD Vega-derived chip, people understandably have higher expectations for the MI25. With a 16GB HBM2 and a 512 GB/s memory bandwidth, plus its name, assumptions land MI25 with a 25 TFLOPs capability. But according to The Country Caller, that level simply isn't plausible. Half of 25 TFLOPs is, apparently, a safer assumption.

For its first foray into artificial intelligence, there's a lot banking on Radeon Instinct's three GPUs. AMD has also stated that Radeon Instinct has been designed in such a way that allows programmers to give more time and attention to training neural networks.