For years, AMD has been struggling in the HPC GPU market. They had recently improved when the Radeon Open Compute Platform (ROCm) started paying dividends. Since 2015, ADM has already laid a good part of the ground work that is needed when tackling the deep learning market.

Now, Advanced Micro Devices (AMD) is the process of computing its prowess to release the next big thing in the server market - the machine learning and artificial intelligence. The new machine learning is called "Radeon Instinct." The upcoming device is said to provide server designers and developers a compelling set of infrastructure to track the machine learning.

The new Radeon Instinct is composed of hardware and software aspects that have the ability to deliver a full machine intelligence platform. AMD wants that all industries like financial services, life sciences, and the cloud that are now in the machine learning solutions and infrastructure will use Radeon Instinct for their computing requirements.

According to ExtremeTech, AMD has developed a set of GPU accelerators for server farms which are mainly geared for machine intelligence workloads. There are three Radeon Instinct accelerators which are configured to be release in the first half of 2017.

The said accelerators will use three generations of AMD GPU architectures which include Vega that will be available in the consumer-oriented cards with a similar time frame.

MI6 is considered as the most basic accelerator that uses Polaris GPU with 5.7 TFLOPs of computing performance with 16GB memory. The said accelerator can step up to a Fiji GPU with 8.2 TFLOPs of performance that includes 4GB of HBM in the small form factor card which is similar to the Radeon R9 Nano.

MI25, on the other hand, uses Vega with AMD's next-generation compute units which are packed with a huge 25 TFLOs of FP16 compute in a sub-300W power envelope.

Lastly, the most interesting things which were recently announced by AMD is the combination of Radeon Instinct Accelerators with the upcoming Zen 'Naples' platform. According to Anandtech, server designers will be able to pair with many accelerators with a single CPU with Naples platform. Using this platform, GPU heavy servers will be possible at a lower cost and footprint compared before.