AMD follows the AI trend with its upcoming Radeon Instinct GPU for machine learning.
Deep Learning & AI Market
According to Extreme Tech, in the year 2016 artificial intelligence (AI) and deep learning have been trendy topics. So far, most of the interest has focused on either custom hardware such as Google's TensorFlow or Nvidia GPUs.
At its turn, Intel is also developing its upcoming Xeon Phi-class silicon. AMD plans to enter the market as well, working on a new lineup of graphics cards. AMD upcoming GPUs are based on three different product families.
In order to accelerate these workloads, AMD will also offer its customized libraries and software tools. The deep learning and AI markets are still in their fairly early days and AMD aims to catch a good share in the industry by bringing some innovative tech to the table.
Three AMD Radeon Instinct GPUs
AMD's new Radeon Instinct brand will consist of three new cards deployed under three different product families. The MI6 is derived from Polaris with 16 GB total onboard RAM, but running at a slightly lower clock than the boost frequencies on consumer parts.
The MI8 is a smaller GPU with 4 GB RAM limitation, built around R9 Nano and working at the same frequencies. The MI25 is expected to be significantly faster than the other two cards, being a Vega-derived chip. No details on its core are provided yet by AMD.
At the moment, we do not know any ship date for any of these products, but they are expected sometimes H1 2017. Tech experts believe that MI6 and MI8 cards will be launched first, in order to test the market.
History Of AMD Vs. Nvidia Competition
It might seem surprising that AMD aims to compete against Nvidia with midrange and older consumer hardware. However, at a closer analysis, this might be a smart move.
At it turn, Nvidia still sells a range of HPC products based on older Kepler and Maxwell hardware. In a number of compute workloads, AMD's GCN was actually a very strong competitor against Nvidia.
AMD continues to offer a CUDA compatibility layer and, if it brings pricing in appropriately, the company could make a compelling argument for its upcoming Radeon Instinct GPUs. This depends, however, on whether AMD can overcome Nvidia's lead in GPGPU computing and how many resources AMD will be able to dedicate to the software side.
During the Maxwell and Kepler eras, AMD held a leadership position in multiple compute benchmarks and workloads, often by enormous margins. But, unfortunately, OpenCL could not capitalize on the strength of AMD's underlying hardware, as the company did not have the resources to spend on a major enterprise computing push.
Advantages Of AMD Radeon Instinct GPUs Lineup
Since those times, AMD has made some incremental progress on this field. Among them are included various server and virtualization product launches, last years' Boltzmann initiative and now the Radeon Instinct brand. According to PC World, AMD claims that the fastest Radeon Instinct GPU, the MI25, will outpace Nvidia's current Titan X Pascal card.
AMD's MIOpen GPU-accelerated library will be used in Radeon Instinct products to provide standard routines GPU-tuned implementations such as activation functions, pooling, convolution, normalization and tensor format. According to AMD, its ROCm deep learning network has also been optimized for acceleration of popular deep learning frameworks, Tensorflow, Torch 7 and Caffe. This will allow programmers to focus more on training neural networks. ROCm aims to become the foundation of the next evolution of machine intelligence problem sets.