By Darren Allan
AMD has unveiled a trio of new GPUs aimed at the machine learning and AI arena to rival Nvidia’s Tesla range, backed up with a suite of open source software offerings for an all-round solution.
The Radeon Instinct range are high-performance GPU solutions for machine intelligence – deep learning and training – with three models to address different levels of performance.
The base card is the Radeon Instinct MI6 which uses AMD’s Polaris graphics and is passively cooled, offering 5.7 Tflops of peak FP16 performance alongside 16GB of video memory (drawing 150W of power).
Moving on up the scale, we have the Instinct MI8 which is driven by a Fiji Nano GPU (older tech) and boasts 8.19 Tflops in terms of performance, with 4GB of HBM video RAM on board (drawing 175W).
And finally, the Radeon Instinct MI25 uses AMD’s next-gen Vega technology and is billed as a ‘passively cooled training accelerator’, although the company didn’t spill any early spec details on this board save for the fact that it will utilise 300W of power.
Killer Instinct
As mentioned, along with the Instinct hardware, AMD also has offerings on the software front, and it revealed MIOpen, a new open source library for GPU accelerators …read more
Source:: techradar.com – Computing Components