UPDATE The official Radeon Instinct announcement from AMD is available over here. Here's a copy listing the three models:
The Radeon Instinct MI6 accelerator based on the acclaimed Polaris GPU architecture will be a passively cooled inference accelerator optimized for jobs/second/Joule with 5.7 TFLOPS of peak FP16 performance at 150W board power and 16GB of GPU memory
The Radeon Instinct MI8 accelerator, harnessing the high-performance, energy-efficient "Fiji" Nano GPU, will be a small form factor HPC and inference accelerator with 8.2 TFLOPS of peak FP16 performance at less than 175W board power and 4GB of High-Bandwidth Memory (HBM)
The Radeon Instinct MI25 accelerator will use AMD's next-generation high-performance Vega GPU architecture and is designed for deep learning training, optimized for time-to-solution
AMD says all three products will launch in the first half of 2017.
AMD is trying to gain a foothold in the deep-learning market with the upcoming Radeon Instinct MI25. This is a passively-cooled card based on the upcoming Vega architecture, it offers 25 teraflops of FP16 computing power and 12.5 teraflops of FP16 computing power.
The MI25 has a typical board power of "under 300W" and according to AMD's leaked slides it's quite a bit faster than NVIDIA's Pascal-based Titan X in DeepBench GEMM. The specifications slide mentions the MI25 has "2x packed math", not sure what this means but it could be that this is a dual-GPU card, which would also explain the long board size.
Besides the Vega-based MI25, the lineup also includes the Radeon Instinct MI6 and MI8. The latter seems based on the Fiji-powered Radeon R9 Nano, while the MI6 uses a Polaris 10 GPU. The MI6 offers 5.7 teraflops of FP32 computing power while the MI8 pack 8.2 teraflops.
Some more slides and estimations can be found at VideoCardz.