AMD demonstrates the industry's first 7nm GPU (Radeon Instinct Vega) with 32GB HBM for datacenter AI acceleration.

At Computex 2018, AMD's Lisa Su displayed the world's first 7nm GPU die, complemented by 32GB of HBM2 memory and designed for the company's Radeon Instinct Vega GPUs targeting the exploding data center AI and machine learning market. Su assured the crowd that the company plans to bring the new process to consumer GPUs in the future, though she did not specify whether the new gaming graphics card would feature the Vega architecture, with expectations pointing toward the next-gen Navi architecture instead. AMD is currently sampling the 7nm Vega GPU to partners and will launch it to the general market in the second half of 2018—a full quarter ahead of expectations—featuring the fifth-gen GCN microarchitecture with numerous optimizations specifically for AI workloads.

The 7nm process delivers substantial technical advantages: AMD claims it is twice as dense as its 14nm process, with the 7nm Vega die appearing roughly 40% smaller than its predecessor. The new process affords a 2x increase in power efficiency and provides a 1.35x increase in performance, metrics that suggest AMD is relatively far down the development path with its working silicon. David Wang announced that the company is striving to produce a new graphics product every year for the next three years, a commitment that bodes well for a consumer graphics market that has become somewhat stagnant in terms of recently released high-end graphics cards.

A fundamental change comes through the Infinity Fabric, a coherent interconnect AMD designed to facilitate communication between components inside discrete packages, as seen with Ryzen and Threadripper processors. AMD's new tactic extends the fabric outside the GPU to speed peer-to-peer communication between graphics cards, an approach similar to Nvidia's NVLink that purportedly reduces latency and boosts throughput. It is logical to expect the Infinity Fabric to eventually extend to communication between the CPU and GPU, which could provide AMD an advantage as the only producer of both x86 processors and GPUs. AMD originally planned to release 12nm GPUs announced the previous year but made the strategic decision to skip to the 7nm process instead, with 7nm Navi up next on its roadmap followed by new graphics cards with a 7nm+ process that should arrive before the end of 2020.

AMD's open-source commitment continues through the Radeon Open Ecosystem (ROCm) software solutions and a demonstration of the GPU running a Cinema4D rendering workload with the company's open-source Radeon Pro Render ray tracing solution, standing in stark contrast to proprietary solutions like CUDA. The 32GB of HBM2 puts the company on par with Nvidia's recent adjustment that boosts the Tesla V100 up to 32GB, though AMD did not reveal whether it will also offer versions with 16GB of HBM2, which could address current sky-high HBM2 memory pricing. AMD did not reveal specifications, pricing, or a definitive timeline for mainstream graphics cards.