What’s a GPU guy doing praising a CPU design? It’s true that the two processing architectures have traditionally played completely separate roles in the computing ecosystem, but AMD recognized long ago that a radical change in system architecture was needed to continue expanding available compute performance while keeping power consumption in check. This challenge requires blending the performance of the CPU and the GPU in a heterogeneous system, which opens up the possibility of a step function increase in performance-per-watt. One of the most exciting applications of heterogeneous computing, one that’s proving to have tremendous potential, is deep learning (and its close cousin, artificial intelligence). The possibilities for these technologies are myriad, from self-driving cars to advanced robotics to healthcare to fraud detection, but what all these applications have in common is the need to crunch massive amounts of data.
Starting with the introduction of the Accelerated Processing Unit (APU), AMD began to explore what is possible when a CPU and GPU are more tightly interconnected. AMD took an early lead in this approach to computing. By bringing together a collection of like-minded companies in the Heterogeneous System Architecture (HSA) Foundation, AMD helped establish an open, royalty free standard that could be leveraged across the industry. Those strategic decisions were made with the intention of better harnessing both architectures to address the demands of future generations of computing.
With the introduction of Radeon™ Instinct and the promise of the upcoming “Vega” architecture, AMD is accelerating the machine intelligence era in server computing. This is possible through a new suite of hardware and open-source software offerings designed to dramatically increase performance, efficiency, and ease of implementation of deep learning workloads. As excited as I am about our future in GPU computing and everything we are doing in high-performance graphics, as a technologist, I know how important the CPU remains in everything we do at AMD and how critical it is to the future of deep learning.
AMD is preparing now to launch its latest CPU, codenamed “Naples”, in Q2 of this year. This 32-core, 64-thread powerhouse will mark a new high-water mark for AMD in performance potential, leveraging the new “Zen” x86 core alongside exceptional memory and I/O capability. Deep learning is highly complex, and its foundational elements require the kinds of innovations in multiple cores, memory capacity, bandwidth and I/O capacity that “Naples” will deliver.
- With the highest number of cores available on an x86 processor today, “Naples” supports greater parallelism across the deep learning process, and combined with GPUs it can support faster and deeper decision-tree analysis, which helps to speed-up artificial intelligence algorithms.
- With the highest number of cores available
- The massive parallelism and computational intensity of the deep learning training phase is perfectly suited for GPUs, in combination with “Naples” rich I/O to attach multiple GPUs in a server.
- And in the inference phase, high memory bandwidth and large memory capacity from “Naples” can combine with our GPUs for efficient, high performance execution of larger deep learning applications.
Truly accelerating the pace of deep learning and addressing the broad needs of the datacenter requires a combination of high performance compute and graphics innnovation. Radeon Instinct and “Naples” connect these two processing architectures so we can support the ever-growing demands and nearly unlimited potential of machine intelligence. This radical change in architecture holds the promise of significantly expanding compute capabilities while helping lower total cost of ownership for businesses. I can tell you, this GPU guy is completely on board.
Ogi Brkic is Senior Director, Radeon Technologies Group, AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.