With the rise of intelligent artificial intelligence, Unified Memory Architecture (UMA) has gained widespread attention. 

These UMA products tightly couple CPU, GPU, and memory to provide a unified chip-level system and have become a widely adopted solution in the field of artificial intelligence.

With the release of the AMD Ryzen AI MAX series and the addition of products from other vendors such as NVIDIA RTX Spark, these architectures are gradually becoming the general foundation for artificial intelligence.

In a discussion with AMD's David McAfee about whether more products would adopt the UMA architecture, he stated that these architectures would be a key focus going forward. 

AMD's first-generation MAX products offered up to 128GB of video memory and could allocate up to 112GB of system memory to the GPU. NVIDIA's RTX Spark also works similarly, dynamically allocating memory between the GPU and CPU based on workload. These dynamically unified memory architectures are suitable for a wide range of applications.

UMA's help

AMD believes UMA will help shape its next-generation architecture, products, and roadmap. They are already working on the upcoming Ryzen AI MAX 400 series, processors that can offer up to 192 GB of VRAM and allocate up to 160 GB of VRAM to the GPU, supporting over 300 billion AI Hierarchical Models (LLMs).

David McAfee believes that with the release of Strix Halo and NVIDIA's entry into this field, UMA systems will receive a lot of attention. People will be very interested in finding the right architecture for these UMA systems and how to improve upon the existing state-of-the-art technologies on these platforms.

David McAfee acknowledges that he doesn't know which direction the technology will take in the next few years, but believes it opens up a world of possibilities because it represents a completely new workload. AMD is addressing a completely new area of computing, and he thinks it will influence AMD's product choices, product roadmap, deployment strategies, and so on.

AMD also endorses NVIDIA's RTX Spark because it's an upgrade to the UMA architecture. Interestingly, David also mentioned high-performance desktops when discussing UMA.

Of course, he wasn't saying that desktop platforms would soon adopt UMA technology. However, the previous question did explicitly ask whether AMD would see Ryzen desktop CPUs with UMA technology, or Ryzen MAX SoCs with 3D vertical cache or higher-end packaged memory designs, to further enhance UMA's performance through tighter integration of fast, low-latency technologies.

What excites David is that, in his view, AMD is standing on the threshold of an exciting era that will fundamentally change AMD's understanding of concepts like high-performance desktops and unified systems. He believes that continuously improving support for the unified memory architecture and getting more vendors in the ecosystem to adopt it will only further enhance the performance and value of these devices and strengthen ecosystem support for them.

AMD remains convinced that its unified architecture built on Halo is the best choice for these platforms. NVIDIA's announcement corroborates this, as they also believe this architecture is suitable for these systems. AMD believes that the emergence of Agenta Compute, and the ability of these systems to run extremely large models with a unified memory pool, brings unique value to the entire computing field. For AMD, these systems play a dual role in its overall product portfolio strategy.

Unified memory architecture is no longer a niche experiment but is rapidly becoming the cornerstone of next-generation computing. With intelligent artificial intelligence driving the demand for massive shared memory pools, both AMD and NVIDIA have fully supported UMA technology, demonstrating its widespread industry acceptance.

AMD's confidence in this direction is evident in the ambitious Ryzen AI MAX 400 series and its openness to high-performance platforms like Strix Halo, indicating that AMD is just getting started. As unified systems blur the lines between CPU, GPU, and memory, they promise to unlock entirely new performance, efficiency, and capabilities, not only for AI workloads but also potentially for gaming and high-end desktops. The future of UMA is incredibly bright, and the entire ecosystem is just beginning.

Source: Compiled from wccftech

View more at EASELINK