2. RDNA2 Architecture
Laura: Our graphics architecture roadmap has a clear goal: gaming leadership across the ecosystem.
The GCN was a great architecture, but we knew we needed to rethink graphics and build the best engine for the future. We wanted to build a set of technologies that game developers could use across all platforms. The first major step was RDNA — introduced last year.
RDNA was a new architecture and use the 7-nanometer process node, which helped us with the large initial gains.
We committed to you that our engineers had plans in the works for RDNA2, and we would be back quickly with another significant improvement. By combining our expertise in CPU design, deep gaming history and broad ecosystem partnerships RDNA2 is built to be the gaming architecture for all gamers across console, PC and beyond.
We had some very ambitious goals with RDNA2. Significantlyj, higher performance, 50% improvement in performance per watt, and new hardware features.
Individually, each of these goals is challenging. When combined, they become extremely difficult because as you push performance, a natural side effect is more power.
New features, also being more computation, further challenging both speed and power metrics. For RDNA2, we made significant gains in all these areas. The compute unit is the core of the architecture. We advanced all aspects of the compute unit design, leveraging techniques first developed in CPUs.
We deployed pervasive fine-grain clock gating. Re-imagined the pipeline to aggressively maximize performance while minimizing data movement. In combination, these elements made the compute units 30% more energy efficient allowing them to be pushed faster.
Gaming at 4K requires more data to be brought into the chip. The amount of data more than doubles as we move from 1440p to 4K. The more efficient the memory subsystem is… the higher fidelity the images are, and the more you can push the framerates.
We took a new approach to solving this technical challenge for gaming and deployed a graphics optimize high-density, high-speed cache based on the Zen L3 cache. We call this the Infinity Cache.
The Infinity Cache minimizes DRAM bottlenecks, latencies, and power consumption. Our specialized Infinity Cache combined with a 256-bit G6 interface delivers more than twice the Effective Bandwidth of a traditional 384-bit G6 solution — and at lower power consumption, high frequency is important for gaming performance.
RDNA2 is built with custom libraries and high-speed design flows, maximizing frequency in 7nm.
With RDNA2 — our silicon design and architectural improvements, we are able to deliver 30% faster frequencies in the same 7nm process node.
Pulling all this together, the RDNA2 architecture beats our aggressive performance-per-watt goal — achieving 54%.
With RDNA2, we also focus in bringing new hardware features into the architecture. We partner closely with Microsoft on DirectX 12 Ultimate features, including:
- DirectX Ray Tracing
- Variable Rate Shading
- Mesh Shaders
- Sampler Feedback.
RDNA2 will also support Microsoft DirectStorage API. We’re excited about this new API that will enhance the gaming experience by significantly reducing game asset load times from large SSDs.
We compared the performance of RDNA2 to our prior generation. And when we put it all together — on average — RDNA2 delivers double the performance and with great power efficiency.
We are so proud of what we’ve accomplished with RDNA2, and we are not stopping. The team is executing to plan. RDNA2 is here and the RDNA3 team is well into execution. Let me welcome Scott Herkelman to show RDNA2 in our new RX 6000 series products.
| AMD RADEON RX 6000 EVENT TRANSCRIPT | ||
| 1. Intro | 2. RDNA 2 Architecture | 3. AMD Radeon RX 6000 Series |
| 4. AMD Partners: Game Developers | 5. AMD Radeon RX 6800 XT Pricing | 6. AMD Radeon RX 6900 XT Pricing |
Hope you enjoyed this article. Please, support Blizzplanet via PayPal or Patreon, and follow us on Twitter, Facebook, YouTube, and Twitch for daily Blizzard games news updates. |
 |
