Work Graphs on Radeon GPU, up to 39% faster rendering
AMD demos the potential of fully GPU-driven rendering.
Microsoft officially announced Work Graphs last week, but the formal introduction is taking place at GDC (Game Developer Conference). Attendees, primarily game developers, are shown the techniques and methods to utilize this new technology in games, which are designed to provide visible performance improvements.
As mentioned before, Work Graphs allow game developers to put more work on the GPU, rather than rely on the CPU to submit work for the graphics processor. The GPU-driven rendering is something that can limit the CPU bottleneck, generate fewer calls and, as a result, deliver higher throughput and rendering performance, something that has been advocated by Epic Game Unreal Engine developers for years.
AMD has showcased that Radeon RX 7900 XTX using Work Graphs can work 39% faster compared to conventional method. In other words, the standard technique involving ExecuteIndirect is 64% slower. Those are ‘super early numbers’, likely to be optimized in the future.
According to AMD, this demo is using the following settings:
- 6600 draw calls/frame (after coalescing)
- 13 million triangles/frame
- 200.000 work items passing through the graph
- 37 nodes and 9 draw nodes
- < 200 MiB of work graph backing store memory
The demo was put together by Cuburg University, who call Work Graphs a major step in graphics programming.
AMD has conducted tests on Radeon RX 7900 XTX GPU featuring a pre-release driver that enables “ExecuteIndirect” command and Work Graphs with Mesh Shading extension. The system was also equipped with Ryzen 7 5800X processor and 32GB of DDR4 memory.
Source: GPUOpen