Microsoft’s Graphics Technology Work Graphs, part of the latest version of DirectX 12, has been released. Aimed at optimizing systems with low-power CPUs, the software enables more efficient task scheduling on the GPU to lighten the load on the CPU.
Traditionally, the CPU and GPU work together to form the final picture on the monitor. However, at times, the CPU fails to prepare tasks for the GPU swiftly enough, leaving the GPU idle while waiting for instructions. Work Graphs is expected to make GPUs more autonomous and reduce their reliance on the CPU for work orders.
With Work Graphs, game developers can streamline their pipelines, shifting heavy workload from CPU to GPU efficiently. A stream functioning on the GPu (the producer), can create a request for other tasks (consumer). The GPU schedules the work once resources are available. If needed, the producer can also switch roles and become the consumer.
Shader Model 6.8 GPU Upload Heaps feature is set to further accelerate the interaction between CPUs and GPUs. It forgoes the need for data transfer from video memory to the CPU through the PCI bus. This function leverages ResizableBAR technology, permitting the CPU to quickly write data into video memory with minimal latency.
Work Graphs, along with Shader Model 6.8, have been provided to game developers via the Agility SDK 1.613. This SDK can be used with the newest drivers from AMD and NVIDIA.
Work Graphs technology is supported by NVIDIA’s Ampere and Ada Lovelace (GeForce RTX 3000 and RTX 4000) graphics processors, given that they have drivers of version 551.76 or newer installed. On AMD’s end, the technology is currently functional only on RDNA3 architecture GPU’s (Radeon RX 7000 graphics cards).
However, it is crucial to note that updating the video driver alone will not enhance performance. For Work Graphs to function optimally, it must be implemented in games, pushing game developers to support the initiative.
NVIDIA has also published a technical demonstration to evaluate the application of Work Graphs. In the video above, the technology’s performance can be assessed (performance growth can be noticed around the 1:20 mark) during the testing of GeForce RTX 4090 and Ryzen 7 7700X processors.
