Microsoft’s (NASDAQ: MSFT) upcoming DirectX 12 has a ton of exciting features and optimizations that we already know about. However the most interesting new feature comes via a report from Tom’s Hardware: the ability to pool resources from multiple GPUs from different manufacturers.
The ability to work with two different GPU architectures in the same system comes from DirectX 12’s support of something called Explicit Asynchronous Multi-GPU capabilities and Split Frame Rendering (SFR). Effectively this pools all GPU resources into a bucket and allows them to be utilized as one.
SFR allows developers to manually divide data between the two GPUs to allow them to work together on each frame with the work for each frame divided between the two cards. The older method, called Alternate Frame Rendering (AFR). AFR is considered to be less efficient because it required both GPUs to have all of the data already in their frame buffers. Each GPU would render an alternate frame. The downside of this, however, is that the cards would work in parallel and not independently. If you had two cards with 4GB of memory each, you’d only have 4GB of useable space in the frame buffer in total.
The kicker is that Tom’s source said that SFR will be supported across multiple GPU architectures in the same system. It will treat both GPUs as one. Reportedly this would allow a system with an AMD (NASDAQ: AMD) APU and an Nvidia (NASDAQ: NVDA) GPU to have the two GPUs work together. The same would go for a discrete AMD and Nvidia GPU in the same system.
In the end this will be up to developers to make use of and optimize their code accordingly. AMD’s Mantle already supports a form of SFR, so in a way the precedent is already there.
More details on DirectX 12 — and a confirmation of this report — will likely be available during March’s Game Developer Conference in San Francisco.