GDC kicks off today, and there’s news from Khronos that the next version of Open GL (dubbed gl Next), will be officially known as Vulkan, the German word for volcano.
Like Mantle, Direct X 12, and Apple’s Metal API, Vulkan is designed to be a low-overhead API that offers superior multi-threaded and multi-GPU performance options, giving developers more control over how workloads are dispatched and processed across multiple graphics cards.
Power VR has posted an early video (below) of what Vulkan can accomplish, though it should be noted that this is an exceptionally early product running on alpha drivers and a draft version of the specification.
The company notes that Vulkan, unlike Open GL ES, is designed to run as close to the GPU architecture as possible, and includes multiple features that reduce CPU overhead and allow for more efficient task execution.
One of the advantages to Vulkan is that the specification has been designed to meet the needs of more GPUs than just Mantle.
Vulkan now supports features like the render pass, which allows the GPU to explicitly tell the application how render targets should be loaded in and out of the GPU at the start and end of each render.
This is critically important to the function of tile-based renderers, and Power VR notes that this new capability helps avoid expensive buffer flushes that could occur under Open GL ES.
As with Mantle and D3D, multi-threading is now far easier with Vulkan, as is memory management.
Imagination Technologies isn’t talking up multi-GPU rendering targets, but explicit memory management options might make it possible to share data across multiple GPU clusters more efficiently (if such products are ever built), or to distribute workloads within the GPU.
Vulkan’s announcement essentially closes the last gaps in low-overhead API availability.
Mantle, Direct X 12, and Apple’s Metal API all focus on offering many of the same performance benefits, but they limit the operating environment in one manner or another.
Vulkan, in contrast, will be available to every operating system and vendor, at least in theory.