AMD has revealed the details of its GPUOpen project, through which it hopes to carve a bigger niche for itself in the crowded market for graphics processors in x86 machines.
The company is being squeezed from above by Nvidia, which dominates the high end of the GPU market, and from below by Intel, which can leverage its larger share of the x86 CPU market to sell its own integrated graphics chipsets. While it can’t do much about Intel, GPUOpen could allow AMD’s Radeon chips to score a few points off Nvidia.
With GPUOpen, AMD is giving software developers the code and specs they need to squeeze the most out of its GPU chips by programming directly to its APIs rather than graphics hardware abstraction layers such as DirectX or OpenGL, which don’t offer the same level of control over specific processor features.
While AMD is showing developers a shortcut to better performance, the path to improved profitability is a long and circuitous one. It first requires programmers to take AMD up on its offer, first announced in December, to use its new APIs to optimize their code for its hardware. With those tweaks, and without the overhead of the abstraction layer for some functions, a Radeon GPU might perform better than a more expensive rival. It’s that promise — that gamers might get more bang for their buck by buying machines with AMD’s Radeon graphics chips inside — that the company hopes will make GPUOpen a source of profit.
This is not AMD’s first offering intended to woo developers away from the hardware-independent DirectX and OpenGL APIs. The company abandoned the last one, Mantle, a year ago, handing the code to OpenGL developer Khronos, which used it as the basis of the specification for OpenGL successor glNext, now known as Vulkan.
Opening the GPUOpen website for business on Tuesday, AMD’s senior manager of worldwide gaming engineering Nicolas Thibieroz wrote that the first goal is “to provide code and documentation allowing PC developers to exert more control on the GPU,” including “many features not exposed today in PC graphics APIs.”
That will help developers more economically code games for both PCs and consoles, where they already have lower-level access to the GPU, he wrote.
While GPUOpen’ primary focus is on games and other CGI applications, another big application area for GPUs is in accelerating other computational tasks. AMD is also targeting these through GPUOpen’s “Professional Compute” branch, offering optimized open-source drivers and standards-based libraries for its chips.
The site already includes a bunch of application examples and sample code, including HIP, a tool for converting code from the Nvidia-backed CUDA parallel-computing API to portable C++ that can then be compiled to run on Nvidia or AMD GPUs. The HIP code is on Github, like much of AMD’s other GPUOpen code.