Vulkan, the Future of Open Graphics APIs

Vulkan, the Future of Open Graphics APIs

By | May 1st, 2015
No Comments on Vulkan, the Future of Open Graphics APIs

Why have some of the biggest players in the industr decided to create

entirely new graphics APIs all of a sudden? Is this an unlikely coincidence

or does it signify some kind of a larger shift in the industry?


														
							

Apple’s Metal, AMD’s Mantle, Microsoft’s DirectX 12, Khronos’ glNext and now Vulkan, the world of graphics computing seems to have exploded with new APIs, standards and frameworks of late. One has to wonder what exactly is going on.

If we look at what all of these have in common, it becomes clear that it is the latter, a growing consensus in the industry about what the future of graphics development should look like. This common base that all the above technologies share, is a focus on making a graphics API that gives developers more direct access to the hardware, with fewer layers of abstraction and code between the hardware a developer writes, and software a developer writes.
To understand why this is important for the future of graphics processing, let us first look at a brief history of GPU technology, and graphics APIs.

A Brief History of the GPU

What we would today call a GPU – the term was coined after the technology had already existed for many years – has been around for over two decades now, and it has not been immune to Moore’s law. In many ways the advancement of GPUs has been even more substantial than that of the CPU. The architecture of GPU and the approach to graphical software development have changed substantially.

One of the biggest such changes has been the move from a fixed-function pipeline to the shader-based pipeline that exists today. The very first GPUs offered what is called a fixed-function pipeline, where the graphics hardware implemented a set of common graphics functions for performing transformations and applying graphical effects. This was the model used till DirectX 7, and OpenGL 1.5 and was the only system supported by the earliest NVIDIA cards.

In the early 2000s, GPU began opening up, allowing more and more of their functionality to be programmed by software developers using small bits of code called shaders rather than using a fixed set of functions in the graphics hardware. OpenGL 2.0 and DirectX 8.0 facilitated the use of this new found flexibility in GPU hardware by including their own shader languages, GLSL (OpenGL Shading Language) for OpenGL and HLSL (High Level Shader Language) for DirectX. Code could be written in these languages and then deployed to be run in the GPU.

GPUs continued opening up to the point that modern GPUs are in essence highly-parallel CPUs. Today you have frameworks such as CUDA and OpenCL that allow a developer to utilise the power of a GPU to perform tasks that would traditionally be performed by a CPU.

To give you an idea, currently the latest NVIDIA GPU architecture is Maxwell, which powers its current top of the line GTX 980 GPU. This GPU has 16 streaming multiprocessors, each of which has 4 warp schedulers, which in turn have 32 CUDA cores. Leave the confusing technology aside, and think of this as equivalent to a 16-core CPU, where each CPU has the equivalent of Intel’s HyperThreading, but instead of making a single core appear as 2, it allows for a single core to appear as 128! In the meantime we have seen the release of OpenGL 3.x and 4.x, and the release of DirectX 9.x, 10.x and 11.x. While graphics APIs continued to evolve, they also abstracted away the true power of modern GPUs in order to provide an API that was designed for old hardware. They also add an overhead, in terms of CPU usage that limits what is possible through the use of these APIs.

The Growing Consensus

The uniformity and simplicity of old OpenGL and DirectX comes with an overhead that can sometimes be limiting. This is more true perhaps of mobile platforms such as Apple’s iOS, where the CPU and GPU are both limited, and every bit of inefficiency means lower battery life, or less optimum use of the available hardware. Imagine a powerful GPU that is capable of brilliant graphical fidelity, but is let down by the slower CPU of a The developer is using the GPU through a layer of software (OpenGL 4.x or DirectX 11.x) that need to process each command sent to the GPU using those APIs.

However these commands themselves use up the CPU to the point where we just can’t send instructions to the GPU fast enough to make full use of it.

cores

Vulkan is better at spreading load across multiple cores

It is for this reason that Apple created their Metal API for iOS, giving game developers for iOS devices more direct control over the graphics hardware, and the capability for better performance, or better visuals with the same hardware. In fact console developers creating games for the Sony Play- Station and the Microsoft XBox have had access to low-level APIs because of which you can see such hardware achieve levels of graphical fidelity that are just not possible with desktop computers with much better specifications.

Both Mantle, by AMD, and DirectX 12 by Microsoft, which preceded Apple’s Metal are built around the same principles. Think of them as a reboot of the graphics APIs which were designed for another time period and are now showing their age.

This was also a little unfortunate, since Mantle (when announced) was an AMD-only technology and would only work on AMD hardware, that too only on Windows. DirectX 12 is a Microsoft-only technology that would only work with Microsoft platforms such as Windows, Windows Phone and the XBox. Metal was an Appleonly API that would only run on Apple platforms, and currently is only available for iOS.

What about a standard graphics API that could work on all operating systems, whether it is OSX, Linux, Windows, iOS or Android, and on all kinds of hardware whether by AMD, NVIDIA, or Intel. OpenGL had always been the solution that could go anywhere due to its open nature, but it hasn’t suddenly disappeared in a puff of smoke now that these newer APIs are available. It will continue to be possible to use OpenGL to create games and other applications needing hardware acceleration.

Go to Page: 1 2

Google
Ashutosh Gupta
Share a little biographical information to fill out your profile. This may be shown publicly. In WordPress version 2.9, a new filter was added to make changing the [...] string in the excerpt. Do not use both of these methods in