Founded in 1997, Futuremark is a company that continuously develops one of the greatest benchmark tools for measuring performance of computer hardware and software. Formerly known as ORB (Online Result dataBase), Futuremark today owns the world’s largest and most comprehensive hardware performance database with over 50,000,000 available benchmark results, mixing thousands of different CPU’s, GPU’s and mobile devices.
Until recently, it was impossible to offer apples-to-apples comparison between different devises as they all used different API (Application Programming Interface). Luckily, with the work of The KHRONOS Group, Vulkan API is now gaining in popularity. In order to keep with times, Futuremark added Vulkan support their 3DMark API Overhead feature test.
This enables the benchmark to compare the API performance between DirectX 12, DirectX 11 and Vulkan API within a single benchmark. In general, Vulkan API derives from core elements of AMD’s Mantle project, but the API evolved significantly since then and today it represents a valid API supported by all the major players such as Apple, Google, Intel, NVIDIA, Qualcomm and of course, AMD themselves.
Vulkan puts less burden on the CPU and should be able to better distribute work inside the multi-core CPU. When it comes to the GPU portion, Vulkan represents a step up in efficiency as it provides a low-level access to the modern GPUs. Key advantages are streamlining the code execution and eliminating software bottlenecks (particularly draw calls). A draw call happens when the CPU tells the GPU to draw an object on the screen. In a modern 3D or VR application, CPUs will make thousands of draw calls per frame, with each call creating a performance-limiting overhead for the CPU itself. At one point, contemporary graphics engines will become flooded by an API overhead (software bottleneck).
Vulkan, DirectX 12 and Metal API all focused on reducing that overhead, enabling more objects, textures and effects to be be drawn on the screen. Which brings us to Futuremark and 3DMark feature tests. This benchmark is more artificial than real-world uses, and will truly test the practical capabilities of the platform it is being tested on. This feature test draws a very large number of individual buildings using simple shaders and no lighting effects. Its main purpose is to compare the relative performance of Vulkan, DirectX 12 and DirectX 11 APIs on a single system. The focus on single-system testing is the primary reason why the API Overhead Test is called a feature test instead of being a benchmark. The test increases a large number of draw calls and lastly, measures the number of draw calls per second achieved by each API before the frame rate drops below 30 FPS. The API Overhead feature test also works for iOS devices, but than test measures the difference between OpenGL ES 3.0 and Metal APIs.