Intel Core i5-3470 Ivy Bridge Processor and HD 2500 Graphics Review
Rightware’s Basemark CL provides compelling feature set to coordinate parallel computation across heterogeneous processors utilizing OpenCL. Image manipulation tests run filters on image data and produce the filtered output on the screen. The image manipulation filters are applied to video streams as a separate test, which allows benchmarking of bandwidth limitations from moving data from CPU to GPU, providing a solid real-world case for benchmarking. Physic tests enable leveraging of extra computing power that OpenCL brings into different platforms. The feature tests provided by Basemark CL enable the testing of performance of single or several features on the hardware.
Physical simulations are good area for leveraging extra computing power which OpenCL brings. Physics simulations are already heavily used in PC-games, but not to a greater extent in mobile games or in applications doing heavy 3D physics simulation. With the help of good OpenCL performance it is possible for future mobile games and applications to include more physics based animations and game elements.
Basemark CL contains fluid and cloth simulations. Cloth simulations are widely used in games to add extra realism to cloths of characters, flags and other soft body items. Fluid simulations are used in games to simulate smoke, water and other materials. The realistic simulation adds possibility for player interaction with visuals, which is not possible with currently used artist made animations.
The Mandelbulb fractal is a type of 3D fractal. The fractal was recently discovered by Daniel White and Paul Nylander.
3D fractal is rendered using ray tracing. The ray tracing for the 3D fractal uses the basic ray tracing formula. For each point in the canvas a ray is casted towards to the object to be rendered. Nearest distance of a point to the surface of the Mandelbulb is calculated. The power of the Mandelbulb is changed constantly so the polynomial solution for power 8 Mandelbulb is not used, therefore this test benchmarks the performance of built-in mathematical functions in the same vein as Julia fractal benchmarks the raw ALU performance. If the point is within predefined distance of the fractal surface the ray tracing is stopped and lighting calculations are done, if not the point is moved along the ray forward until the surface is found or predefined maximum iteration count is exceeded.
Benchmarking image manipulation performance is useful due to fact that image manipulation filters are widely used in image processing and computer machine vision applications. Various image manipulation filters are used in machine vision to find out discontinuities and remove unwanted features from images. Image processing in embedded devices has been increasingly important since the advent of cheap digital cameras. OpenCL devices may allow fast image enhancement on a modern smartphone without degrading usability nor requiring application specific circuits to perform this task.
Image manipulation tests run filters on image data and produce the filtered output on the screen. The implemented filters are listed below in separate sub paragraphs. The images in the paragraphs demonstrate how the corresponding filters affect the image. In addition performance of some of the image processing filters produces a good indication towards performance of other signal processing tasks, such as audio processing.
The image manipulation filters are applied to video streams in a separate test. This allows benchmarking the bandwidth limitations from moving data from Host to OpenCL device (e.g from CPU to GPU) and provides a good real world use case. Time spent decoding the video is not taken into account when determining benchmark score.