PowerVR MBX - Creating an IP standard for advanced mobile multimedia
Kings Langley, UK
Abstract :
This paper will describe Imagination Technologies’ PowerVR MBX graphics core, which has emerged as the de facto standard IP solution for advanced mobile multimedia applications.
Mobile multimedia needs
Providing high-performance, low-power IP cores for today’s mobile multimedia SOCs is incredibly challenging. Handheld multimedia platforms must provide digital audio and digital video playback fidelity equal to their set-top or component counterparts, yet must also provide extremely long battery life. Most challenging of all is the provision of high-performance, high-quality 3D graphics. Handset manufacturers demand 3D graphics capabilities equal to or exceeding desktop systems only 2 generations old, yet which consume only milliwatts of power. Of course, the shrinking feature size of silicon processes and reduction in switching voltages in SOCs both contribute to solving this problem; however these advances are lost without an architecture designed to provide the best blend of power, performance and features.
No wonder then that PowerVR MBX has emerged as the IP core of choice for the most demanding SoC designs. MBX delivers millions of triangles per second at fillrates of 100s of million of perspective-corrected textured pixels per second, yet dissipates orders of magnitude less power than equivalent designs for the desktop, and its architecture ensures that it places lower demands on the memory subsystem than other architectures, saving bandwidth and power.
MBX history
MBX is an IP Core born of over ten years development – the same architecture which has been used in the Sega Dreamcast console and Naomi arcade system and also powered a range of PC chips including ST Microelectronics KYRO and KYROII. MBX development began in 2000 and was first licensed to lead partners in 2002.
MBX design goals
MBX’s brief was simple: Create an IP Core for handheld multimedia and gaming devices with the features and performance to allow games developers to deliver compelling content, and power consumption low enough that console-class 3D gaming on the move could become a reality.
3D standards
Open standards have been key to wide adoption of 3D Graphics in the handheld market. Imagination Technologies has been a leading promoter of the Khronos group working to bring OpenGL ES to the market. It also works with Microsoft, Symbian and other leading companies in the handheld market to ensure availability of standards on the MBX core which is compliant with both OpenGL ES and the upcoming Direct3D Mobile API for WinCE. This has provided content developers a stable, standard platform upon which they can build compelling content.
MBX architecture
Handheld graphics has been mainly software based to date. In order to enable the features and performance desired in the power required, MBX has eschewed the traditional brute force approach of most 3D graphics hardware and instead adopted Tile-based rendering, an alternative 3D rendering technique.
A traditional Immediate Mode Renderer (IMR) rasterises (textures and shades) all geometry submitted to it, and decides which pixels to display by a Z Buffer test before committing them to the framebuffer: PowerVR’s Tile based rendering determines which pixels are visible prior to rasterisation and only renders those which contribute to the final scene. This approach, which reduces the number of pixels processed by the graphics engine, saves bandwidth and power with improved quality.
Figure 1 – MBX Architecture
Reduced Bandwidth
Bandwidth is reduced because non-visible pixels are rejected prior to texturing. This is achieved by performing all the geometry processing prior to “tiling” the transformed and lit data; this operation stores the entire scene’s geometry and references it by which tile (area of the screen) it occupies. Once complete, each tile containing all the data required to render that screen area is processed in turn. All objects in the tile are tested for depth and only the visible objects are forwarded for shading and texturing. Bandwidth is reduced because the depth values for the tile are held on-chip, completely eliminating Z-buffer read modify write accesses; in addition, because only visible pixels are textured, texture read bandwidth is vastly reduced – often by up to a factor of 4; finally, because the entire pixel data for the tile is stored on chip in an accumulation buffer, the read modify write bandwidth associated with pixel blending is also eliminated.
Lower Power
Reduced memory bandwidth leads directly to reduced power dissipation by reducing IO activity between devices.
Because MBX is required to shade and texture fewer pixels, it can achieve the same performance as an immediate mode renderer with fewer gates devoted to texturing and shading – again saving power.
MBX was designed with low power operation in mind and its processing pipelines are automatically shut down at the completion of rendering.
Improved Quality
Because of its unique architecture, MBX is able to use a number of techniques not available to Immediate Mode Renderers (IMRs) in handheld devices and which improve quality significantly.
FSAA: MBX supports full scene anti-aliasing for significantly improved image quality especially on lower resolution handheld screens. MBX’s architecture allows the FSAA to be performed entirely within the core without impacting Z or pixel bandwidth requirements which would otherwise make this technique too costly in bandwidth and memory footprint.
Figure 2 – zoomed image without antialiasing.
Figure 3 – same scene with 2x2 antialiasing.
The increase in image quality from exploiting 2x2 anti aliasing is critical when rendering on low resolution (320x240 or 480x272) handheld devices, quadrupling the apparent display resolution.
Floating point Z: Because MBX stores its Z values on chip, it can calculate them using floating point precision rather than the 16-bit fixed point Z invariably used by IMRs to keep bandwidth down.
Internal True Colour: MBX processes all shading and blending operations internally in 32 bits colour precision even when using lower colour resolution displays and framebuffers. This is only possible on tile-based architectures where all colour processing is done in the 32 bits ARGB accumulation buffer.
On immediate mode renderers, image quality is a direct result of the frame buffer colour depth. With a 16 bits colour depth frame buffer, multiple reads and writes will result in a loss of precision for the final colour. This problem usually creates a “banding” effect on the screen. Standard dithering techniques do not help IMRs and can potentially create worse result by giving the image a “grainy” look. On PowerVR, each pixel is only written once to the frame buffer, making dithering a benefit when converting the final 32bit image data to its target colour resolution. The benefits are shown in the two screenshots below.
Figure 4 – IMR blending.
Figure 5 – MBX blending.
Advanced Features
MBX boasts a range of advanced features to support modern content development. These include:
Vertex Shading: A fully programmable Vertex Shader, the VGP, allows the developer to write custom lighting models and transforms and includes the ability to pre-process vertex data back to the host as well as fully programmable hardware clip planes.
Tesselation: Fractional tessellation of Bezier patches allows geometry compression as well as adaptive control of geometry complexity depending on distance.
Matrix blending: Matrix blending is an important feature in the animation of models in order to ensure that geometry at joints (e.g. the elbow) remains continuous.
MBX performance
MBX offers quite extraordinary performance for a handheld 3D solution. Polygon rates in excess of 4M triangles/sec can be achieved from a 100MHz clock, with pixel fill-rates in excess of 400M perspective-correct textured, MIP-mapped pixels/sec. Achieving this performance in Unified Memory Architectures with cost effective memory sub-systems is only possible because of the bandwidth saving benefits of the Tile Based architecture.
MBX Software Support
With complex IP, broad software support is also a key factor in enabling the market. The handheld 3D market has reacted quickly to the challenge of producing standards as mentioned above, and Imagination Technologies, as well as supporting the creation of the standards, has developed a wide range of driver software which exposes these APIs for MBX across a range of Operating Systems including Linux, SymbianOS and WinCE.
Complemetary SOC cores
Imagination Technologies’ offers a range of complementary video and display cores allow complete multimedia systems to be built around the core graphics capabilities of PowerVR MBX. MBX itself has features that ease integration of video and 3D into SOCs including YUV texture format support, so that video decoded can be used directly as textures by MBX. Other cores include multi standard video decode accelerators, configurable display pipelines and TV Encoders are also available.
MBX adoption
Since its first availability in 2002, MBX has become the de-factor hardware graphics standard in Handheld Multimedia, Mobile Phones and Care Information Systems.
Licensees include five of the world’s top 10 semiconductor manufacturers:
Figure 6 – MBX Licensees
Announced SOCs targeted at the mobile market and incorporating MBX include Intel’s 2700G, Renesas’ SH Mobile3 family and TI’s OMAP 2 devices. Intel’s 2700G is shipping in the Dell Axim X50v PDA.
In Car Information Systems Renesas’ SH7770 which also includes MBX is shipping in systems from Mitsubishi and Pioneer.
A wide range of other SOCs using a range of CPUs and targeting a broad range of applications are in development using PowerVR MBX technology.
Conclusion
By developing an IP Core based on a novel graphics architecture designed to address the needs of highly integrated SOCs, focusing on reducing system power and cost, targeting a rich feature set and high performance and delivering a complete hardware and software solution Imagination Technologies have succeeded in making the PowerVR MBX family a de-facto standard for graphics acceleration in the advanced multimedia market.
Acknowledgements
Registered names and trademarks are the property of their respective owners.
|
Imagination Technologies Group plc Hot IP
Related Articles
- Using audio codecs IP as the digital audio hub in mobile multimedia systems
- How customer-specific standard products ease mobile device design
- Analysis: CEVA's ''Lite'' Mobile Multimedia Platform
- Creating multi-standard, multi-resolution video engines using configurable processors
- Advanced Packaging and Chiplets Can Be for Everyone
New Articles
- Quantum Readiness Considerations for Suppliers and Manufacturers
- A Rad Hard ASIC Design Approach: Triple Modular Redundancy (TMR)
- Early Interactive Short Isolation for Faster SoC Verification
- The Ideal Crypto Coprocessor with Root of Trust to Support Customer Complete Full Chip Evaluation: PUFcc gained SESIP and PSA Certified™ Level 3 RoT Component Certification
- Advanced Packaging and Chiplets Can Be for Everyone
Most Popular
- System Verilog Assertions Simplified
- System Verilog Macro: A Powerful Feature for Design Verification Projects
- UPF Constraint coding for SoC - A Case Study
- Dynamic Memory Allocation and Fragmentation in C and C++
- Enhancing VLSI Design Efficiency: Tackling Congestion and Shorts with Practical Approaches and PnR Tool (ICC2)
E-mail This Article | Printer-Friendly Page |