The Next Standard for PC Graphics Memory

Pressure is mounting on PC graphics system designers to boost performance as interest grows for larger displays, more colors and more sophisticated graphics capabilities such as 3D and video. Until now, designers have been able to squeeze enough performance from their frame buffers to meet system requirements by using faster synchronous DRAM or video RAM along with ever wider data paths to the graphics controller. But this architectural approach has hit a performance wall, and further increasing the complexity will only make these solutions impractical for mainstream PC applications. An alternative architecture using high-bandwidth Rambus DRAM meets the most demanding performance requirements while significantly reducing overall system cost, opening the door to a new generation of high-performance, low-cost graphics systems.

In the days of simple Microsoft Windows applications, about the most demanding tasks placed on the frame buffer were some rectangle drawing, a lot of character generation and a few BitBlt operations as a result of fast scrolling of text or dragging of a window across the screen. Today, however, the attention of the PC industry is moving toward high-resolution displays, true color and sophisticated multimedia applications incorporating three-dimensional graphics and full-motion video. With all this video and rendering data passing through the frame buffer along with the text, windows and cursor activity, graphics systems are being brought to their knees because of a bandwidth bottleneck.

Today, with the "sweet spot" in the graphics industry at a display of 1024 by 768 pixels with a 16-bit-per-pixel color depth, just refreshing this display, cycling through the frame buffer to send the screen image to the monitor, requires a memory bandwidth of about 160 megabytes per second (MB/s). 2D graphics and line drawing operations add about another 50 MB/s, 3D graphics operations need about 300 MB/s, and MPEG video requires about 150 MB/s more, for a total bandwidth of about 660 MB/s. Larger displays or more colors increase the bandwidth requirements even further.

Function	Sustained Bandwidth (MB/s)
Screen Refresh	160
2D Graphics Operations	50
3D Graphics Operations	300
MPEG Video	150
Total Bandwidth	660 MB/s

A graphics system designer really has only three alternative to increase frame buffer bandwidth: increase the width of the port, increase the number of ports, or increase the data rate through the port. For the past several years, designers have chosen the first method, but increasing the bus width has already reached its practical limits.

A frame buffer using single-ported EDO DRAM or SDRAM and a 64-bit data path, for example, requires a 105-pin interface between memory and the graphics controller. The large number of pins increases the minimum die size of the controller, which in turn may require a larger, more costly package, especially if additional functionality such as 3D graphics and video are to be integrated in a single controller. The complexity of the printed circuit board layout requires many layers and more board real estate, also increasing system cost. Finally, the large voltage swings on the interface lines can cause electromagnetic interference problems, especially at the higher clock rates used by SDRAMs, requiring greater expertise and time to design the board layout, and extra shielding. Even assuming these to be acceptable tradeoffs, the resulting sustained bandwidth using EDO DRAM or SDRAM is only 200 to 400 MB/s, not enough for multimedia applications.

A dual-ported VRAM design, with its 110-pin memory interface and complex board layout considerations, shares all the spatial disadvantages of a single-ported design, with the further disadvantage of potentially higher controller pin usage. Today, the trend in PC graphics controllers is to integrate the RAMDAC on the controller chip to increase performance and reduce costs. Such a design would require the VRAMs' serial ports to be connected back to the controller to access the internal RAMDAC, requiring even more pins on the controller. While overall bandwidth may appear to be higher, the partitioned display bandwidth and 200-MB/s transfer rates for line drawing suffer under real-world multimedia applications.

Rambus DRAMs take a completely different approach that avoids the traditional performance bottlenecks and paves the way for entirely new system designs. Through a combination of high-speed data transfer and other graphics acceleration innovations to enhance performance, Rambus technology delivers the high-integration and low-cost benefits PC graphics system designers need.

A single-channel, 2-MB Rambus frame buffer implementation using two 8-Mb RDRAMs provides 600 MB/s peak and about 480 MB/s sustained bandwidth, while a dual-channel design can deliver 1200 MB/s peak and 960 MB/s sustained bandwidth, clearly more than enough to support even today's most advanced PC graphics applications. The 31-pin/channel Rambus interface makes it possible to cost-effectively integrate additional functionality onto the controller die with attendant lower packaging costs. The simple two-chip solution requires no glue logic, uses less board real estate and less power, and offers up to a tenfold reduction in electromagnetic interference compared to other DRAM solutions.

3D graphics rendering, Garaud shading and texture mapping as well as MPEG video make heavy use of complex algorithms and other sophisticated mathematical operations to manipulate frame buffer data. The Rambus DRAM component design has built-in features to accelerate these functions. Rambus DRAMs provide the means to execute quick writes within any 8-byte block in memory though random access transfers at the full 600-MB/s bandwidth, faster than any other technology. Used in conjunction with tiling or address mapping of the display memory, this technique substantially accelerates line drawing by reducing latency. Write-per-bit and mask-per-bit operations provide significant advantages when running either 2D or 3D graphics applications by enabling individual bits in the data stream to more easily be manipulated.

More than any other technology, the architectural solution developed by Rambus meets the requirements for creating high-performance, low-cost graphics systems. Only Rambus technology delivers the high bandwidth, low overall system cost and room to integrate multimedia features that today's and tomorrow's PC graphics require.

# # #