Everywhere you look, stunning digital imagery is evoking a great deal of adrenaline-surging excitement. It also helps to sell products, whether they’re thin-film-transistor LCDs, new GUI operating systems (like Microsoft Vista), or high-end PCs.
While new display and graphics processing unit (GPU) technologies get the glitz, the brick and mortar supporting these images is high-speed graphics memory. Today’s high-end memory is producing images with exceptional depth, clarity and animation as the graphics memory market advances in frequency, bus width, density, and voltage.
Today, 1.4- to 2.4-Gbit/s GDDR3 and GDDR4 memories serve the high end of the graphics market, while 700- to 800-Mbit/s DDR2 services the value segment. This results in GDDR3 and GDDR4 offering raw performance that’s superior by a factor of 2.5 to 3.
In addition to the marked difference in frequency, GDDR3 and GDDR4 memories have the widest bus width (x32) per device, whereas the value segment memory (DDR2) is half of that (x16). This means that at the same clock frequency, DDR2 would transfer only half as much data as GDDR3 or GDDR4.
The overall combination of frequency and bus width results in high-end GDDR3 and GDDR4 performing at a level approximately five to six times that of lower-performing DDR2 memory.
Typically, current high-end graphics cards ship with a total of 512 Mbytes of graphics memory, while the value segment uses 128 to 256 Mbytes. These total densities are best achieved by combining 256- or 512-Mbit component densities.
Until the last few years, supply voltage wasn’t a factor in graphics memory selection. Everywhere, 2.5 V was the norm. But as total frame buffer sizes and frequencies increase, thermal and power budget concerns have quickly brought about the need for lower voltage. Today’s 1.8-V supply voltages will see a progression to 1.5 V and eventually 1.35 V over the next two years.
WHO NEEDS IT?
These advances have made a huge impact. Today’s visually compelling, totally immersive video games would not be possible without the availability of extremely high-speed GDDR3 and GDDR4.
Even in the mid-range and value PC segments, there’s industry-wide recognition that discrete memory (DDR2) will make Microsoft Vista Premium run more efficiently from a graphics perspective, providing more headroom than PCs with integrated graphics.
XDR DRAM is another high-speed memory technology that’s been adopted in some game console and consumer applications. It was developed three years ago as the memory with the highest bandwidth per pin.
If a designer wants to save on the pin count of a controller, XDR DRAM could fit that system well. On the other hand, if the number of pins isn’t a limitation for highest overall bandwidth without ballooning the total size of the memory, GDDR4 is now the best choice (see the table).
While this discussion has focused on high-performance and value graphics segments, the fact is that graphics solutions are tiered into at least six or seven segments, each offering a different combination of performance, technology, and density for different customers, much like the overall PC market. At any given time, most, if not all, of the above solutions likely will be used in the market. (Reportedly, GPU manufacturers have yet to adopt XDR DRAM.)
THE FUTURE LOOKS BRIGHT
GDDR5 technology will be introduced in 2008, but it probably won’t reach the “sweet spot” for high-end graphics technology on a volume scale until 2009. This technology targets 5.0-Gbit/s data rates, surpassing all current technologies in performance. In addition to OEMs embracing it for premium PCs, there’s a high likelihood that next-generation game consoles will adopt GDDR5.
When GDDR5 technology is introduced for the high end, GDDR3 and GDDR4 devices will “waterfall” down into the mid-range performance graphics segments and continue to be used beyond 2009.
Mueez Deen is the director of graphics memory for Samsung Semiconductor Inc. He is responsible for mobile memory (low-power DRAM, NOR Flash, SRAM and MCP), graphics memory, and consumer DRAM product lines at Samsung Semiconductor. He has more than 25 years of experience in the semiconductor industry, with the past 18 years at Samsung and the balance in design engineering at Advanced Micro Devices.