by Anton Shilov
10/07/2002 | 06:50 AM
Graphics chips and cards have definitely been the fastest developing computer devices in the last four years. It is quite strange, but now it appears that the high-end graphics cards and systems are going to be the most advanced devices in personal computers. Graphics chips are usually more complex compared to modern microprocessors, they utilise the fastest memory available and they are still not enough to show their whole potential. A week ago I told you that both ATI Technologies and NVIDIA want JEDEC to approve DDR-II standard for graphics applications (see this news-story), today I read over EE Times and EBN that ATI Technologies and its arch-rival want to push the so-called GDDR III (G in this case stands for “Graphics”) already in 2003 because they need it for their upcoming graphics chips.
It may take JEDEC several years to fully approve one or another memory standard. Since graphics processors are extremely bandwidth hungry, developers of the GPUs cannot wait so long. As a result, they need to influence the market and memory manufacturers so that the latter introduced even faster memory chips. In case such devices do not meet certain standards due to several reasons, they may propose their own specifications and make use of such chips. I believe it is quite logical for them, since memory types that are pushed by JEDEC are mostly intended for system memory of personal computers and working over a bus at much lower frequencies than those offered by modern graphics cards’ memory sub-systems. All in all, the demands of graphics cards and system platforms for memory types are very different nowadays.<%BANNER[article]%>
According to certain industry insiders, DDR-II on graphics cards is going to appear in early 2003 and work at the resulting speeds of about 1GHz. The GDDR-III will come later in 2003, probably in the fourth quarter or in the very end of the year, and will be able to function at clock-speeds of up to 1.5 – 1.6GHz. Although GDDR-III will evolve from DDR-II, it will have some pretty important differences, as stated by the sources. Firstly, GDDR-III will make use of a single-ended, unidirectional strobe that separates the reads and writes. DDR-II, by contrast, uses differential bi-directional strobes. Secondly, GDDR-III will utilise a "pseudo-open drain" interface technique that is based on voltage rather than current. This was done so that graphics chips can be compatible with DDR, DDR-II and GDDR-III. Like DDR-II, GDDR-III interface uses 1.8-Volt SSTL. Such memory is better suited to point-to-point links used on graphics cards and will allow the GPU developers to reach the new performance and feature heights with their products.
Apart from ATI Technologies and NVIDIA, GDDR-III is supported by leading memory manufacturers, including Micron, Hynix and Infineon. Well, it seems that the NV40 and R400 based graphics cards will be equipped with brand-new types of memory. Amazing how two fabless semiconductor companies can drive the industry.