Radeon X1950 XTX
This problem became even more urgent for ATI after the arrival of the GeForce 7950 GX2 (for details see our article called Two for One: Nvidia's Dual-Chip GeForce 7950 GX2 Reviewed) because the company lost its competitive edge in the sector of top-end single-card graphics solutions. The fact that the GeForce 7950 GX2 is actually a SLI tandem didn’t make life easier for ATI. They had to find a way to improve the performance of top-end models in the Radeon X1000 series. ATI didn’t have time to equip its R580 chip with additional TMUs notwithstanding the flexibility of the Radeon X1000 architecture which allows making a chip from ready-made functional units as if from Lego pieces. This approach might have been fruitful, but would have also meant projecting a new chip almost from scratch. Although this could have been made faster than with a classic graphics architecture, it didn’t make sense in view of the imminent arrival of the next generation of GPUs. But where to find new resources then?
Actually, such resources were built into the Radeon X1000 architecture originally. We mean the memory controller. As we said above, using high resolutions along with FSAA raises the requirements bar not only in terms of texturing speed but also in terms of graphics memory bandwidth. So if there is no way to improve the former parameter, the other may be tinkered with. That’s exactly what they did to create the Radeon X1950 XTX!
The memory controller in the Radeon X1000 series had originally been designed to support GDDR4, so ATI decided to use this new memory type on the new card as having a much higher frequency potential than GDDR3 while having comparable electrical and thermal characteristics. ATI Technologies played a key role in the development of GDDR4, so the company had had a chance to try the new memory under real conditions before the release of graphics cards on next-generation processors where GDDR4 is expected to be used.
Mass production of the new memory was launched by Samsung in June of this year. The slowest chip in the GDDR4 series has an access time of 0.91 nanoseconds and is capable of working at 1100 (2200) MHz frequency whereas the fastest GDDR3 chip is only rated for 900 (1800) MHz. If clocked at the same frequency of 1000 (2000) MHz, a GDDR4 chip would consume 45% less power than a GDDR3 chip. It means that the sky-high operating frequency of the new memory type does not lead to higher heat dissipation and does not make some special cooling systems necessary.
After BIOS-based reprogramming of the memory controller the R580 chip becomes R580+ and in that capacity is installed on a new family of graphics cards called Radeon X1950. The minor addition to the chip name and to the graphics card model name looks reasonable considering that the product characteristics have changed so little.