pigpen65
June 24th, 2004, 8:43:03 PM
Here's a cool little article about the newest board from ATI. My computer came with Intel's integrated graphics. I added an ATI Radeon 9200 with 128mb dedicated RAM (about $50 on eBay.) You wouldn't think a graphics accelerator would make that big of a difference, but the effects were incredible. Most notably with the speed of pulling up web pages. I can't image what effect these $400-500 boards would have.
http://www.computerpoweruser.com/editorial/article.asp?article=articles/archive/c0407/08c07/08c07.asp&guid=1B4C8DFCE7C748BAA76E1AF617F9CDA1
"ATI’s Next Wave
Radeon X800 XT Platinum Edition & X800 Pro
Last month we toyed with NVIDIA’s NV40 chip in big-daddy “Geforce 6800 Ultra” form, with its massive transistor count and Shader Model 3.0 features. Based on what we saw, it was going to take something rather special for ATI to equal, never mind top, NVIDIA’s efforts. The new R420 architecture’s paper specs look just as exciting but perhaps not quite as forward thinking. Clearly the race is on, as we were surprised at just how potent the review units we received from ATI were when it came to the benchmarks.
First off is the high-end Radeon X800 XT Platinum Edition, which sports 160 million transistors—some 50 million more than the previous-generation R360 chip. Based on a low-k dielectric 0.13-micron TSMC FAB process, the 500MHz core clock speed is more aggressive than the 6800 Ultra’s. The GDDR3 memory, which still uses a 256-bit interface, clocks in at 525MHz x 2 (effectively 1.12GHz), which is in line with NVIDIA’s top draw effort. The X800 XT’s peak fill and texture rates of 8.3 gigapixels per second actually raise the bar even higher than the 6800 Ultra’s 6.4 gigapixels per second, which was the previous best. With 16 pixel pipelines and six vertex pipelines at its disposal, the tale of the tape is markedly similar to NVIDIA’s. The stonking 36.8GBps memory bandwidth is also just ahead of NVIDIA’s and pales in comparison to the previous generation R360’s 23.4GBps.
Architecture. Architecturally there is a difference in the choices the ATI and NVIDIA engineers made. NVIDIA decided to push the boat out and support Shader Model 3.0, whereas ATI stuck with 2.0; this is probably reflected in the smaller number of transistors on the R420. With its 16 x 1 pixel pipeline, it can render up to 16 single-texture pixels in parallel per clock cycle. Just like NV40, R420 actually serves up its pipes in groups of four, with the X800 XT dishing out four of these quads. Putting this in perspective, ATI has exactly doubled the R360’s 8 x 1 pipeline, which was grouped into two quad pipes. Each pixel shader unit can perform up to five (a collection of two three-component vector operations, two scalar ops, and one texture address op that can all work in parallel) floating-point instructions per clock and sports two vector ALUs and a couple of scalar ALUs.
When all 16 pipes are pegged, the theoretical specs look tasty: 80 floating-point operations per clock are possible from the pixel shading units. ATI didn’t just double R3xx’s architecture, but the company dug in deep to refine the number of temporary registers from 12 to 32 and added a facing register, which allows for better efficiency in determining a polygon’s front or rear. This comes in handy for double-sided lighting scenes. The shader instructions limit has been raised all the way up to 1,536 (512 scalar, 512 vector, and 512 texture operations) per pass. Even though R3xx was able to do “only” 160, games such as Half-Life 2 max out with no more than 30 pixel shader instructions per scene. The F-buffer, carried over from R3xx, has been beefed up with a “temporary” staging area where a PSU is able to store/stage a pixel’s value that will be revisited during the heavier multipass rendering instances."
...more...
http://www.computerpoweruser.com/editorial/article.asp?article=articles/archive/c0407/08c07/08c07.asp&guid=1B4C8DFCE7C748BAA76E1AF617F9CDA1
"ATI’s Next Wave
Radeon X800 XT Platinum Edition & X800 Pro
Last month we toyed with NVIDIA’s NV40 chip in big-daddy “Geforce 6800 Ultra” form, with its massive transistor count and Shader Model 3.0 features. Based on what we saw, it was going to take something rather special for ATI to equal, never mind top, NVIDIA’s efforts. The new R420 architecture’s paper specs look just as exciting but perhaps not quite as forward thinking. Clearly the race is on, as we were surprised at just how potent the review units we received from ATI were when it came to the benchmarks.
First off is the high-end Radeon X800 XT Platinum Edition, which sports 160 million transistors—some 50 million more than the previous-generation R360 chip. Based on a low-k dielectric 0.13-micron TSMC FAB process, the 500MHz core clock speed is more aggressive than the 6800 Ultra’s. The GDDR3 memory, which still uses a 256-bit interface, clocks in at 525MHz x 2 (effectively 1.12GHz), which is in line with NVIDIA’s top draw effort. The X800 XT’s peak fill and texture rates of 8.3 gigapixels per second actually raise the bar even higher than the 6800 Ultra’s 6.4 gigapixels per second, which was the previous best. With 16 pixel pipelines and six vertex pipelines at its disposal, the tale of the tape is markedly similar to NVIDIA’s. The stonking 36.8GBps memory bandwidth is also just ahead of NVIDIA’s and pales in comparison to the previous generation R360’s 23.4GBps.
Architecture. Architecturally there is a difference in the choices the ATI and NVIDIA engineers made. NVIDIA decided to push the boat out and support Shader Model 3.0, whereas ATI stuck with 2.0; this is probably reflected in the smaller number of transistors on the R420. With its 16 x 1 pixel pipeline, it can render up to 16 single-texture pixels in parallel per clock cycle. Just like NV40, R420 actually serves up its pipes in groups of four, with the X800 XT dishing out four of these quads. Putting this in perspective, ATI has exactly doubled the R360’s 8 x 1 pipeline, which was grouped into two quad pipes. Each pixel shader unit can perform up to five (a collection of two three-component vector operations, two scalar ops, and one texture address op that can all work in parallel) floating-point instructions per clock and sports two vector ALUs and a couple of scalar ALUs.
When all 16 pipes are pegged, the theoretical specs look tasty: 80 floating-point operations per clock are possible from the pixel shading units. ATI didn’t just double R3xx’s architecture, but the company dug in deep to refine the number of temporary registers from 12 to 32 and added a facing register, which allows for better efficiency in determining a polygon’s front or rear. This comes in handy for double-sided lighting scenes. The shader instructions limit has been raised all the way up to 1,536 (512 scalar, 512 vector, and 512 texture operations) per pass. Even though R3xx was able to do “only” 160, games such as Half-Life 2 max out with no more than 30 pixel shader instructions per scene. The F-buffer, carried over from R3xx, has been beefed up with a “temporary” staging area where a PSU is able to store/stage a pixel’s value that will be revisited during the heavier multipass rendering instances."
...more...