ATI & Nvidia next-gen parts coming out next year?
Posted: Fri Mar 08, 2002 10:38 pm
http://www.penstarsys.com/editor/SO3D/SO3D.htm The article is long so the relevant info has been posted:
"There are two upcoming parts from ATI that should cause quite a stir. The R250 is the successor to the Radeon 8500, and should offer greater speed and a few more features. While the current Radeon 8500 is produced on a .15 micron process, the R250 will be produced on a full .13 micron process. This will allow the R250 to be clocked at around 350 MHz. The new BGA form factor DDR memory will also allow memory clock speeds to approach 350 MHz. This one-two punch will give overall performance a shot in the arm, and the tweaked R250 core will also help in the performance area. So far there are no plans for this part to be 100% DX9 compliant, as DX9 does not yet have its feature set fully specified. The R250 will stay DX 8.1 compliant, but its speed will be greatly increased. Staying with the same basic core will also allow for better drivers upon release, which is absolutely key when competing with NVIDIA and thier new GeForce 4 Ti series of chips. The tweaks to the R250 will also allow for better Anti-Aliasing performance and support, but as of now the AA solution is not as standalone as is NVIDIA's Accuview technology. This product is expected in late Summer 2002.
The next generation part out of ATI is known only as the R300. This is the chip that ATI is designing around the known DX9 specifications, and ATI is actively trying to get Microsoft to agree with their interpretation of what DX9 should enable. Work is continuing on this chip, and it will be a slight departure from the R200 and R250 chips. One of the things that DX9 promises is easier utilization of advanced Pixel and Vertex functions. The current Radeon 8500 chip posses quite a challenge to adequately program for because of the way the Pixel and Vertex Shaders are set up (and also how DX 8.1 handles these calls). To make a fully DX 8.1 compliant game is quite a chore, which is why there are only a handful of games out there that can truly make use of today's Pixel and Vertex Shaders. The jump from DX 8.1 to DX 9 can almost be compared to the same jump from DX 5.1 to DX 6 in terms of ease of use and advanced features. The R300 does promise more programmability and features, as well as ease of use over previous generations. There have been rumors that the R300 will have 6 pixel pipelines, but this is most likely false. Fillrate is no longer a problem for today's accelerators, but advanced shading is the bottleneck. ATI will most likely stay with the 4 pixel pipelines, but more pixel functions per pass will be added to each pipeline. The R300 will also continue to have the dual Vertex Shaders, but again the functionality of these Shaders will be increased by a significant margin. Anti-Aliasing will also be more thoroughly addressed with this part, and a much more advanced AA unit will be included. Many are suspecting that this will be a stocastic type of AA (random pattern) that will utilize a 5x5 matrix. The AA results produced by this type of setup are much more advanced than any current AA function, and the visual effect will be impressive. This product will most likely hit the market in Spring 2003."
and
"Six months from now NVIDIA will release a refresh of its GeForce 4 line. The Ti series will get a speed boost to around 350 MHz as well as feature higher clocked memories. The overall architecture will get polished a bit, but no major changes will be seen. It is thought that the .13 micron line from TSMC and UMC will be online, and there will be .10 micron samples available at that time. Most likely the GF4 Ti and MX will be transferred to the .13 micron line, hence the increase in clockspeed. More features hidden in the GF4 architecture will also be revealed through software releases, so it will stay very competitive in the face of products such as the ATI R250.
Next year will be a very different proposition for NVIDIA. We will finally see a product that is a fusion of NVIDIA and 3dfx/GigaPixel technology, and the combination will be very impressive. Quite a few of the features of Rampage will be present in the rendering portion of this product, including support for 52 bit (and higher) color, a more advanced version of the "Texturing Computer" that will allow 8+ texturing functions to be applied to a pixel per pass, with 16 functions applied when performing multi-pass (currently the GF4 Ti can do 4 functions per pass/8 functions in multi-pass and the Radeon 8500 can do 6 functions per pass with 12 functions per multi-pass). This is also the reason why internal color rendering will be increased from the current 32 bits to 64 bits (though 52 bit color will be available, and most likely be used extensively due to bandwidth reasons). The more passes and functions applied to a pixel, the more accuracy is needed so that the output will not appear "washed out". This texturing computer can also dynamically allocate resources to the different pixel pipelines depending on the need. The Vertex Shaders will also be revamped, gaining more registers than before to allow more floating point functions to be integrated. Currently NVIDIA is designing this part around what it thinks DirectX 9.0 should be. Whether it all pans out is a question for the future. Needless to say, the NV-30 part will be much more flexible, powerful, and programmable than any other part in NVIDIA's history. This will also most likely feature a more random sample AA part with greater performance than the current Accuview technology from NVIDIA. This chip will weigh in around 100 million transistors and (depending on TSMC) show up using the .10 micron die process. Initial clock speeds will be between 350 MHz and 400 MHz."
"There are two upcoming parts from ATI that should cause quite a stir. The R250 is the successor to the Radeon 8500, and should offer greater speed and a few more features. While the current Radeon 8500 is produced on a .15 micron process, the R250 will be produced on a full .13 micron process. This will allow the R250 to be clocked at around 350 MHz. The new BGA form factor DDR memory will also allow memory clock speeds to approach 350 MHz. This one-two punch will give overall performance a shot in the arm, and the tweaked R250 core will also help in the performance area. So far there are no plans for this part to be 100% DX9 compliant, as DX9 does not yet have its feature set fully specified. The R250 will stay DX 8.1 compliant, but its speed will be greatly increased. Staying with the same basic core will also allow for better drivers upon release, which is absolutely key when competing with NVIDIA and thier new GeForce 4 Ti series of chips. The tweaks to the R250 will also allow for better Anti-Aliasing performance and support, but as of now the AA solution is not as standalone as is NVIDIA's Accuview technology. This product is expected in late Summer 2002.
The next generation part out of ATI is known only as the R300. This is the chip that ATI is designing around the known DX9 specifications, and ATI is actively trying to get Microsoft to agree with their interpretation of what DX9 should enable. Work is continuing on this chip, and it will be a slight departure from the R200 and R250 chips. One of the things that DX9 promises is easier utilization of advanced Pixel and Vertex functions. The current Radeon 8500 chip posses quite a challenge to adequately program for because of the way the Pixel and Vertex Shaders are set up (and also how DX 8.1 handles these calls). To make a fully DX 8.1 compliant game is quite a chore, which is why there are only a handful of games out there that can truly make use of today's Pixel and Vertex Shaders. The jump from DX 8.1 to DX 9 can almost be compared to the same jump from DX 5.1 to DX 6 in terms of ease of use and advanced features. The R300 does promise more programmability and features, as well as ease of use over previous generations. There have been rumors that the R300 will have 6 pixel pipelines, but this is most likely false. Fillrate is no longer a problem for today's accelerators, but advanced shading is the bottleneck. ATI will most likely stay with the 4 pixel pipelines, but more pixel functions per pass will be added to each pipeline. The R300 will also continue to have the dual Vertex Shaders, but again the functionality of these Shaders will be increased by a significant margin. Anti-Aliasing will also be more thoroughly addressed with this part, and a much more advanced AA unit will be included. Many are suspecting that this will be a stocastic type of AA (random pattern) that will utilize a 5x5 matrix. The AA results produced by this type of setup are much more advanced than any current AA function, and the visual effect will be impressive. This product will most likely hit the market in Spring 2003."
and
"Six months from now NVIDIA will release a refresh of its GeForce 4 line. The Ti series will get a speed boost to around 350 MHz as well as feature higher clocked memories. The overall architecture will get polished a bit, but no major changes will be seen. It is thought that the .13 micron line from TSMC and UMC will be online, and there will be .10 micron samples available at that time. Most likely the GF4 Ti and MX will be transferred to the .13 micron line, hence the increase in clockspeed. More features hidden in the GF4 architecture will also be revealed through software releases, so it will stay very competitive in the face of products such as the ATI R250.
Next year will be a very different proposition for NVIDIA. We will finally see a product that is a fusion of NVIDIA and 3dfx/GigaPixel technology, and the combination will be very impressive. Quite a few of the features of Rampage will be present in the rendering portion of this product, including support for 52 bit (and higher) color, a more advanced version of the "Texturing Computer" that will allow 8+ texturing functions to be applied to a pixel per pass, with 16 functions applied when performing multi-pass (currently the GF4 Ti can do 4 functions per pass/8 functions in multi-pass and the Radeon 8500 can do 6 functions per pass with 12 functions per multi-pass). This is also the reason why internal color rendering will be increased from the current 32 bits to 64 bits (though 52 bit color will be available, and most likely be used extensively due to bandwidth reasons). The more passes and functions applied to a pixel, the more accuracy is needed so that the output will not appear "washed out". This texturing computer can also dynamically allocate resources to the different pixel pipelines depending on the need. The Vertex Shaders will also be revamped, gaining more registers than before to allow more floating point functions to be integrated. Currently NVIDIA is designing this part around what it thinks DirectX 9.0 should be. Whether it all pans out is a question for the future. Needless to say, the NV-30 part will be much more flexible, powerful, and programmable than any other part in NVIDIA's history. This will also most likely feature a more random sample AA part with greater performance than the current Accuview technology from NVIDIA. This chip will weigh in around 100 million transistors and (depending on TSMC) show up using the .10 micron die process. Initial clock speeds will be between 350 MHz and 400 MHz."
