|
nVidia to announce Turbo cache marchitecture |
|
Written by Mavke
|
|
Monday, 04 October 2004 |
nVidia has an answer for ATI's Hypermemory. nVidia is not ready to talk about it publicly, but we already know quite a bit about it. That's because nVidia is already talking with its big customers about it and telling them how great this marchitecture will be. It will use the PCIe bi-directional bus to use system memory RAM as graphics memory.
 nVidia to announce Turbo cache marchitecture
You can use this marchitecture only with PCIe cards since they are bi-directional and their buses can provide enough bandwidth. AGP cards and chips won't benefit from this marchitecture - this ball is in the PCIe court only.
Still, addressing system memory and using it for graphics sounds awfully slow. Even mainstream cards such as the X300 have 6.4 GB/s, faster than the PCIe bus can provide at 16X.
If we are talking about 6800 Ultra you have to face 35.200 GB/s of internal GPU to GDDR3 memory bandwidth versus a thick 4096 MB. That's all you will get between the graphic card through the PCIe bus to system memory.
It has to be something that graphic programmers could actively use, but ATI and nVidia still don't want to spell out the tiny details to the world. I guess that you can store some of the textures in system memory and even 4 GB/s might be enough.
Anything that uses the word "hyper" is, of course, deeply suspicious, whichever vendor employs it.
|
NiBiTor v6.06