Nvidia Turbo Cache

[zoomy942]

you know something? in this round or graphics releases, i have to give nvidia credit for coming up with idea after idea that are full of inovation. being unorthodox is a powerful weapon

http://theinquirer.net/?article=20110

[msxyz]

Hardly original. The idea behind AGP was that the onboard memory would have served only as a texture cache and for the framebuffer while the card could allocate as much system memory as needed to store all the data needed for rendering.

The only GPUs that used this approach were Intel's own 740 chip and S3 Savage. Both those chips were outperformed by the competitors and S3 eventually abandoned the idea of AGP texturing with the next Savage4 chip (though they kept the 64 bit bus which killed the performance of this otherwise fine GPU). Intel 740 found its way inside the northbridge of many Intel chipset as a low cost integrated graphic solution.

While PCIe offers more bandwitdh than AGP and the ability to both read and write from system memory (AGP was sort of asymmetrical) it still does not provide the necessary bandwidth of a today mid/high range card. Not to mention that sharing the memory with other devices (and with the microprocessor) cause a general slowdown of the system.

[zoomy942]

i know it doesnt help performance, but having a Dedicated GPU using the system memory and not some integrated stuff, is better for people who have the occasional 3d gameplay, and it keeps cost down.

so i agree that for uber-gamers, like us, its not so great, but for people that dont have much $$$ or dont play often, its a neat approach to an old idea.

[jolle]

you know something? in this round or graphics releases, i have to give nvidia credit for coming up with idea after idea that are full of inovation. being unorthodox is a powerful weapon

http://theinquirer.net/?article=20110

Acctually ATI has the same thing (seemingly) called HyperMemory.
Im sorta hoping it resembles 3dLabs Virtual Memory more then ye ole AGP texturing..
If so it could perhaps help highend cards too, and allow developers to make use of more textures then onboard RAM amounts would normally allow.
I guess that depends on how well managed it is, System RAM is ofcource not a good replacement for Onboard RAM with the speed difference and added latancy.
The basic idea behind VM is that Onboard RAM is a Cache, and System is sortof the "main memory" ( i think), the larger cache you have the better.
Its similar to BitBoys prototype of a GPU with 12Mb Embedded RAM in the GPU on a 1024bit wide bus, which could be completed with regular DDR, and it would keep the most intensive things in the Embedded RAM in a similar fashion.
And VM was present on Pro cards on AGP, which doesnt necessarly mean it would work well for gaming, but Carmack and Sweeny was pushing for it on gaming cards if i remember right so who knows, and on PCI-E which offers more bandwidth and fullduplex it would make more sense..

Dont know tho, but one can always hope.

At any rate it seems like it will be a spec of WFG and well needed to allow budgetcards to handle "Avalon" in Longhorn, as it seems to put some requirements on the amount of onboard RAM, since it draws things on the desktop as textures (unless im way off here)..
So a 128mb Card could start using System RAM when needed via PCI-E buss.