View Single Post
  #7 (permalink)  
Old 31st December, 2004, 12:07 AM
Aedan Aedan is offline
Chief Systems Administrator
Join Date: September 2001
Location: Europe
Posts: 13,075

Originally Posted by cadaveca
I'm hoping that we will see more "hypermemory"....when the vidcards use system memory(kinda like ATI's SMARTGART, but way more advanced)....will lead to less latency, and way better performance.
That'd drop performance through the floor!! The system memory bus is contended - that's just a fancy word for the fact that several devices will want to use it at the same time. If the graphics card is using the system memory, then no other device can be using it. If the CPU needs to access system memory, it's locked out until the GPU has finished!

Now add to that the fact that graphics cards have memory that handles far higher bandwidths than system ram does, and the picture looks bleak. Take a card like the 6800 Ultra. It has around 35000Mbyte/sec of memory bandwidth. Take an Athlon 64, which can manage some 5510Mbyte/sec of memory bandwidth. Couple that with the fact that the processor can no longer cache the system memory that the graphics card is using, and you also lose any benefits of the cache onboard the CPU. This is because the CPU doesn't know if the GPU has changed memory contents or not, and can only assume that the GPU has altered memory.
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).
Reply With Quote