As we've seen time and time again with low-end boards that pack the memory on, but which use asthmatic memory bus widths to talk to it, there's little point in giving a bandwidth-starved graphics chip more memory which it can't realistically use to accelerate 3D performance. Here, on-card caching schemes work best to make the most of what they've got, especially if you want to use the board for something other than 3D graphics and games. While you might not have the bandwidth on the board to feed the GPU very well (often less than 5GiB/sec on 64-bit memory bus low-end hardware) to render complex scenes, that doesn't mean you have to leave blocks of memory empty. You still want to upload data to the board even if it won't be processed for a few frames. Speculative preloading on 512MiB low-end hardware would make some sense, if the driver is able.

However, the almost overriding factor for low-end hardware is its inability to do much with huge on-card framebuffers. With the rise of PCI Express, here's hoping that extra costs to low-end hardware aren't seen in the push to give the masses a fatter framebuffer, and schemes like HyperMemory and TurboCache are used instead. There's currently a serious false economy to purchasing a large memory low-end board, and that's not going to get any better without investment in driver advancement by the IHVs with 512MiB (or indeed 256MiB) boards.

http://www.hexus.net/content/reviews...lld19JRD0xMTYw