Originally Posted by Sgt Bilko
Originally Posted by Red1776
Thanks for the support guys. This is not going to be like any review you have seen before.
really looking forward to seeing how the 8350 and PCIe 2.0 hold up with 4 cards
Mathematically, x8/x8/x8/x8 should be able to sustain four R9-290Xs up to 4k resolution 32-bit color at 60FPS before starving to death. HT will need to be overclocked significantly.
At 1600p or lower, the risk should be minimal and may even allow for 90-100FPS, though due to the extra data sent to the GPUs scaling in such a way would not be optimal.
HRes * VRes * CBit * FPS * (Cards - 1)
8 * 1024 * 1024 * Cards
^ Amount of bandwidth the lead GPU will need to have available for frame transfer.
At 4k 32-bit 60FPS, the lead GPU will need to be able to receive 1,424 Megabytes of data per second from the other three GPUs. At x8, this leaves it with only 2,672 MB/s of bandwidth for frames form the CPU and bus overhead. It then has to process the data from the CPU and get it out the door in time to be used.
Each of the other three GPUs only needs 475MB/s of bandwidth to transfer to the lead GPU. all of those are fine. However, 2.0 x4 is considered a no-no for GPU usage, and that is cutting it very very close giving the lead GPU only an x5 to deal with.
At a lower resolution such as 1440p, the bandwidth needed for transfers from the other GPUs is cut in half, giving the card much more headroom (~6.5 lanes). However, if the FPS is increased over 60, both the CPU -> GPU and GPU -> GPU bandwidth scales linearly, quickly cutting off remaining supply before it can double.
This basically limits the situation where you would want Quadfire 290Xs in the first place (4k and above) by the PCI-e buss simply not being able to do it. Trifire at x16/x8/x8 would actually be much better suited, or even using a Dual GPU card (not that they exist) to conserve some precious bandwidth.
My two cents. Be very very careful, because the lack of crossfire connector is actually going to hurt here. That extra .9GB/s would have tipped the scales. There is a reason AMD recommends 2.0 x16 and 3.0 x8.