Originally Posted by slaWter
And that's why PCIe 3.0 is important. GPUs don't need the additional bandwidth just yet but PCIe SSDs will. With PCIe 3.0 mainboard manufactures can better design their PCIe layouts for GPUs and SSD cards.
Personally, I prefer SATA as my disk interface because I'd like to use my PCIe slots for "more traditional" cards. And the max bandwidth of SATA 3 is more than enough for me. SSDs are about access time for me. I even took apart my G2 Intel RAID to use the disks in two different systems instead. I'm more than happy with something like a SF2000 on a SATA 3 controller. But when it comes to controllers I just prefer Intel's native solutions over some third party onboard controller. For me it's either Intel onboard or otherwise a real HW controller.
Like I said, if you're happy the way it is, that's cool.
I just think that if Intel would still be on their previous release schedule, we'd have much better and more balanced products on the highend front.
Well, PCI-e 3.0's max bandwith is 10 GB/sec, so in theory that makes it have about 20x the maxium mesured performance of the fastest SSD's on the market for the time being....Even PCI-e 2.0 is still 10x faster than any SSD on the market...
If you want the maximum performance from your SSD's, then for me, it would be using a dedicated controler from either LSI or areca or adaptec....Controlers that feature their own dual core CPU's and up to 512 MB of dedicated cache just for the drives themselves, but they start at 300~350$ for baseline models and go from there, so cheap they are not...
Even the upcoming intel 720 series of SLC SSD's, wich will be released by late this year, will use a PCI-e interface rather than the SATA 3.0 interface, so that right there suggests that the latter won't be fast enough to allow that partcular SSD to perform at it's best...
As for GPU's, i'm not too sure about PCI-e 3.0 either, since we're still seeing the same trends as newer video cards are being released, and that's cards that are not only faster at the GPU level, but also pack more memory and that memory is faster clocked, with single GPU cards now aproaching 200 GB/sec of memory bandwith, wich is 20x faster than PCI-e 3.0, and much lower latency since it's right next to the GPU itself, on the Same PCB...So if you want that GPU performing at it's fastest speed, it won't be because of the 10 GB/sec provided by the PCI-e 3.0 bus specification, but the memory on the video card itself for all effects and overall workloads that the GPU processes often in a given game....Developers try to store the data within the video card memory itself as much as possible, and try to minimize PCI-e bus data traffic as much as possible and the same for system memory too..
Even right now, if you add a video card into a PCI-e 8x slot(electrically of course) wich effectively cuts the bandwith down to half, and measure that performance with the same card, on the same game and settings, but this time running on a PCI-e X16 slot, effectively doubling the bandwith to the video card, the performance difference is practically non existant(2~3 FPS), wich is usually considered margin of error values....
At least with Nvidia, if they maintain a 384 bit memory bus with kepler, just like they have up until now with all their higher end releases over the last few years, that means that the card will obviously have faster clocked memory, but might also offer a 3 GB model by default and that's a lot of memory to completely fill up before PCI-e 3.0 actually makes a difference.
I'm sure there will be a lot of reviews comparing PCI-e 2.0 to PCI-e 3.0 when the time comes, but i'm not just buying into it and holding back upgrades, simply because it's a higher number and it should be better in theory....I want to see actual real world performance numbers, and like i said earlier, previous comparisons between PCI-e 1.0 and PCI-e 2.0 didn't show a difference and reviewers went on that might change in the future....It's 2011 and i'm still waiting here