so pcie 3.0 16x has a memory bandwidth of 16GB/s, and modern high end gpus have bandwidths of 334 like the titan x, so isnt pcie 3 16x, and even the new 4 at 16x 32, bottlenecking basically all modern day, and past day gpus? i mean its being saturated by like 20 times over, and shouldnt the storage also be a consideration as thats where the information to draw what to do such as lighting effects, textures, etc be considered a bottleneck as well?
Nope it doesnt bottleneck at all.
PCIe Bandwidth and memory bandwidth are different things used for different things.
The PCIe bus is used to move data from the to the GPU either for calculations or just moving files from the hard drive for game loading.
The Memory bus on the other hand is used constantly by large texture files and game engine code. This the CPU doesn't really need to touch. And as a result memory bandwidth has to much significantly larger.
PCIe 3.0 is not a bottleneck has been proven time and time again to not be. Memory bandwidth on the other hand is becoming much more needed as game textures and game resolutions are increasing dramatically.
HBM (High bandwidth memory) by AMD Is a step forward to increasing memory bandwidth.
1 Like
Haha you beat me on this one. ☺
But This basicly. ^^
2 Likes
http://www.tomshardware.com/answers/id-2249473/calculate-pci-express-usage.html
But performance hinges on a lot of other factors for a gpu, temps, power "cleanliness", driver optimization. But the general consensus is that a gpu won't be bottle-necked for more than a frame even in pci express 1.0 @ x16.
My limited experience is jumping from gtx 760's in a pci ex 2.0 amd motherboard running both at x8 to an intel motherboard running both at 3.0 x8, i actually gained on average 5 fps in bf4 on ultra @ 1080p, and it was probably just drivers.
I'll throw this out there so I can get updated info if things have changed, but last time i checked, PCI x16 running as a x8 isn't even enough to bottleneck. For example, the mobo i have was made to be able to do 3 way SLI with x16, x16, x8. You could say the x8 doesn't matter at that point due to the SLI strip, but I'm unsure of how that actually works. Anyway, as is said above PCI bandwidth and memory bandwidth are different and if what I know (knew?) still holds true, just x8 is still enough.
The only "real bottleneck" that might occure, if you put 2 highend gpu´s in SLi,
and you play on 1080p for example.
Is a cpu that cannot keep up with it.
Even a titan-x can only saturate PCIE gen3 x8 by a small margin. It's got a lot further to go to saturate x16.
but the texture files, and game engine code travels over pcie.
On initial Loading, Otherwise they're in Vram the whole time
PCI-E sees the most usage when you run out of vram, because then it constantly has to send files over it constantly.
If the OP was true, there would be no way that my 1X PCIe 2.0 eGPU would ever work. It does somehow, though with a crapload of screen tearing. Trigger warning for those FreeSync users among us :P