Solved.

[ybriK]

Solved.

[Glenwing]

Memory communication doesn't go over the PCI Express bus.  The memory chips are on the graphics card, and the GPU communicates directly with them across the graphics card's PCB.

[brownninja97]

No bloody clue but what confuses me is that if you are using a titan on pci-e gen 2 then you loose only like 1fps.

[Tech_Dreamer]

First of all, that's the longest thread title i've ever seen ,

 

PCIE 3 has 32 GB/s bandwidth i presume, and yes , primary offload is to ram & Then the on to the GPU memory,

[Glenwing]

First of all, that's the longest thread title i've ever seen ,

 

PCIE 3 has 32 GB/s bandwidth i presume, and yes , primary offload is to ram & Then the on to the GPU memory,

 

Roughly 16GB/s actually, for a gen3 x16 link.

[D3MON]

because even the highest end gpus today dont even max out pcie 2.0 x16 (8gb/s) even when using 2 cards in 8x/8x (pcie 2.0)

and the gpu core needs high memory bandwidth (through gddr5 and a wide bus i.e 256bit) to the gpu VRAM buffer, pcie bandwidth is to the system itself which is less of a concern

[Tech_Dreamer]

Roughly 16GB/s actually, for a gen3 x16 link.

Wow!Really? they do boast out on their ads, I'm interested in Nvidia's stacked memory release, for volta they're claiming upto 1 tb of bandwidth, don't know how they're going to perform on real world over PCIE 4, maybe a 4.5 release has a higher bandwidth to accommodate that kind of speed.

[Glenwing]

Wow!Really? they do boast out on their ads, I'm interested in Nvidia's stacked memory release, for volta they're claiming upto 1 tb of bandwidth, don't know how they're going to perform on real world over PCIE 4, maybe a 4.5 release has a higher bandwidth to accommodate that kind of speed.

 

Memory bandwidth has nothing to do with the PCI Express bus.  The GPU has the PCI Express bus to communicate with the CPU and system memory if necessary, and it has multiple memory busses to communicate with it's own onboard memory.  The onboard memory does not communicate through the PCI Express bus.

[Briggsy]

you will often hear people refer to video memory as a frame buffer. Lets say you have 60 frames @ 2.1 Megapixels being displayed every second, you need some crazy bandwidth to shift that data around effectively. the pci-e bus isn't transferring frames around, luckily.

[Tech_Dreamer]

Memory bandwidth has nothing to do with the PCI Express bus.  The GPU has the PCI Express bus to communicate with the CPU and system memory if necessary, and it has multiple memory busses to communicate with it's own onboard memory.  The onboard memory does not communicate through the PCI Express bus.

Ahh.. Thanks for clearing that up.. *Hats off to you*

[ybriK]

Memory communication doesn't go over the PCI Express bus.  The memory chips are on the graphics card, and the GPU communicates directly with them across the graphics card's PCB.

 

 

because even the highest end gpus today dont even max out pcie 2.0 x16 (8gb/s) even when using 2 cards in 8x/8x (pcie 2.0)

and the gpu core needs high memory bandwidth (through gddr5 and a wide bus i.e 256bit) to the gpu VRAM buffer, pcie bandwidth is to the system itself which is less of a concern

 

 

Memory bandwidth has nothing to do with the PCI Express bus.  The GPU has the PCI Express bus to communicate with the CPU and system memory if necessary, and it has multiple memory busses to communicate with it's own onboard memory.  The onboard memory does not communicate through the PCI Express bus.

 

Oh that makes sense now, and wow that was obvious why didn't I think of that fully. Then what does the PCIe exactly communicate (I know some of it would be the drivers/configurations for voltage, fan speed, and core/mem speed)? I've been hearing the same thing "we haven't even fully saturated a PCIe x16 3.0" but I've never grasped as to what is exactly being communicated over the bus?

[Glenwing]

Oh that makes sense now, and wow that was obvious why didn't I think of that fully. Then what does the PCIe exactly communicate? I've been hearing the same thing "we haven't even fully saturated a PCIe x16 3.0" but I've never grasped as to what is exactly being communicated over the bus?

 

Communication between the graphics card and the CPU, for instructions and whatnot.

[ybriK]

Communication between the graphics card and the CPU, for instructions and whatnot.

Then how come manufacturer's/ the company responsible for PCIe are making these busses that have really high trasnfer rates when us consumers aren't able to fully saturate them? Is it because of "future proofing" or for high-end business performance servers?

[Glenwing]

Then how come manufacturer's/ the company responsible for PCIe are making these busses that have really high trasnfer rates when us consumers aren't able to fully saturate them? Is it because of "future proofing" or for high-end performance servers?

 

Because people will pay more money for it.  Also, graphics cards are not the only thing that uses PCI Express, and the bandwidth usage depends on the application.  For receiving instructions for frame rendering and everything else associated with gaming, the bandwidth usage is relatively little, but if you use a GPU for computation then you will use a lot more bandwidth because there is a lot more data that needs to be communicated back and forth.

[ybriK]

Because people will pay more money for it.  Also, graphics cards are not the only thing that uses PCI Express, and the bandwidth usage depends on the application.  For receiving instructions for frame rendering and everything else associated with gaming, the bandwidth usage is relatively little, but if you use a GPU for computation then you will use a lot more bandwidth because there is a lot more data that needs to be communicated back and forth.

Ah I see. So that means PCIe SSDs use the bus lane to communicate information and that's why it is able to perform significantly faster than its counterparts SATA/IDE?

[Glenwing]

Ah I see. So that means PCIe SSDs use the bus lane to communicate information and that's why it is able to perform significantly faster than its counterparts SATA/IDE?

 

Yes

[ybriK]

Yes

Ah thanks a lot and to everyone else, as it cleared up some confusion I had.