PCIe 4 gen details: Double the bandwitdth

[Syntaxvgm]

PCIe 4.0 will have a base speed of 16 Gbps per data link. Graphics cards, which are typically among the most data-hungry components in a machine, can use up to 16 of these links for a total throughput of 64 GB/s vs. today's maximum of 32.

I'm surprised no one posted this yet. 

While at first glance, it may seem not really important seeing as we are only now hitting the limitations of PCIe 2.0 for GPUs, but thinking of 1x and 4x makes me giddy. 

Imagine mini PCIe 4.0 eGPUs?  :lol:  :lol:  :lol:  :lol:

I think the day is coming when eGPUs will no longer be majorly throttled. 

Scratch the fact that MXM 3.0 and PCIe are electronically compatible and it is the laptops refusing to recognize anything not registered in the Bios in order to prevent upgrading overpriced mobile GPUs past a certian point preventing gaming laptops from having desktop GPUs using MXM.  :angry:

I'n storage terms, this means 1x will be capable of ~1970 MB/s transfers verses ~985 MB/s now. (Practical numbers, not theoretical)

 

With all the delays in 3.0 though, I wonder if 4.0 will suffer the same fate? 

Article: http://www.tomshardware.com/news/pcie-generation-4.0-bandwidth,27000.html

[Askew]

It's doubled each time damn.

[wpirobotbuilder]

What's more amazing is that it's still done with generic copper traces.

 

So, GPU makers, can we please see cards that can be configured to only use eight or four lanes? Even PCI-E 3.0 doesn't offer much benefit for x8 over x4, and being able to use available lanes for other things would be quite useful.

[Syntaxvgm]

What's more amazing is that it's still done with generic copper traces.

 

So, GPU makers, can we please see cards that can be configured to only use eight or four lanes? Even PCI-E 3.0 doesn't offer much benefit for x8 over x4, and being able to use available lanes for other things would be quite useful.

Not a bad idea really. Thing is it currently take a dual GPU card to see bottlenecks on 2.0 x16 or 3.0 x8, so single GPU cards could be done with 8x. After all a lot of motherboards do multi GPU in 8x still, and it works just fine. Thing is the full 16x actually helps support the card in the slot. Not that this is a good excuse, but worth mentioning. 

[JAKEBAB]

What's more amazing is that it's still done with generic copper traces.

 

What would u like them to use over copper?

[Syntaxvgm]

What would u like them to use over copper?

Unobtainium 

[geforceftw]

Im fine if it gets delayed. 3.0 16x is 32GB/sec.. That should fast enough at least for now. 

[Syntaxvgm]

Im fine if it gets delayed. 3.0 16x is 32GB/sec.. That should fast enough at least for now. 

But it's not just about 16x. 

[Altecice]

Yay more bandwidth we don't use!..... at x1,x4 hah!

[wpirobotbuilder]

What would u like them to use over copper?

Nothing. But we're pushing the limits of what copper can do, eventually we'll need something else. I don't know what that might be.

[MegaDave91]

Graphics card still can't saturate gen 3 16x, or am I wrong?

[Syntaxvgm]

Graphics card still can saturate gen 3 16x, or am I wrong?

If you meant can't, than you are right, yes. Dual GPU cards can saturate gen 2 though

[MegaDave91]

If you meant can't, than you are right, yes. Dual GPU cards can saturate gen 2 though

Ah yes, meant can't. Thanks for pointing that out.

[geforceftw]

But it's not just about 16x. 

I know i was talking in general. I had a feeling someone was going to point this out.

[Syntaxvgm]

I know i was talking in general. I had a feeling someone was going to point this out.

You do have a point though for desktop enthusiasts this doesn't mean much. 

I'm posting this from a laptop with Xfired GPUs, and to people like me , it means a lot, both in terms of SSDs and eGPUs

[geforceftw]

You do have a point though for desktop enthusiasts this doesn't mean much. 

I'm posting this from a laptop with Xfired GPUs, and to people like me , it means a lot, both in terms of SSDs and eGPUs

Yeah. I was going to bring up what you use it for but i have never really used a laptop in that way. Can you post a pic? 

[Syntaxvgm]

Yeah. I was going to bring up what you use it for but i have never really used a laptop in that way. Can you post a pic? 

I don't have an eGPU on this laptop, it's integrated. But it would cost over 1000$ to upgrade these cards. They are dual 6970m 2GB cards, and while that just fine now it won't be eventually. 

eGPUs like this use either express card or mini PCIe (even thunderbolt now) to provide a 1x connection to the GPU for an obviously throttled performance, but better than integrated. 

The alienware I'm using would be a perfect as it has multiple mini pcie connections, and had an HDMI input, meaning you can plug stuff into the screen on the laptop directly. It's awesome, i sometimes plug desktops into it and I frequently use game consoles on it. 

Otherwise (No idea how this guy here has it on his laptop screen) you would need to use an external screen or use a screen controller and a spliced LVDS cable as alienwares and 2 or so old clevos are the only laptop I know of that support direct input to the screen. 

EDIT

an i series CPU with HD gma graphics can be displayed on the screen if the card is nvidia fermi/kepler or newer using optimums. 

[pixeldensity]

so i'm a little lost here, what does a increase in pcie bandwidth mean for pc hardware

[Syntaxvgm]

so i'm a little lost here, what does a increase in pcie bandwidth mean for pc hardware

with 1x links you can have more parts on 1 lane. I guess. Cost savings they claim. 

[The Crazed Child]

Imagine mini PCIe 4.0 eGPUs?  :lol:  :lol:  :lol:  :lol:

Exactly my thoughts. Imagine a small GPU the size of an mSATA SSD than plugs right in to the mini PCIe slot.

[Cedimedi]

Nothing. But we're pushing the limits of what copper can do, eventually we'll need something else. I don't know what that might be.

 

I could image something like optical(fibre). The connectors tho would be painful. But i think its impossible to do this in near future. Must be expensive. Also, the chipset to cpu connection must be something other too.

[LukaP]

I could image something like optical(fibre). The connectors tho would be painful.

either we go fibre/get a quantum breakthrough, or we will only increase bandwith by making the slots 32x or more (to remain backwards compatible, something similar to how USB did the more pins thing would need to be implemented)

[NInety]

There needs to be a way you can stack VRAM when doing Crossfire or SLI. Maybe just a dedicated VRAM card?

[LukaP]

There needs to be a way you can stack VRAM when doing Crossfire or SLI. Maybe just a dedicated VRAM card?

NVLink. 

[NInety]

NVLink. 

I do not understand.  :mellow: