Fouth Gen PCIe Sees Bandwidth Double

[Kaapstad]

Just found this on another website.

This week we got the first details about the fourth generation of the PCI express spec. PCIe 4.0 will have a base speed of 16 Gbps per data link.

PCI has been the de facto standard for PC expansion slots and now we have the first details about the fourth generation of the PCI express spec. 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.

Right now, there are several companies all vying for the chance to draft the new standard. Hardware manufacturers try to agree on one specific plan for these updates to help streamline the experience for end users. And with more devices needing greater speeds to take full advantage of their potential, this is a welcome update.

It's also expected to cut costs, as odd as that sounds. In the same way that shifting to a smaller manufacturing process for CPUs and GPUs can help cut costs by reducing component waste and helping systems run more efficiently, moving to a quicker expansion slot speed means manufacturers can use fewer data links or assign more tasks to each one, cutting down the total cost dramatically.

PCIe 4.0 could be even more valuable in the mobile space, where size, cost and efficiency are even greater concerns. With M-PCIe, manufacturers are beginning to adapt to the shifting landscape of computing. Mobile devices can take advantage of a second interlink called M-Phy, which can help enhance the performance of video cameras and communications chips in tablets and smartphones.

No matter how you slice it, PCIe's fourth generation is an exciting and welcome boost to current tech. Here's hoping we'll see that standard surface sooner, rather than later.

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

We may not need it yet but when we start looking at resolutions beyond 4K we will need the bandwidth.

 

[bakes0310]

Isnt there like a tiny difference between pci express 2.0 and 3.0 in terms of performance in games. Literally 1-3fps difference, im not holding my breath on 4.0, but you never know.

 

[Kaapstad]

Isnt there like a tiny difference between pci express 2.0 and 3.0 in terms of performance in games. Literally 1-3fps difference, im not holding my breath on 4.0, but you never know.

There is a larger difference when you start using very high resolutions and multi GPU setups.

For normal resolutions in use now there is hardly any difference though.

 

[kylew_is_back]

We may not need it yet but when we start looking at resolutions beyond 4K we will need the bandwidth.

Higher resolutions will not consumes more pcie bandwidth, in fact it is the opposite, lower resolutions will make the GPU ask data from the CPU more frequently (higher fps) therefore more pcie bandwidth consumption with low resolutions.

 

[StriderX]

I assume this is mainly for industrial uses rather than commercial, since we have nothing that can use that amount of throughput and frankly, by the time we get anywhere close, 64x PCIe slots may start coming out or Nvidia succeeds in making a new socket.

Silly really.

 

[Kaapstad]

Higher resolutions will not consumes more pcie bandwidth, in fact it is the opposite, lower resolutions will make the GPU ask data from the CPU more frequently (higher fps) therefore more pcie bandwidth consumption with low resolutions.

Don't work like that

It has been demonstrated when switching between PCI-E 2.0 and 3.0 that there was a big jump in performance when using a 3 way SLI setup at an uber resolution.

 

[Jokester]

That's because it's feeding data to 3 separate cards, it will likely become worse as well since AMD have already moved to putting inter card data over the bus and Nvidia will likely follow so bandwidth will be an issue, even on 2 card setups.

 

[Orangey]

I assume this is mainly for industrial uses rather than commercial, since we have nothing that can use that amount of throughput and frankly, by the time we get anywhere close, 64x PCIe slots may start coming out or Nvidia succeeds in making a new socket.

Silly really.

In TFA it explains why it's about more than just throughput.

And speaking of beyond 4K I found these last night, which I didn't want to start a thread for and have someone say "omg ooooold":

http://www.overclock.net/t/1481154/4k-eyefinity-crossfire-scaling-from-1-2-3-4-gpus-benchmarks

http://www.overclock.net/t/1487232/56k-warning-49-megapixel-gaming-test-warning-cell-phones-too

 

[andybird123]

I assume this is mainly for industrial uses rather than commercial, since we have nothing that can use that amount of throughput and frankly, by the time we get anywhere close, 64x PCIe slots may start coming out or Nvidia succeeds in making a new socket.

Silly really.

as mentioned in the article, this will make low power / smaller chips more efficient, as they can use half the lanes to get the same throughput, pcie isn't just for graphics cards

 

[whyscotty]

Don't work like that

It has been demonstrated when switching between PCI-E 2.0 and 3.0 that there was a big jump in performance when using a 3 way SLI setup at an uber resolution.

^^^
This

 

[kylew_is_back]

Don't work like that

It has been demonstrated when switching between PCI-E 2.0 and 3.0 that there was a big jump in performance when using a 3 way SLI setup at an uber resolution.

For SLI and CF yes will consumption more pcie bandwidth, but as you increase the resolution you will consume less pcie bandwidth because CPU send less fps(less data) to GPU.

Benchmarks show exactly that:

 

[Kaapstad]

as mentioned in the article, this will make low power / smaller chips more efficient, as they can use half the lanes to get the same throughput, pcie isn't just for graphics cards

Do you have the link to that 3 way GTX 670 setup tested at ultra high resolution on BF3 with PCI-E 2.0 and 3.0 please.

 

[Tonester0011]

Don't work like that

It has been demonstrated when switching between PCI-E 2.0 and 3.0 that there was a big jump in performance when using a 3 way SLI setup at an uber resolution.

Having more cards makes a difference as you tend to be running lower speeds (8x, 4x etc) i don't see where running a higher resolution comes into play?

 

[Kaapstad]

For SLI and CF yes will consumption more pcie bandwidth, but as you increase the resolution you will consume less pcie bandwidth because CPU send less fps(less data) to GPU.

Benchmarks show exactly that:

Wrong

The CPU will send less fps but the ones it does send are made up of bigger chucks of data.

 

[Kaapstad]

Having more cards makes a difference as you tend to be running lower speeds (8x, 4x etc) i don't see where running a higher resolution comes into play?

Because the individual frames @4K or higher are made up of a lot more data than @1080p for example. It is all about width rather than speed.

A good example @4K is the 512bit bus on the Hawaii cards getting the upper hand over the 384bit bus on the GK110 cards even though the bus on the Hawaii cards are running at a lower speed.

 

[andybird123]

here you go kaap

http://forums.evga.com/tm.aspx?m=1537816

 

[Kaapstad]

here you go kaap

http://forums.evga.com/tm.aspx?m=1537816

Thanks Andy.

I think the opening post in the link says it all.

 

[kylew_is_back]

Wrong

The CPU will send less fps but the ones it does send are made up of bigger chucks of data.

The ones it does send are made up from the same chucks of data regardless of the resolution. CPU have noting to do with resolution.

Benchmarks show that at 2560x1600 pcie bandwidth variations have small impact on performance, but in 1280x800 the impact on performance is larger.

 

[Kaapstad]

The ones it does send are made up from the same chucks of data regardless of the resolution. CPU have noting to do with resolution.

Benchmarks show that at 2560x1600 pcie bandwidth variations have small impact on performance, but in 1280x800 the impact on performance is larger.

We are not talking about 1600p, we are talking about 4K and higher for future use.

 

[Tonester0011]

Because the individual frames @4K or higher are made up of a lot more data than @1080p for example. It is all about width rather than speed.

A good example @4K is the 512bit bus on the Hawaii cards getting the upper hand over the 384bit bus on the GK110 cards even though the bus on the Hawaii cards are running at a lower speed.

Data is data, it's the same size irrespective of screen size (made up of the same size chunks rather) the overall data size increases, obviously hence you see a performance increase from 1.0 to 3.0, or when you increase the name of cards as the lanes become saturated. The bus width on the GPU bares no relevance to the PCI-E speed.