PCI-Express Gen 4 to arrive next year

[Godlygamer23]

3 hours ago, Curufinwe_wins said:

Dah! I fell prey to the OP's mistyping. My bad.

The article made the mistake too.

[elkenrod]

I get the "feels" just thinking that my first computer had ISA slots....

 

In 15 years PCI-E 4 will give someone those same feels.  I take comfort in that.

 

[NumLock21]

1 hour ago, elkenrod said:

I get the "feels" just thinking that my first computer had ISA slots....

 

In 15 years PCI-E 4 will give someone those same feels.  I take comfort in that.

 

Pcie 4 looks the same as any other. In 15 years there won't be any nostigla on the old board with old pcie slots. Pci might, just like isa, mca, and agp.

 

[Bananasplit_00]

On 2016-08-20 at 2:52 AM, Sammael said:

Probably faster ssd drives.

we dont have SSDs that are fast enough to use 16x(or even 8x IIRC) PCI-e 3.0 ffs, its just getting insaine lol, not that im against tech mooving forward but still...

[Sauron]

The bandwidth is irrelevant to consumers for now, pcie 3.0 is far from being saturated (frankly so is 2.0) by video cards and ethernet adapters. What is interesting is the higher power delivery, which could mean most graphics cards won't require a direct connection to the power supply on supporting motherboards. I wonder if motherboards will require more expensive components in order to provide that power and if the ATX 24-pin will be enough even on lower end psus.

[Godlygamer23]

14 minutes ago, Sauron said:

The bandwidth is irrelevant to consumers for now, pcie 3.0 is far from being saturated (frankly so is 2.0) by video cards and ethernet adapters. What is interesting is the higher power delivery, which could mean most graphics cards won't require a direct connection to the power supply on supporting motherboards. I wonder if motherboards will require more expensive components in order to provide that power and if the ATX 24-pin will be enough even on lower end psus.

I'm certain that the 24-pin won't be able to handle the additional current on any power supply unless larger wires will be used. I mean think about it this way: Motherboard manufacturers of expensive boards that support three and four cards in SLI/Crossfire will have supplemental connectors on them in the form of molex and PCIe 6 pins. The total power draw from the slots would be 300W for all four(300W/4=75W each). If that's going to be integrated into a single slot, the 24-pin connector might need to change to support the additional power draw, or we just have supplemental connectors on normal boards as well.

[Sauron]

Just now, Godlygamer23 said:

I'm certain that the 24-pin won't be able to handle the additional current on any power supply unless larger wires will be used. I mean think about it this way: Motherboard manufacturers of expensive boards that support three and four cards in SLI/Crossfire will have supplemental connectors on them in the form of molex and PCIe 6 pins. The total power draw from the slots would be 300W for all four(75*4). If that's going to be integrated into a single slot, the 24-pin connector might need to change to support the additional power draw, or we just have supplemental connectors on boards as well.

I think the key is "UP TO" 300W - I don't expect every slot to be able to provide that much at the same time. As for the 24 pin, technically speaking a single rail power supply can provide all of its power through a single connector iirc. Of course this is outside specification and not recommended at all, however it is possible and might even work well enough on a high quality psu. If I had to take a guess however I'd say they'll add a dedicated 8-pin connector on high end motherboards and disable the feature on those that can't handle it.

[Godlygamer23]

Just now, Sauron said:

I think the key is "UP TO" 300W - I don't expect every slot to be able to provide that much at the same time. As for the 24 pin, technically speaking a single rail power supply can provide all of its power through a single connector iirc. Of course this is outside specification and not recommended at all, however it is possible and might even work well enough on a high quality psu. If I had to take a guess however I'd say they'll add a dedicated 8-pin connector on high end motherboards and disable the feature on those that can't handle it.

It doesn't matter if it's only on one slot. My post is only really considering at least one slot being provided 300W at maximum. But, we have no idea how this is all going to work.

 

In regards to going outside of spec, I would surely hope that PCIe would not go outside of the ATX spec in order to achieve what it wants. That'd be kinda weird - setting a standard that violates another standard.

[79wjd]

Maybe they'll redesign/modernize the ridiculous 24-pin cable into something....less obnoxious to solve the potential power issue (rather than just adding auxiliary power headers to the board). 

[Sauron]

6 minutes ago, Godlygamer23 said:

In regards to going outside of spec, I would surely hope that PCIe would not go outside of the ATX spec in order to achieve what it wants. That'd be kinda weird - setting a standard that violates another standard.

That doesn't need to be the case, it's up to the board manufacturers to deliver the power to the slot, how they do it is their problem to solve. All the pcie spec says is that it can theoretically deliver 300W to an expansion card.

[Godlygamer23]

Just now, Sauron said:

That doesn't need to be the case, it's up to the board manufacturers to deliver the power to the slot, how they do it is their problem to solve. All the pcie spec says is that it can theoretically deliver 300W to an expansion card.

And herein lies the problem...You might still need to update the standard if they decide they're not going to use supplemental power connectors, and quite frankly, I think updating the ATX standard is the most elegant way of doing this to allow higher integration of components.

 

It doesn't matter if they're saying "up to." You still need to allow it to happen, otherwise it makes no sense. Companies might try to do it, and they need a means to do it, which may involve supplemental connectors, but I'd rather see the 24-pin updated so it can be integrated there, instead of having supplemental connectors all over the board.

[Sauron]

Just now, Godlygamer23 said:

And herein lies the problem...You might still need to update the standard if they decide they're not going to use supplemental power connectors, and quite frankly, I think updating the ATX standard is the most elegant way of doing this to allow higher integration of components.

 

It doesn't matter if they're saying "up to." You still need to allow it to happen, otherwise it makes no sense. Companies might try to do it, and they need a means to do it, which may involve supplemental connectors, but I'd rather see the 24-pin updated so it can be integrated there, instead of having supplemental connectors all over the board.

That's true, although there already are boards with an extra cpu 8-pin

[mariushm]

I'd like to see how they're going to push up to 300w through a bunch of tiny contacts in the pci express slot. It's not like there's many unused contacts to be repurposed for 12v . Maybe they can implement something to the effect of switching the 3.3v contacts to 12v by sending a command to the motherboard.

 

You guys are right that there's not enough 12v wires in the 24 pin connector to make it sensible to allow up to 300w in each pci express slot. Those wires can probably push up to about 200w in total to the motherboard and the 24 pin connector is so far away from pci-e slots that it would be a significant voltage drop through the motherboard.

If the ATX connector is about to be redesigned, it would make sense to add higher voltage wires like 20v to the power supply standard and dropping the -5v (already gone) and -12v (barely used for serial ports) which would make it possible to power some motherboards from 18.5-19v laptop adapters without major redesign of motherboards.

 

Right now there's 5.5A of current capability through the slot, which gives 75w .. with 20v it would already have 110w

[Godiwa]

On 21/8/2016 at 1:11 PM, Bananasplit_00 said:

we dont have SSDs that are fast enough to use 16x(or even 8x IIRC) PCI-e 3.0 ffs, its just getting insaine lol, not that im against tech mooving forward but still...

There is the thing called "controller cards" and "raid cards" which connect more then one drive, which could benefit of that extra speed.

[rm -rf]

On 20/08/2016 at 7:24 AM, Bananasplit_00 said:

WHAT NEEDS THIS BANDWITH? like its getting freaking silly, everything should be PCI-e lol, screw USB, just have a 1x slot  but still the through put speed is just getting insaine...

Well to misquote Bill Gates: "640k RAM is as much as anyone will ever need"

[Curufinwe_wins]

On 8/20/2016 at 9:34 PM, NumLock21 said:

Pcie 4 looks the same as any other. In 15 years there won't be any nostigla on the old board with old pcie slots. Pci might, just like isa, mca, and agp.

 

Agp does this to me. I fully admit it.

[mariushm]

In a few years, we may get rid of copper altogether and use laser through optical fibers to communicate with devices. A single fiber has the potential for 40-100 gbps or 12+ GB/s when pci-e v3.0 x16 is now less than 16 GB/s

It wouldn't seem far fetched for me to imagine optical fibers hidden inside the motherboard layers, going to various slots.

 

There are already silicon chips with transmitters and receivers built inside them, but they're still too big I guess to have several transmitters / receiver combo, one for each slot a motherboard could have. I don't see how we would have 8-16 separate transceivers inside a CPU die, to have 8-16 separate "lanes" like pci-express has. Some processors (high end Intel for example) have up to 40 pci express lanes in them.

 

Intel wanted to do something with fiber in the Thunderbolt specs, but switched to copper to make it possible to transmit power (10w) to connected devices for better acceptance (like USB was received). The initial specs for optical thunderbold were for 10 gbps, but the aim was for 100 gbps through optical fiber (about 12.5 GB/s - pci-e v3.0 x16 is about 15.6 GB/s and v4.0 x16 would be )

[patrickjp93]

10 minutes ago, mariushm said:

In a few years, we may get rid of copper altogether and use laser through optical fibers to communicate with devices. A single fiber has the potential for 40-100 gbps or 12+ GB/s when pci-e v3.0 x16 is now less than 16 GB/s

It wouldn't seem far fetched for me to imagine optical fibers hidden inside the motherboard layers, going to various slots.

 

There are already silicon chips with transmitters and receivers built inside them, but they're still too big I guess to have several transmitters / receiver combo, one for each slot a motherboard could have. I don't see how we would have 8-16 separate transceivers inside a CPU die, to have 8-16 separate "lanes" like pci-express has. Some processors (high end Intel for example) have up to 40 pci express lanes in them.

 

Intel wanted to do something with fiber in the Thunderbolt specs, but switched to copper to make it possible to transmit power (10w) to connected devices for better acceptance (like USB was received). The initial specs for optical thunderbold were for 10 gbps, but the aim was for 100 gbps through optical fiber (about 12.5 GB/s - pci-e v3.0 x16 is about 15.6 GB/s and v4.0 x16 would be )

The optronic computing community said the same thing. It's not viable for short, tiny communication paths. It takes more space and hardware than a metal or metaloid interconnect. Graphene nanowires will take over from Copper eventually. Also, PCIe 5 is likely to be optical for communication, but by the time it rolls around, we may have fixed many of the mass production snags for graphene.

 

There is a Fiberoptic variant for long-distance thunderbolt. It's just 5x as expensive per port and per cable.