PCIe 5.0 Ready For Prime Time

Good for compute I guess, consumer graphics cards still don't use enough bandwidth for 3.0 let alone 4 or 5, but bringing 5.0 motherboards out would be nice cause it would mean more pcie lanes for storage and things other than graphics cards.
 
Basically, yeah, PCIe 4.0 is being leapfrogged in favor of 5.0. As for AMD's 4.0 implementation, while not many devices will develop with 4.0, their MB should still be cross compatible with coming 5.0 devices but obviously stepped-down to 4.0 speeds.
 
Yea maybe they should worry about getting PCI-e 4.0 out the door before going to PCI-e 5. It seems everyone is in such a hurry these days to get the tech out asap because the older tech that probably came out like 1 year earlier is so slow now yet when it came out a year before it was labeled as blazing fast. Yes PCI-e 3.0 is 7 years old now and yet we are just now starting to come near it's limits with graphics cards. Oh but I forgot those PCI-e based SSD drives max it out and the one shown at CES was doing over 4GB's a second read and write on PCI-e 4.0 and of coarse that is not fast enough for the want to be tech nerds because hey we need moooore faster.

As for AMD PCI-e 4.0 boards or should I say CPU's with support for it. I read on Anandtech that even the older x370,x470 boards with a new Ryzen 3000 CPU will at least support PCI-e 4.0 on the first 16x slot because the travel distance form the CPU to the slot is more than short enough to support the signals fine as well as voltage requirements needed but they were not sure if the slots farther down would work in 4.0 mode. Either way probably better to just get a x570 board with direct support for PCI-e 4.0 that way it will most likely support PCI-e 5.0 at least in theory on the first slot unless AMD has already worked that all out and when 5.0 does drop the Ryzen 4000 or 5000 CPU's which might support 5.0 will drop into x570 boards and have 5.0 support. Either way it makes no difference to me the next card I get even if it supports PCI-e 4.0 and I get a x570 board I probably won't be upgrading to a card with 5.0 support for at least 2-3 years after 5.0 drops onto the market.
 

GPUs only require enough bandwidth to receive frame setup data from the CPU and we're barely at the point of needing 3.0x8 with the GTX2080. You'll need games with double the amount of scene data to max out 3.0x16. Probably won't need 4.0x16 for gaming before 2030.
 

Yes, we need 'more faster' but not necessarily for the reason you have in mind: if you double the per-lane PCIe bandwidth, you halve the number of lanes required to meet a device's bandwidth requirements and halve the number of PCIe lanes required from the CPU and chipset. Conversely, it doubles the amount of stuff you could conceivably connect to a system without increasing the CPU and chipset IO lane count.

10GbE requires two PCIe 3.0 lanes, one PCIe 4.0 lane, or you can squeeze 3x10GbE on a single 5.0 lane. Same thing with USB3.1-gen2 (10Gbps) if you want non-blocking bandwidth. ThunderBolt 3.0 goes from x8 to x4 and x2. Etc.

More bandwidth per lane opens up many options that wouldn't be viable otherwise.
 
Was just talking with a co-worker dealing with big data and VM infrastructure. Right now 10gb is the mainstream for that, but there's already a 25 and 40gb standard. When dealing with host servers talking ISCSI back to SANs, you'll want lots of redundancy and throughput.

For example many host servers will have a few 8x PCIe slots. At PCIe 5.0 standards, an x8 lane card is enough to support 30GB/s of throughout. That's more than enough for a 4 port 40Gb network card as it only needs 20GB/s of bandwidth.
 
Wonder if Pcie5 will finally allow dismissal of supplementary power cables, that would be cool for cable management. Last i read Pcie 4 could theoretically provide 300w from the slot but motherboard manufacturers were having trouble phasing it?
 
The power limitation comes from the connector design, each connector leaf has a fixed surface area and thickness and is only rated to handle "X" amps per leaf, coupled with only being able to use a percentage of connections on the connector for power the only way you can up the power available through it is to up the voltage per connection.
 

Unless they change the slot design, that isn't going to happen as you would need to at least quadruple the number of 12V pins in the slot.

Even if it did, you'd still need to get all of those however many extra amps per slot on the 12V rail from the PSU to the motherboard and PCIe slots, which means you'd still need auxiliary 12V cables from the PSU to somewhere near the PCIe slots similar to how the EPS12 connector(s) are right next to the CPU VRM. This could easily turn far more nightmarish than going directly to each card.
 
While PCIe 4.0 is extremely short-lived as a product, and I suspect Intel might even launch 5.0 with their next platform if they decide to hurry (which would go especially well if their GPUs launched as PCIe 5.0 cards, which they should be aiming for if they want to market them as superior products), one has to look at AMDs decision with future-proofing in mind. I think it was a fabulous move in favor of consumers, especially those who got older 3.0 boards that will now get 4.0 speeds enabled.

In the mainstream, where GPU support is the most relevant factor, 3.0 has not been maxed out yet, and 4.0 offers twice that speed. Even if cards supporting 5.0 speeds launch in the upcoming years, they will still not even reach anything close to actual 4.0 bandwidth usage (twice the speed of current 3.0), meaning they will work on 4.0 boards as well as they would on 5.0 boards. I actually suspect those 4.0 boards will be able to support max speeds of GPUs for many, many years to come.

While 3.0 might approach bottleneck area for highest-end GPUs in 2 or 3 generations, 4.0 won't for a long time. 5.0 simply has an insane additional reserve, considering it's quadrupling 3.0's bandwidth. 5.0 cards won't work faster on 5.0 boards than 4.0 boards unless 4.0 can't provide enough bandwidth though, which isn't likely to be the case for a very long time with the 3.0 not even bottlenecking the 2080Ti (it actually seems to be the very first card that merely shows any advantage in full 16x3.0 vs 8x3.0). Again, 4.0 is twice as fast.

While at first it might seem that there's no point to bother with 4.0 and we should've jumped straight to 5.0, we can actually get 4.0 right now, and for most users it will be as good as 5.0 for the lifetimes of their platforms.
 
Intel will go most likely directly to 5.0... maybe Nvidia goes Also to 5.0 2020 just to piss off the amd that ”only” have 4.0 and that time 😉

But Lets see what happens. Amd 500 series comes too soon to get 5.0 support. So it may be so that only Navi is supporting that standard and nothing else if everyone else moves directly to 5.0. It would put AMD somewhat tricky situation, but maybe amd 600 series motherboard Also support 5.0... who knows.
 

It doesn't really matter: as long as the AM4 socket has sufficiently good electrical properties to handle PCIe 5.0, nothing stops AMD from doing PCIe 5.0 on its 2020 CPUs and chipsets. GPU-wise, it makes very little difference as we're barely at the point where PCIe 3.0x8 can somewhat of a bottleneck in some games.

For most people, PCIe 4.0 is more of a preemptive measure (get the bandwidth out there before anyone really starts feeling the IO squeeze) than an immediate necessity.

AMD adopting 4.0 now is a win for everyone regardless of how short-lived it might be as the current standard since the alternative is being stuck at 3.0 speeds. With 4.0, at least you get to use 5.0 devices at 4.0 speed when those come out instead of 3.0.
 
In a typical SLI configuration, how much communication is there between two GPUs over PCIe? Or is most if not all coordination done over an external link module/bridge.
 


Devices come to market at 0.9. Also, they expect to ratify 1.0 this quarter.