It's kinda funny with everyone rushing in with PCI-e 4.0 stuff fully knowing PCI-e 4.0 will have a fairly short shelf life with PCI-e 5.0 right around the corner.
The real interesting part of this is that PCI-e 4.0 has extremely limited market and PCI-e 5.0 should be bigger - it would be extremely likely that PCI-e 4.0 products will be compatible with PCI-e 5.0
You're only bashing PCI-e 4.0 because AMD has the lead on support for it. Pretty much anyone that's read any of your other posts can figure that one out.
Before you can get to 14 GB/s you need to get to 7 GB/s. If anything, the controller manufacturers need to come out with these controllers even quicker so they can come out with a PCIe 5.0 controller by the time motherboards start supporting them.
Judging from the Heat Sink in other PCI-E 4.0 SSD demo, it seems they are having a hard time getting 5 / 6 GB/s while keeping it within 5 / 6W. And if we had to move to 7nm for even the SSD Controller, the questions becomes will there be enough market for an PCI-E 5.0 10+GB/s SSD that gets a ROI with its 7nm controller.
Article says "The plan is to make SSDs featuring a PCIe 4.0 x4 interface available sometime in the second quarter of 2020 and aligning it with a launch of a next-generation mainstream PC platform."
I guess we can safely assume that this means Intel as the other next gen mainstream PC platform (AMD) is launching now.
If there is a demand for 5/6 GB/s NVMe drives you can be certain it won't matter if a NVMe drive uses 5 watts similar to current drives or 25 watts similar to hard dives. Of course it will matter in legacy applications but the real market will be in the sever industry where they buy new hardware every couple years. And you can bet server makers can easily crank out new designs that allow for little heat spreaders on NVMe drives.
No it is not. Also even if PCIe 5 hosts became widespread in 2020 (2021?) does it really matter if devices can't saturate a 4 lane PCIe 4 link? This same controller on a PCIe 5 link wouldn't be any faster. It took years for m.2 devices to saturate the 4 lane PCIe 3.0 m.2 connection. Maybe in 2021 or 2022 we might see devices maxing out what a PCIe 4.0 connection can provide.
The one nice thing is that PCIe is backwards and forward compatible. You can put a 5.0 device in a 4.0 slot or put a 4.0 device in a 5.0 slot albeit operating at 4.0 not 5.0 speed.
4.0 just launched and we're already 5/8th of the way to saturation, it won't take a year to get there. 4.0 is only double and market's already dev'ing for m.2 whereas it only started on 3.0 and before with sata which it had to move away from, somewhat(I wonder what the market share is between sssd sata vs m.2).
TheUnhandledException i think shabby for to add /sarcasm at the end of that :-) cause intel has been saying for 4 years now that 10nm chips are on track.. :-)
No it won't. Intel's PCIe 4.0 support is still some way off, so PCIe 5.0 support is even farther back in time. It is still unclear if Intel will manage to release mid and high power Ice Lake CPUs fabbed at 10nm+. If they don't (and they probably won't) then PCIe 4.0 support will have to wait for Tiger Lake (with Willow Cove cores), which will be fabbed at the 10nm++ node iteration, and apparently released in 2H 2020 (i.e. Q4 2020).
While they could have added a PCIe 4.0 controller in their upcoming Ice Lake-U CPUs for laptops I strongly doubt they did, since their thermals and clocks are already bad enough, and that would only make them worse. So don't expect PCIe 4.0 from Intel anytime soon, not in their CPUs anyway (they are offering PCIe 4.0 *and* PCIe 5.0 in their newest FPGAs, which are an entirely different thing).
PCIe 4.0 spec was released October 5, 2017. PCIe 4.0 products hit shelves July, 2019. (21 months later). PCIe 5.0 spec was released May 28, 2019. PCIe 5.0 products hit shelves ??? January 2021 maybe ???
That makes sense to me, 2 years to capitalize on for amd if intel skips 5.0 which they won't. But god do I hate this extra point of failure with those fans on mobos...
PCIe 5.0 will not come to consumer boards at all. It's a strictly server oriented technology. The problem is that even PCIe 4.0 is on the edge of what is possible in current consumer motherboard set ups as signal length is only 16cm (6 inches) without a repeater. For PCIe 5.0 that is halved again, so you need a power hungry repeater for every 8cm (3 inches). To even make PCIe 5.0 realistic in consumer boards the current ATX standard will have to go, and motherboards will have to have the CPU socket centred on the board with PCIe expansion slots on either side instead of the current set up with the CPU socket at one end of the board.
The real market for any new PCIe standards is not on the consumer side of things. So the length restrictions and power requirements won't really matter there. They of course still effect the sever side of things but they are needing more PCIe bandwidth everyday now. As such I'm sure they will make it work. You can fix the length issues with PCIe repeaters and new chase designs. The increased power requirements even currently don't matter. Sure a 5 or 10 watts for every gen5 controller adds up but when severs often have a thousand watts power miminmun that is hardly a design limitation. After gen5 launch in a few generations those advance meant will trickle down to the consumer market. When presumable it will start mater in the consumer sector. The power requirements will have gone down due to chip lithography getting smaller and PCIe repeaters or something other solution will be common place. All taking
Current repeaters are at least 96% efficient. Smaller lithography will not make them any more, or any less efficient. The issue is that they have to supply a fixed amount of power the PCB traces at a fixed voltage. Small lithography gets it power savings from the ability to use lower voltage which is not the case here. At least for the active part. Smaller lithographies also tend to be more leaky (higher stray capacitance) which you certainly don't want when dealing with fast switching serial interfaces. Anyway, there's also the question of why consumers would need PCIe 5.0 at all, and have to pay the power penalty involved. If I take an X570 platform and use the x16 slot on a bifurcation card to make a 4 SSD RAID 0, and we assume these drives max out the PCIe 4.0 standard for the full 30GB/s transfer rate that means that the SSDs combined have a transfer rate faster than dual channel DDR3-1866 RAM, or as fast as single channel DDR4-3750 RAM.
Due to the added board complexity over PCI-e 4.0, Need for retimers, and cost I dont see most consumers seeing PCI-e 5.0 for many years. PCI-e 5.0 I think will solely be used in the enterprise/server side while consumers will use PCI-e 4.0. The cost and complexity is just too great to abandon PCI-E so quickly
So what do you suggest they do, just wait for 5.0 instead? Why? 5.0 works with 4.0, so even if 5.0 comes a month after 4.0, why not be a month ahead of everyone else on a doubling of the speed? It will take a while to saturate 4.0 anyway, so even if 5.0 existed now, the first speed bumps over 3.0 will likely be 4.0 anyway.
You're right that that at the moment it looks like PCIe gen4 won't be that long lived. As intel is going strait for PCIe gen5. But since gen5 should be backwards compatible gen4 drives will work for a long time to come. As it seems at least on the high end some manufactures are getting close to maxing out the band with of PCIe gen4 x4 on NVMe drive already meaning there will likely be really tangible benifents to moving to PCIe gen 5 soon. Even if we don't need the speed of gen 5 we should be able to get the same speed at have the PCIe lanes which will be arguable more useful in the near future.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
27 Comments
Back to Article
rocky12345 - Tuesday, June 4, 2019 - link
It's kinda funny with everyone rushing in with PCI-e 4.0 stuff fully knowing PCI-e 4.0 will have a fairly short shelf life with PCI-e 5.0 right around the corner.DigitalFreak - Tuesday, June 4, 2019 - link
Not really. PCIe 4.0 exists now, at least in the AMD world. Any PCIe 4 device will work in a PCIe 5 slot, so there's no downside.HStewart - Tuesday, June 4, 2019 - link
The real interesting part of this is that PCI-e 4.0 has extremely limited market and PCI-e 5.0 should be bigger - it would be extremely likely that PCI-e 4.0 products will be compatible with PCI-e 5.0shabby - Tuesday, June 4, 2019 - link
You are extremely likely right...Korguz - Tuesday, June 4, 2019 - link
yea right... not likely...PeachNCream - Tuesday, June 4, 2019 - link
You're only bashing PCI-e 4.0 because AMD has the lead on support for it. Pretty much anyone that's read any of your other posts can figure that one out.Korguz - Tuesday, June 4, 2019 - link
heh.. trying to put a positive pro intel spin on this, like always ??Someguyperson - Tuesday, June 4, 2019 - link
Before you can get to 14 GB/s you need to get to 7 GB/s. If anything, the controller manufacturers need to come out with these controllers even quicker so they can come out with a PCIe 5.0 controller by the time motherboards start supporting them.ksec - Tuesday, June 4, 2019 - link
Judging from the Heat Sink in other PCI-E 4.0 SSD demo, it seems they are having a hard time getting 5 / 6 GB/s while keeping it within 5 / 6W. And if we had to move to 7nm for even the SSD Controller, the questions becomes will there be enough market for an PCI-E 5.0 10+GB/s SSD that gets a ROI with its 7nm controller.genzai - Tuesday, June 4, 2019 - link
is there any info, or even speculation, on when intel will offer PCIe support on their CPUs and chipsets?Irata - Wednesday, June 5, 2019 - link
Article says "The plan is to make SSDs featuring a PCIe 4.0 x4 interface available sometime in the second quarter of 2020 and aligning it with a launch of a next-generation mainstream PC platform."I guess we can safely assume that this means Intel as the other next gen mainstream PC platform (AMD) is launching now.
Skeptical123 - Wednesday, June 5, 2019 - link
If there is a demand for 5/6 GB/s NVMe drives you can be certain it won't matter if a NVMe drive uses 5 watts similar to current drives or 25 watts similar to hard dives. Of course it will matter in legacy applications but the real market will be in the sever industry where they buy new hardware every couple years. And you can bet server makers can easily crank out new designs that allow for little heat spreaders on NVMe drives.Death666Angel - Tuesday, June 4, 2019 - link
"with PCI-e 5.0 right around the corner"Oh, so I can buy it before Christmas? Because that would be "right around the corner".
shabby - Tuesday, June 4, 2019 - link
It'll be bundled with intel's 10nm desktop chips... which are right around the corner.TheUnhandledException - Tuesday, June 4, 2019 - link
No it is not. Also even if PCIe 5 hosts became widespread in 2020 (2021?) does it really matter if devices can't saturate a 4 lane PCIe 4 link? This same controller on a PCIe 5 link wouldn't be any faster. It took years for m.2 devices to saturate the 4 lane PCIe 3.0 m.2 connection. Maybe in 2021 or 2022 we might see devices maxing out what a PCIe 4.0 connection can provide.The one nice thing is that PCIe is backwards and forward compatible. You can put a 5.0 device in a 4.0 slot or put a 4.0 device in a 5.0 slot albeit operating at 4.0 not 5.0 speed.
rems - Tuesday, June 4, 2019 - link
4.0 just launched and we're already 5/8th of the way to saturation, it won't take a year to get there. 4.0 is only double and market's already dev'ing for m.2 whereas it only started on 3.0 and before with sata which it had to move away from, somewhat(I wonder what the market share is between sssd sata vs m.2).Korguz - Tuesday, June 4, 2019 - link
TheUnhandledException i think shabby for to add /sarcasm at the end of that :-) cause intel has been saying for 4 years now that 10nm chips are on track.. :-)Santoval - Tuesday, June 4, 2019 - link
No it won't. Intel's PCIe 4.0 support is still some way off, so PCIe 5.0 support is even farther back in time. It is still unclear if Intel will manage to release mid and high power Ice Lake CPUs fabbed at 10nm+. If they don't (and they probably won't) then PCIe 4.0 support will have to wait for Tiger Lake (with Willow Cove cores), which will be fabbed at the 10nm++ node iteration, and apparently released in 2H 2020 (i.e. Q4 2020).While they could have added a PCIe 4.0 controller in their upcoming Ice Lake-U CPUs for laptops I strongly doubt they did, since their thermals and clocks are already bad enough, and that would only make them worse. So don't expect PCIe 4.0 from Intel anytime soon, not in their CPUs anyway (they are offering PCIe 4.0 *and* PCIe 5.0 in their newest FPGAs, which are an entirely different thing).
pogostick - Tuesday, June 4, 2019 - link
PCIe 4.0 spec was released October 5, 2017. PCIe 4.0 products hit shelves July, 2019. (21 months later).PCIe 5.0 spec was released May 28, 2019. PCIe 5.0 products hit shelves ??? January 2021 maybe ???
rems - Tuesday, June 4, 2019 - link
That makes sense to me, 2 years to capitalize on for amd if intel skips 5.0 which they won't. But god do I hate this extra point of failure with those fans on mobos...SaturnusDK - Wednesday, June 5, 2019 - link
PCIe 5.0 will not come to consumer boards at all. It's a strictly server oriented technology.The problem is that even PCIe 4.0 is on the edge of what is possible in current consumer motherboard set ups as signal length is only 16cm (6 inches) without a repeater. For PCIe 5.0 that is halved again, so you need a power hungry repeater for every 8cm (3 inches).
To even make PCIe 5.0 realistic in consumer boards the current ATX standard will have to go, and motherboards will have to have the CPU socket centred on the board with PCIe expansion slots on either side instead of the current set up with the CPU socket at one end of the board.
Skeptical123 - Wednesday, June 5, 2019 - link
The real market for any new PCIe standards is not on the consumer side of things. So the length restrictions and power requirements won't really matter there. They of course still effect the sever side of things but they are needing more PCIe bandwidth everyday now. As such I'm sure they will make it work. You can fix the length issues with PCIe repeaters and new chase designs. The increased power requirements even currently don't matter. Sure a 5 or 10 watts for every gen5 controller adds up but when severs often have a thousand watts power miminmun that is hardly a design limitation. After gen5 launch in a few generations those advance meant will trickle down to the consumer market. When presumable it will start mater in the consumer sector. The power requirements will have gone down due to chip lithography getting smaller and PCIe repeaters or something other solution will be common place. All takingSaturnusDK - Thursday, June 6, 2019 - link
Current repeaters are at least 96% efficient. Smaller lithography will not make them any more, or any less efficient. The issue is that they have to supply a fixed amount of power the PCB traces at a fixed voltage. Small lithography gets it power savings from the ability to use lower voltage which is not the case here. At least for the active part. Smaller lithographies also tend to be more leaky (higher stray capacitance) which you certainly don't want when dealing with fast switching serial interfaces.Anyway, there's also the question of why consumers would need PCIe 5.0 at all, and have to pay the power penalty involved. If I take an X570 platform and use the x16 slot on a bifurcation card to make a 4 SSD RAID 0, and we assume these drives max out the PCIe 4.0 standard for the full 30GB/s transfer rate that means that the SSDs combined have a transfer rate faster than dual channel DDR3-1866 RAM, or as fast as single channel DDR4-3750 RAM.
BigMike678 - Wednesday, June 5, 2019 - link
Due to the added board complexity over PCI-e 4.0, Need for retimers, and cost I dont see most consumers seeing PCI-e 5.0 for many years. PCI-e 5.0 I think will solely be used in the enterprise/server side while consumers will use PCI-e 4.0. The cost and complexity is just too great to abandon PCI-E so quicklyajp_anton - Wednesday, June 5, 2019 - link
So what do you suggest they do, just wait for 5.0 instead? Why? 5.0 works with 4.0, so even if 5.0 comes a month after 4.0, why not be a month ahead of everyone else on a doubling of the speed? It will take a while to saturate 4.0 anyway, so even if 5.0 existed now, the first speed bumps over 3.0 will likely be 4.0 anyway.Skeptical123 - Wednesday, June 5, 2019 - link
You're right that that at the moment it looks like PCIe gen4 won't be that long lived. As intel is going strait for PCIe gen5. But since gen5 should be backwards compatible gen4 drives will work for a long time to come. As it seems at least on the high end some manufactures are getting close to maxing out the band with of PCIe gen4 x4 on NVMe drive already meaning there will likely be really tangible benifents to moving to PCIe gen 5 soon. Even if we don't need the speed of gen 5 we should be able to get the same speed at have the PCIe lanes which will be arguable more useful in the near future.Qasar - Thursday, June 6, 2019 - link
where has it been said that intel will skip pcie 4, and go strait to pcie 5 ?