PCIe 6.0 Specification Reaches Milestone, Remains on Track for a 2021 Release

[Uskompuf]

PCI-SIG has recently confirmed that version 0.7 of the PCIe 6.0 Specification has been officially released to its members. The PCI-SIG organization has already ratified the PCIe 5.0 standard and plans to release the full PCIe 6.0 specification in 2021. PCIe 6.0 will bring the same doubling in data rates seen with previous generations of the standard including a 2x improvement over PCIe 5.0, 4x boost over PCIe 4.0, and an 8x increase in speeds over the common PCIe 3.0 standard.

The organization has been speeding up its timeline for new versions of the PCIe specification with PCIe 3.0 being released in 2010 followed by PCIe 4.0 in 2017, PCIe 5.0 in 2019, and with PCIe 6.0 expected in 2021. The PCIe 6.0 standard is designed to fulfill the needs of future devices in the PC and enterprise markets such as 800 Gb/s Ethernet cards. The earliest we can expect to see PCIe 6.0 devices in the PC market would be 2023 or 2024 depending on requirements.



View at TechPowerUp Main Site

 

[lexluthermiester]

Seriously... Do we even need PCIe 6.0? No one is or has a reason to use PCIe 5.0 and we're only just getting into PCIe 4.0...

Hey Industry pundits:
DING DING DING!! This is your 2020 wake up call, let's work on things that actually NEED work and updating. SATA specs for example. Seriously, get on it!

 

[Prima.Vera]

Bro, you can never have too much of a speed , bandwidth, etc.

 

[randomUser]

how many redrivers and PCB layers will we need to support PCIe 6?

Just by supporting PCIe4 prices shot up. Image how high will they go when pcie6 comes into play.

 

[Nuckles56]

This isn't for home users, not even for prosumers. This is for data centre use only as they can always use more speed, especially for more storage and GPUs and alike. I doubt we're likely to see anything beyond PCI-e 5.0 for home users before 2030.

 

[olstyle]

how many redrivers and PCB layers will we need to support PCIe 6?

Just by supporting PCIe4 prices shot up. Image how high will they go when pcie6 comes into play.

Exactly my thoughts.
That they will go for doubling speed again was pretty much a given, but some info on how they plan to achieve it (QAM?) would have been appreciated.
Edit: Ok, the slide says PAM-4. Yeah that's actually feasible, more clock seemed kind of improbable.

 

[hardcore_gamer]

Exactly my thoughts.
That they will go for doubling speed again was pretty much a given, but some info on how they plan to achieve it (QAM?) would have been appreciated.
Edit: Ok, the slide says PAM-4. Yeah that's actually feasible, more clock seemed kind of improbable.

Yup, it's still 16GHz Nyquist (same as PCIe 5.0) and uses PAM4 to double the bandwidth. Doubling the frequency is impractical on FR4.

 

[ExcuseMeWtf]

Even PCI-E 2.0 x16 is shown not to bottleneck hard current video cards lol and yet they keep coming with new ones .

 

[Ferrum Master]

Everyone is just eyeballing the 16x speed, while most important are 1-2x ones like for peripherals including Ethernet switches. Pushing more through one line, thus greatly reducing costs.

Frequencies will be raised, it can work on SoC level also on interposer, the future lies there creating a stacked combos. We are not limited to classic motherboards on FR4. PCIe is everywhere actually. Is it a mobile phone or router or any other device having a SoC...

So basically, if you don't need it, that's your issue for not grasping the whole picture.

 

[Metroid]

Seriously... Do we even need PCIe 6.0? No one is or has a reason to use PCIe 5.0 and we're only just getting into PCIe 4.0...

Hey Industry pundits:
DING DING DING!! This is your 2020 wake up call, let's work on things that actually NEED work and updating. SATA specs for example. Seriously, get on it!

Troll news I guess hehe and is explicit for 2021 ehhe

 

[Deleted member 24505]

Only really any use for storage controllers

 

[R0H1T]

No not really not even for storage, PCIe 6.0 1x is the equivalent of PCIe 2.0 16x & you can imagine there aren't too many drives coming close to utilize that amount of bandwidth. The only exception is probably the upcoming Optane Gen 2 (gen 3?) drives which are fundamentally different from regular NAND.

 

[Flaky]

let's work on things that actually NEED work and updating. SATA specs for example. Seriously, get on it!

For what reason, exactly? HDDs and ODDs are fine with current speeds.
SSDs are meant to be connected via PCIe, as it allows using protocol optimized for them (NVMe).

There's no point in maintaining and improving old tech when there's a better one already.

 

[londiste]

Seriously... Do we even need PCIe 6.0? No one is or has a reason to use PCIe 5.0 and we're only just getting into PCIe 4.0...

PCIe 4.0 took a long time to get the spec together.
PCIe 4.0 preliminary spec was in 2011 and final spec announced in June 2017. Same dates for PCIe 5.0 are June 2017 and May 2019.
For PCIe 6.0 preliminary spec was in June 2019 and this news is just the next version of it (updates are electrical spec and test chips data AFAIK). If it keeps being on track, the spec should be finalized in summer 2021.

When actual implementations in hardware happen is a different topic. Usually the first controllers or implementations take a few months to half a year from final spec.
Desktop and x86 CPUs will take longer, usually having to wait for a new platform generation. Whether or not newer versions are worth it for desktop is a matter of discussion and arguments for the foreseeable future, speed upgrades are awesome but the cost and complexity increases are also significant this time around.

 

[Mysteoa]

Seriously... Do we even need PCIe 6.0? No one is or has a reason to use PCIe 5.0 and we're only just getting into PCIe 4.0...

Hey Industry pundits:
DING DING DING!! This is your 2020 wake up call, let's work on things that actually NEED work and updating. SATA specs for example. Seriously, get on it!

PCIe 6.0 is mainly for servers where they could possibly saturate it. And Sata is dead, developing has stopped as HDD can't max out SATA3 and SSD are moving to PCIeX.

 

[bonehead123]

hummm...what SIG is NOT telling you is that PCIe-891.62 is just around the corner, BUT, in the mean time, you will first HAVE to buy stuff that uses 4, 5, 6 and all the other versions before you can use it, hahahah ......

 

[TechLurker]

I wouldn't be surprised if AMD and Intel didn't push PCIe 6.0 into mainstream around 2026-28 or so; building upon lessons learned from implementing them in the commercial/enterprise end (EPYC and Xeon).

AMD would definitely push it; given that their future vision for the Infinity Architecture partially scales with PCIe (more bandwidth for total heterogeneous computing), and given that their Zen archs are mostly uniform from the top-down. There's also the fact that AMD said they would pursue the leading edge, so it's likely that they will adopt it as soon as possible.

Intel would push it more to either try and remain ahead somewhere (like their original plan to skip PCIe 4.0 for 5.0 in 2020 before delays), or just to stay in sync with AMD given the bit of marketing brouhah that Ampere 4.0 marketing gave Intel and Intel mobo-makers, as well as in the highly competitive enterprise sector.

On the other end, with AMD having lit a fire under PCIe device makers towards faster components after pushing PCIe 4.0 into mainstream, and Intel supposedly aiming for mainstream PCIe 5.0 mid-late 2021 with AMD following late-2021 or early-2022, it wouldn't be all that surprising either if PCIe controllers continue to race towards PCIe 6.0 capability (instead of settling into a lull like during the PCIe 3.0 period). I feel though, that PCIe 6.0 will become the mainstay like PCIe 3.0 was. It'll take quite some time for tech to catch up to those speeds.

 

[TumbleGeorge]

I wouldn't be surprised if AMD and Intel didn't push PCIe 6.0 into mainstream around 2026-28 or so

If in this period mainstream exist. I think AMD will end evolution in mainstream with AM5 socket 2021-2025. Intel maybe just one year later. After that next "PC" will be just weak terminal to clouds like Stadia and other services. AMD maybe focused to HPC(servers, workstations, supercomputers) and car electronics. Maybe also will acompanies to mobile devices with GPU part of SoC's, like Exinos by Samsung.

The reason for stopping the development of even more powerful and productive mainstream components is that no one will need them. The latest models(in 2025) will be enough to meet all "home" consumer needs. And the prices of development and, respectively, the selling prices of the components after 2025 would be completely different from the usual ones for the mainstream.

 

[lexluthermiester]

For what reason, exactly? HDDs and ODDs are fine with current speeds.

Oh wow! You need lay off whatever you're doing because no, they're not. SATA SSD's can run sooo much faster but the SATA bus holds them back.

SSDs are meant to be connected via PCIe, as it allows using protocol optimized for them (NVMe).
There's no point in maintaining and improving old tech when there's a better one already.

Opinion and a very poor one. M.2 slots are not cost effective means to connect multiple drives in a single system, SATA is. Then there's the concerns about space for 6 or 8 connectors on the motherboard. Just not going to happen with M.2. Then there is the drive cooling question, how would one do that with M.2, again cost and space effective? Hmm? The notion of SATA being "old tech" and therefore useless is beyond short sighted and narrow minded. It needs to be updated, not replaced.

 

[londiste]

Oh wow! You need lay off whatever you're doing because no, they're not. SATA SSD's can run sooo much faster but the SATA bus holds them back.

U.2?

 

[lexluthermiester]

U.2?

Hmm. Hadn't seen that. Interesting. SATA is more ubiquitous however. It would much easier and cost effective to simply update existing specs.

 

[Flaky]

Oh wow! You need lay off whatever you're doing because no, they're not. SATA SSD's can run sooo much faster but the SATA bus holds them back.

In that particular sentence I've precisely mentioned HDDs and ODDs without SSDs. Please be more mindful when reading other's comments - you may be replying to something that hasn't been said.

Opinion and a very poor one. M.2 slots (...)

I though it's the SATA interface you actually want improved, but I see now it's mainly the form factor and connector count you're interested in.
I haven't used the "M.2" term - I've referred to PCIe as an interface. PCIe itself is flexible and fast enough to allow using single (x1) link for SSDs, and already be faster than SATA.
The tech is ready - there's simply no such connector to utilize only a singular lane.
And I think that's because there's no demand for it - either someone is fine with SATA limitations, or wants something much more, what requires more lanes (M.2 PCIe).

Some kind of a stop-gap may be using PCIe x1->M.2 adapters. Faster than SATA? Check. Have a free connector for it? Check... in most cases.

Then there's the concerns about space for 6 or 8 connectors on the motherboard. Just not going to happen with M.2.

It may happen if you consider vertical M.2 connectors

The notion of SATA being "old tech" and therefore useless is beyond short sighted and narrow minded. It needs to be updated, not replaced.

Old tech doesn't mean "useless". I have no idea where you got that from.

It would much easier and cost effective to simply update existing specs.

You may want to read SATA-IO FAQ. Mentions why there's no speed bump and what are the goals for SATA specs now.

 

[lexluthermiester]

In that particular sentence I've precisely mentioned HDDs and ODDs without SSDs. Please be more mindful when reading other's comments - you may be replying to something that hasn't been said.

I read what you said. Even HDD's burst mode functionality would benefit from SATA specs being updated. But I was specifically pointing out the SSD aspect because it is the main issue at hand with the current SATA spec limitations and you seemed to deliberately gloss over it.

I haven't used the "M.2" term - I've referred to PCIe as an interface.

It's still an M.2 variant slot regardless of the interface protocol.

It may happen if you consider vertical M.2 connectors

SATA connectors are much smaller and more space efficient.

Old tech doesn't mean "useless". I have no idea where you got that from.

Your statement directly implied it with statement "There's no point in maintaining and improving old tech when there's a better one already.".

 

[Rei]

Hey Industry pundits:
DING DING DING!! This is your 2020 wake up call, let's work on things that actually NEED work and updating. SATA specs for example. Seriously, get on it!

Oh wow! You need lay off whatever you're doing because no, they're not. SATA SSD's can run sooo much faster but the SATA bus holds them back.

Opinion and a very poor one. M.2 slots are not cost effective means to connect multiple drives in a single system, SATA is. Then there's the concerns about space for 6 or 8 connectors on the motherboard. Just not going to happen with M.2. Then there is the drive cooling question, how would one do that with M.2, again cost and space effective? Hmm? The notion of SATA being "old tech" and therefore useless is beyond short sighted and narrow minded. It needs to be updated, not replaced.

Well, according to SATA-IO's FAQ, here is what they claim: Doubling the native SATA speed would take too much time to catch up with the advancements in SSD technology, would require too many changes to the SATA standard & would result in a much greater power consumption compared with the existing PCI Express bus. As a widely adopted computer bus, PCI Express provides sufficient bandwidth while allowing easy scaling up by using faster or additional lanes.

While I do think that NVMe M.2 has replaced SATA as a boot drive bus but as a general storage PC bus, SATA is far from being replaced especially since much of legacy hardware is still supported. Though sooner or later, USB will eventually replace SATA once the software department could no longer support legacy systems.

 

[Aquinus]

Do we even need PCIe 6.0? No one is or has a reason to use PCIe 5.0 and we're only just getting into PCIe 4.0...

Fewer PCIe lanes should be a reason. If you quadruple PCIe bandwidth from what you're using now, then 16 lanes sounds like a whole lot more. Storage might only need a single PCIe lane and GPUs only 4. In that situation, 16 lanes sounds like more than enough. Cheaper chips could even have fewer PCIe lanes if they run faster, which makes for a smaller PCIe controller, fewer traces in the motherboard, and fewer pins/contacts on the CPU. Across the board, PCIe improves are a win. They just take time to be adopted in the consumer market.

A great example is the Radeon Pro 5600m in my laptop. It's only connected with 8 PCIe lanes at 3.0. Imagine if you only needed two lanes to do the same thing. Or even if you don't have wide adoption, what if Intel or AMD used it as the interconnect to their chipset/PCH. Then it starts sounding like a much better PCIe switch if it's 4 lanes to 20 lanes. It's just progress and there is nothing wrong with progress.

Edit: Also, you need a spec before you can implement it. This stuff takes time and doesn't happen overnight.