News PCIe 6.0 SSDs for PCs won't arrive until 2030 — costs and complexity mean PCIe 5.0 SSDs are here to stay for some time

[thestryker]

U2 and M2 are not the same nor interchangeable.

Actually they are pretty much the same thing with different physical connectors. They both support a PCIe connection utilizing NVMe which is why you can connect U.2 drives to M.2 ports and vice versa. Otherwise U.2 has SATA/SAS capability as well if enabled by the platform and can leverage a wider voltage range.

 

[Nolandc]

But they're already talking about pcie7 releasing soon, so is 6 just going to get skipped? I think it will..

 

[bit_user]

But they're already talking about pcie7 releasing soon,

The spec did just get finalized.

https://www.tomshardware.com/tech-i...ets-1tb-s-for-8-0-as-exploration-phase-begins
​

There's typically a lag of years, between the spec being finalized and products reaching the marketplace. The above article includes this slide:

​

So, the first systems with PCIe 3.0 didn't ship until 2012 - 2 years after it was finalized. With PCIe 4.0, IBM had systems with it in 2018, but AMD didn't until 2019 and Intel didn't until 2020. PCIe 5.0 only made it to shipping hardware in 2021 - another 2-year delay.

so is 6 just going to get skipped? I think it will..

Concerning PCIe 6.0, Intel and AMD will feature it in their next server platforms, which I think ship early next year. I think they're both on about a 2-year cadence, which means we're not likely to see PCIe 7.0 before 2028.

The other thing that has to happen is there need to be devices which support it. If there aren't going to be devices on the market, before 2029, then there's no great need for CPUs to rush adoption much before then, right?

Finally, these speeds are not free. They add costs to the products which support them and take more power to run. So, nothing is going to adopt them until there's a market need that justifies the added cost & complexity. That could be among the reasons why PCIe 6 is taking longer than its predecessors to be implemented in shipping products.

 

[palladin9479]

Actually they are pretty much the same thing with different physical connectors. They both support a PCIe connection utilizing NVMe which is why you can connect U.2 drives to M.2 ports and vice versa. Otherwise U.2 has SATA/SAS capability as well if enabled by the platform and can leverage a wider voltage range.

They are different electrically for a reason, the data is the same but the control and signal requirements are different. M2 is really just a horizontally mounted PCIe slot, it's cheap and expects to be close to the PCIe host controller. U2 (Really U.3 now) expects to not be close and has different requirements, this also makes it more expensive. Since most PCIe host controllers are now device agnostic you can use a ghetto converter to plug a U2 device into a M2 slot, assuming the max signal distance is observed. Going the other way around is very dicey as the M2 devices do not tolerate being farther from the PCIe host controller. And this is with PCIe 4 and some 5, 6 has such short lengths that the usual suspects are jumping through hoops to figure out viable enterprise implementations. That is why I mentioned optical, it's really a thing and used to send PCIe signals longer then the specification length requirements because each interface has it's own termination and regenerates the electrical signal.

Again the scope here was that instead of using a PCIe 6 x4 M2 NVMe, having two PCIe 6.0 x 2 cables that went to two separate drives so that you'd get the same bandwidth as having two PCIe 5 x4 drives but at a distance.

 

[bit_user]

They are different electrically for a reason, the data is the same but the control and signal requirements are different.

I don't think anyone here expects to plug a passive cable into a M.2 slot, but I'd definitely need to see a source on your claim that the control signals are different.

The U.2 standard allows for drives that can support SAS and SATA, but I'm not sure if this works exactly like M.2 SATA or if it involves a different mechanism.

U2 (Really U.3 now)

No, you cannot use the terms interchangeably. You cannot put a U.2 drive in a U.3 host, although a U.3 drive is supposed to work in a U.2 host.

Going the other way around is very dicey as the M2 devices do not tolerate being farther from the PCIe host controller.

Whether or not that's true, there certainly do exist U.2 enclosures that can host a M.2 drive.

Amazon.com

 

[Nolandc]

According to "news sources" PCI-E 4.0 won't be available until 2028 at the earliest....and pci-e 5 hardware MIGHT be available to enthusiasts by 2030.....

This smacks of saying "just spend money NOW, don't hold onto it". I wonder who would benefit from buying a journalist/reviewer to say this?

The spec did just get finalized.

​

https://www.tomshardware.com/tech-i...ets-1tb-s-for-8-0-as-exploration-phase-begins​

​

There's typically a lag of years, between the spec being finalized and products reaching the marketplace. The above article includes this slide:

​

​

So, the first systems with PCIe 3.0 didn't ship until 2012 - 2 years after it was finalized. With PCIe 4.0, IBM had systems with it in 2018, but AMD didn't until 2019 and Intel didn't until 2020. PCIe 5.0 only made it to shipping hardware in 2021 - another 2-year delay.


Concerning PCIe 6.0, Intel and AMD will feature it in their next server platforms, which I think ship early next year. I think they're both on about a 2-year cadence, which means we're not likely to see PCIe 7.0 before 2028.

The other thing that has to happen is there need to be devices which support it. If there aren't going to be devices on the market, before 2029, then there's no great need for CPUs to rush adoption much before then, right?

Finally, these speeds are not free. They add costs to the products which support them and take more power to run. So, nothing is going to adopt them until there's a market need that justifies the added cost & complexity. That could be among the reasons why PCIe 6 is taking longer than its predecessors to be implemented in shipping products.

Didn't we just get the first wave of pcie5 hardware this year with the Samsung 9100pro and nvidias 5k series gpus though?

 

[bit_user]

Didn't we just get the first wave of pcie5 hardware at the end of last year with the Samsung 990evo and then this year with nvidias 5k series gpus though?

The first consumer PCIe 5.0 SSD came in Q2 of 2023, about half a year after AMD launched Ryzen 7000 and the AM5 platform (which supported an extra PCIe 5.0 x4 link from the CPU).

• https://www.tomshardware.com/reviews/inland-td510-ssd-review
• https://www.tomshardware.com/reviews/gigabyte-aorus-10000-ssd-review
• https://www.tomshardware.com/reviews/corsair-mp700-ssd-review
• https://www.tomshardware.com/reviews/crucial-t700-ssd-review

FYI, Intel's LGA1700 had a PCIe 5.0 x16 interface, but the only way motherboards could add PCIe 5.0 M.2 slots was to bifurcate that link, which I think not many did for the sake of adding PCIe 5.0 M.2 slots.

 

[thestryker]

Going the other way around is very dicey as the M2 devices do not tolerate being farther from the PCIe host controller.

This is just false and I'm not sure where you came up with it, but I'd be really curious to find out. IcyDock and OWC have had 2.5“ U.2 to M.2 adapters for years and they work perfectly fine with backplane or cabling just like native U.2 drives do. I'm not sure if anyone has PCIe 5.0 certified versions, and with the massive increase in enterprise SSD capacity I'm not sure there's a market, but they both have at least through PCIe 4.0.

 

[palladin9479]

This is just false and I'm not sure where you came up with it, but I'd be really curious to find out

It's in the PCIe spec, specifically the maximum signal length. Now check those adapters and look for some sort of conditioning circuitry. You will find that the reliable and more expensive adapters them on there while the cheap adapters that are basically just brackets do not. Running your NVME out of spec is a good way to kill it.

 

[thestryker]

It's in the PCIe spec

Feel free to cite specifics regarding the M.2 specifications instead of this hand waving.

 

[bit_user]

Now check those adapters and look for some sort of conditioning circuitry.

So what? I don't consider it materially significant, if they have redrivers or retimers. None of us was conditioning their statements on the absence of such necessities.

In fact, I've even read that PCIe retimers have been commonly utilized on motherboards since PCIe 4.0 first went mainstream. I'm not sure if that's true, but it'd support the notion that such supporting ICs have become a matter of course and certainly not a deal-breaker.

Feel free to cite specifics regarding the M.2 specifications instead of this hand waving.

I would also be interested in know where this information is coming from. I tried to find the relevant specs, but they're behind a paywall.

 

[Nolandc]

The first consumer PCIe 5.0 SSD came in Q2 of 2023, about half a year after AMD launched Ryzen 7000 and the AM5 platform (which supported an extra PCIe 5.0 x4 link from the CPU).

• https://www.tomshardware.com/reviews/inland-td510-ssd-review
• https://www.tomshardware.com/reviews/gigabyte-aorus-10000-ssd-review
• https://www.tomshardware.com/reviews/corsair-mp700-ssd-review
• https://www.tomshardware.com/reviews/crucial-t700-ssd-review

FYI, Intel's LGA1700 had a PCIe 5.0 x16 interface, but the only way motherboards could add PCIe 5.0 M.2 slots was to bifurcate that link, which I think not many did for the sake of adding PCIe 5.0 M.2 slots.

Ahh ok, it's just the GPUs from Nvidia and AMD that were late to pcie5. Thanks for all the info!

 

[bit_user]

Ahh ok, it's just the GPUs from Nvidia and AMD that were late to pcie5.

Whether or not they were "late" is very much a matter of perspective. The way I see it, Intel and AMD were too "early" in adding PCIe 5.0 to their platforms. It added cost and complexity for benefits too small and too far in the future.

TechPowerUp has a series of articles, going back several generations, where they test the impact of different PCIe speeds. Even with the very fastest GPU available, they found "you lose about 1% performance, across all three resolutions" when stepping down from PCIe 5.0 x16 to either PCIe 4.0 x16 or PCIe 5.0 x8.

https://www.techpowerup.com/review/nvidia-geforce-rtx-5090-pci-express-scaling/28.html
(above quote is taken from page 33: Conclusion)
​

We should also note that they achieved this using CPUs not available, back when PCIe 5.0 support was first launched. If they restricted themselves to i9-12900K and R9 7950X, I'm sure the gains would've been even smaller. This reinforces my belief that these products were not as "future proof" as users might've believed.

As TechPowerUp points out, the main benefit of using a PCIe 5.0 card is still really just that you could bifurcate the x16 link to x8, without really sacrificing anything. That's not nothing, but probably not relevant for many or even most, who just thought they were investing in future performance.

 

[Nolandc]

Whether or not they were "late" is very much a matter of perspective. The way I see it, Intel and AMD were too "early" in adding PCIe 5.0 to their platforms. It added cost and complexity for benefits too small and too far in the future.

Not prospective. GPUs only got pcie5 this year compared to ssds 2 years ago... and pcie6 is coming next year? Yeah late .

 

[logainofhades]

The average user doesn't even need Gen 4 speeds, for SSD, much less 5.0 or 6.0. We need more affordable high capacity drives, more than we do speed.