Mainstream PCIe 5.0 SSDs Only Expected in 2024

[TheLostSwede]

Although we've already seen some companies both tease and announce PCIe 5.0 consumer SSDs, it seems like we shouldn't expect mainstream PCIe 5.0 SSDs until 2024, at least if Silicon Motion's earnings call is anything to go by. Wallace Kou, Silicon Motion's CEO was quoted as saying ""It is likely that PCIe Gen4 will last a few years since Intel, AMD both continue to bring new upgrade variant of CPU with PCIe Gen4 to the market," continuing "Similarly, we are preparing for the launch of our third-generation PCIe Gen4 controller next year before transitioning to PCIe Gen5 in the following year."

This obviously has something to do not only with development time, but also with recuperating the costs of developing the previous generation of PCIe 4.0 SSD controllers before introducing new products that would have limited market share due to at least initially high prices. Silicon Motion is said to be working on multiple PCIe 5.0 SSD controllers with the enthusiast grade SM2508 expected towards the end of this year and it's set to compete with Phison's PS5026-E26 and InnoGrit's IG5666 controllers, plus unannounced inhouse controllers from the likes of Samsung and WD. Silicon Motion's mainstream PCIe 5.0 controller is said to be the SM2507, which is also said to be more suitable for mobile applications, most likely due to lower power consumption. However, the first PCIe 5.0 controller from Silicon Motion should be the SM8366 enterprise grade controller that was announced last year and which should be shipping in the second half of this year. For most users, PCIe 5.0 SSDs are unlikely to make a huge difference, unless we see some major improvements when it comes to random read and write performance, as this is currently the real bottleneck with SSDs.



View at TechPowerUp Main Site

 

[jesdals]

Well wonder if we will see mainstream integration of PCIe 5.0 on platforms before anyway considering recent status of both Intel and AMD platforms and current GPU standing

 

[TheLostSwede]

Well wonder if we will see mainstream integration of PCIe on platforms before anyway considering recent status of both Intel and AMD platforms and current GPU standing

I presume you mean PCIe 5.0?
Obviously we're moving that way, but for some reason Intel decided to only use it for the "GPU interface" which still seems like an odd move.
We still don't really know what AMD is planning for PCIe 5.0, beyond them saying there will be support for it on the AM5 platform.

 

[TheoneandonlyMrK]

So pciex 5 = pointless for consumers.

Good to know.

Save me a pointless upgrade.

 

[TheLostSwede]

So pciex 5 = pointless for consumers.

Good to know.

Save me a pointless upgrade.

We don't really know yet, but until we have some better NAND flash that improves the random performance, it's likely to offer no tangible benefits.
It's not as we're going to see consumer level controllers that goes up to 16-channels of NAND, as it gets way too expensive.

 

[Valantar]

Given that few if any applications today can make meaningful use of even the full speed of PCIe 3.0 (and if they do, the duration of the task is unlikely to be long enough for a further 2-3-4x speedup to be noticeable) .... meh. Maybe we should focus on improving Windows' and various software's ways of accessing data instead of just blindly chasing benchmark numbers? These things are no doubt great for bandwidth hungry servers, but consumers? Nope.

 

[mechtech]

That pic looks like a sata express?? Does pcie version affect them? Or is that a U2 plug?

look at all those power failure caps on that pcb. I think looks cooler without the case

 

[Daven]

Given that few if any applications today can make meaningful use of even the full speed of PCIe 3.0 (and if they do, the duration of the task is unlikely to be long enough for a further 2-3-4x speedup to be noticeable) .... meh. Maybe we should focus on improving Windows' and various software's ways of accessing data instead of just blindly chasing benchmark numbers? These things are no doubt great for bandwidth hungry servers, but consumers? Nope.

I would be interested in getting higher capacity SSDs at manageable prices. Also a rethinking of motherboard layouts to accommodate the addition of little PCBs connected onboard vs the cable connected drives (HDD, 2.5 SSD, optical drives) in bays of the past.

My dream scenario is system RAM like HBM on the CPU package and then converting DIMM slots into vertically inserted SSDs. Four SSDs could be connected in a very small area into four DIMM like slots. Then the CPU to memory traces in the motherboard could be converted into PCIe traces connected to the SSDs directly from the CPU.

 

[ncrs]

That pic looks like a sata express?? Does pcie version affect them? Or is that a U2 plug?

Yes.

SATA Express and U.2 (SFF-8639) share the same plug design, but I don't think there ever were any SATA Express drives released in the mainstream.
U.2 provides up to 4 PCIe lanes, which can be configured as 2x2 for multipath. It is affected by PCIe version - both sides negotiate maximum speed.

 

[stimpy88]

So it's pointless then? Yes yes, graphics cards will come out with support for it, but it won't make any difference to performance.

 

[Patriot]

That pic looks like a sata express?? Does pcie version affect them? Or is that a U2 plug?

look at all those power failure caps on that pcb. I think looks cooler without the case

That is sata express and while it made it on motherboards, I don't recall any drives making it to market.

u.2 is one of the enterprise nvme interfaces and it shares design with sas.
They are of course conveniently not compatible, just share the same connectors.
There are I think now 2 tri-mode HBA controllers that do support SAS and NVME but that requires backplane support.
U.3 was made by tosh to fix that? https://quarch.com/news/what-you-need-know-about-u3/
It's kind of a mess right now. Then there are multiple smaller form factors that are taking off.

 

[TheLostSwede]

That pic looks like a sata express?? Does pcie version affect them? Or is that a U2 plug?

look at all those power failure caps on that pcb. I think looks cooler without the case

NVMe drives don't have to use the M.2 form factor you know.
That's the SM8266 enterprise controller, so yes, U.2 or U.3 interface.

 

[defaultluser]

And until we fix the latency, the nvme drives will continue to be just as perceptibly-fast as the 1tb SATA drie I bought last year.

If only Optane had a future of higher density?

 

[mechtech]

NVMe drives don't have to use the M.2 form factor you know.
That's the SM8266 enterprise controller, so yes, U.2 or U.3 interface.

How common are they in enterprise world??

 

[Patriot]

How common are they in enterprise world??

Oddly about the same price as sata with all the markups. If you are buying SSDs you buy nvme.
It is the standard form factor for 1u/2u servers. the new high density stuff use edsff.

ASRock Rack 1U4G-ROME AMD EPYC 4x GPU 1U with EDSFF

In our ASRock Rack 1U4G-ROME review, we see how this 1U AMD EPYC GPU server adds an 8x EDSFF SSD array for a more modern capability

www.servethehome.com

 

[zlobby]

Another way for shallow-cache-SSD to ask for more money.

 

[Jism]

So pciex 5 = pointless for consumers.

Good to know.

Save me a pointless upgrade.

They are primarily designed for enterprise market. Consumers dont need PCI-E 5.0 yet unless you really really do shit that requires massive amount of bandwidth.

 

[TheoneandonlyMrK]

They are primarily designed for enterprise market. Consumers dont need PCI-E 5.0 yet unless you really really do shit that requires massive amount of bandwidth.

I've rocked 4x1TB nvme in raid0 <(11GB/s)and tried memory cache Extremereme, but nothing really gets a game to work That much better with the current software and hardware environment and APIs.
I just run two separate nvme now it's as good as it Presently needs to be.

 

[Jism]

Even a S-ata SSD is quite on par with a NVME drive, apart from when you throw in large workloads. Then the NVME ssd overblows the Sata.

 

[Minus Infinity]

And until we fix the latency, the nvme drives will continue to be just as perceptibly-fast as the 1tb SATA drie I bought last year.

If only Optane had a future of higher density?

Exactly. The fastest. PCI-E 4 drives only make a difference to someone doing large sequential file transfers a lot. ll the benchmarks show for real world usage for the vast majority of us, they make little difference over even a decent SATA SSD. Massive improvements in random ops is what we need.

 

[Tek-Check]

I presume you mean PCIe 5.0?
Obviously we're moving that way, but for some reason Intel decided to only use it for the "GPU interface" which still seems like an odd move.
We still don't really know what AMD is planning for PCIe 5.0, beyond them saying there will be support for it on the AM5 platform.

It might have been a vanity move from Intel, to be able to say they were the first to bring PCIe 5.0 to client market. But then again, someone's got to push forward at some point. It looks to me that vendors keep trying to enable one feature by one and see how it works before introducing something else. PCIe 5.0 is hard and on Z690 boars is meant for new GPUs coming in a few months. We have seen that performance gain in GPUs transitioning from x16 3.0 to x16 4.0 was rather minimal for gaming. I have not seen gains in heavy rendering in encoding. Those might be more substantial.

For NVMe gains, 5.0 vs 4.0 will be useful in server market, for sure. For home PC, I would agree with the other member who said more work is needed on OS to be able to utilize better 4.0, let alone 5.0. Hardware Unboxed tested NVMe 3.0 vs 4.0 in games. Improvemnt is really miniscule, so negligible. For example, 2-3 seconds faster loading time. Such high speeds and bandwidth might become more useful once neural engines and more advanced instructions are widely available.

I remember buying LG 9 series TV with 48 Gbps HDMI 2.1 ports almost one and half year before any HDMI 2.1 source came into the market. And then, in late 2020, GPUs and consoles were released with new interface. Release of products is not synced well and manufacturers seem not to play together to enable new tech in similar time.

 

[AusWolf]

Not that we really need it anyway. I can't even feel the difference between 3.0 and 4.0.

 

[Keullo-e]

Not that we really need it anyway. I can't even feel the difference between 3.0 and 4.0.

I can't even tell the difference between a good SATA SSD vs a NVMe drive as my system drive.

 

[Xajel]

I guess AMD postponement of PCIe 5.0 adoption and intel's limited PCIe 5.0 on GPU side only is just a way for them to follow the market.

It was rumoured that the first gen. of AM5 might not support PCIe 5.0. But intel's adoption meant AMD must accelerate (if that rumour was true, it was a rumour after all). And limiting it to the GPU meant they knew how is the market is going, especially that they have some secret roadmaps from their partners.

But, the move toward DirectStorage will mean another push is needed, for both the GPU & the NVMe side, so the system can push the data from the NVMe to the GPU as fast as they can.

AMD did a push with PCIe 4.0 when they released their platform, as they worked closely with Silicon Motion to release Phison E16, which was just an updated E12 but with a PCIe 4.0 PHY, the controller can reach higher performance that E12, but it didn't use the full PCIe 4.0 potential as they wanted to rush it in the market. AMD might as well do the same trick with AM5 later this year.

 

[Valantar]

I guess AMD postponement of PCIe 5.0 adoption and intel's limited PCIe 5.0 on GPU side only is just a way for them to follow the market.

It was rumoured that the first gen. of AM5 might not support PCIe 5.0. But intel's adoption meant AMD must accelerate (if that rumour was true, it was a rumour after all). And limiting it to the GPU meant they knew how is the market is going, especially that they have some secret roadmaps from their partners.

But, the move toward DirectStorage will mean another push is needed, for both the GPU & the NVMe side, so the system can push the data from the NVMe to the GPU as fast as they can.

AMD did a push with PCIe 4.0 when they released their platform, as they worked closely with Silicon Motion to release Phison E16, which was just an updated E12 but with a PCIe 4.0 PHY, the controller can reach higher performance that E12, but it didn't use the full PCIe 4.0 potential as they wanted to rush it in the market. AMD might as well do the same trick with AM5 later this year.

That's a bit of a weirdly convoluted take. The main reason: you're assuming that any of this tech is necessary or useful in some way on consumer platforms. From all available evidence, this is not the case. Not for GPUs, not for SSDs, not for other consumer AICs. There are no meaningful performance increases to be had outside of benchmarks.

So why, then, did Intel push PCIe 5.0, and why are AMD following?
- because people believe tech needs to be a "number goes up" game, and not doing that makes you look bad
- because having a new interface first is and has always been a good marketing point
- because selling future benefits of a tech with no current use is extremely lucrative
Common among all three? It's about marketing, not about the tech being actually useful. Sure, the engineers probably think the tech is cool and have ideas for future uses, but that doesn't make those uses realistic or relevant in the near future.

Remember, tech like DirectStorage reduces pressure on system interfaces, as data has to travel a much shorter path - rather than SSD -[PCIe]-> CPU [decompression] -[RAM bus]-> RAM -[RAM bus]-> PCIe root hub -[PCIe]-> VRAM it becomes SSD -[PCIe]-> PCIe root hub -[PCIe]-> VRAM with decompression happening in-place, so no uncompressed data is ever transferred. So, DS and any similar solutions lower pressure on system interfaces. That obviously doesn't mean that it can't lead to more utilization in the future - it does remove significant bottlenecks, after all - but given the low, low speed of adoption of this type of tech in game development this will be a 5-10-years-in-the-future type of thing, at the very least. There's obviously also the risk of poorly trained developers abusing a system like this to way too aggressively pre-cache anything and everything, but the fix to that is better development, not faster interfaces. Another important point to remember here: the main point of DS is to fix the problem that current games just don't really utilize SSDs well at all. To the degree that in most situations, there's no way of telling the difference between an old beater SATA SSD and a brand-new PCIe 4.0 NVMe drive. There is still tons left in the tank before we get around to needing faster interfaces due to this. The whole point is to make better use of what we already have. Turning that around and making it an argument for needing faster interfaces is entirely missing the point.