News All the PCIe 5.0 SSDs Coming Out in the Next Year or So

[Admin]

We finally have some model names for upcoming PCIe 5.0 drives. Here are all those we know about.

All the PCIe 5.0 SSDs Coming Out in the Next Year or So : Read more

 

[TechieTwo]

What an embarrassment that delivery of the PCIe 5 SSDs are almost a year late. Thankfully most of these SSDs do not require the stupid mile-high heat sinks. Typical low profile mobo supplied heatsinks are reported to work fine and not cause any throttling on the better drives.

 

[USAFRet]

What an embarrassment that delivery of the PCIe 5 SSDs are almost a year late. Thankfully most of these SSDs do not require the stupid mile-high heat sinks. Typical low profile mobo supplied heatsinks are reported to work fine and not cause any throttling on the better drives.

Why is this an "embarrassment" ?
A year late from what?

 

[peachpuff]

We don't need more speed we need bigger sizes, these don't decrease load times of the os/game at all so whats the point of them getting faster all the time?

 

[bit_user]

@apiltch , I couldn't comment on the main CES blog/article, but I wanted to thank you & your staff for the excellent show coverage!

 

[umeng2002_2]

RND4K is still the most important metric along with latency.

 

[bit_user]

What an embarrassment that delivery of the PCIe 5 SSDs are almost a year late.

Or, another way to look at it is that Intel was more than a year early in adopting PCIe 5.0!

If we rewind back to AMD's introduction of PCIe 4.0, there were drives that supported the interface that showed up sooner, but it took over a year since Ryzen 3k's launch for any PCIe 4.0 SSD to meaningfully exceed PCIe 3.0 speeds!

 

[bit_user]

Most of the drives we've seen rely on the Phison PS5026-E26 (or E26 for short). This 12nm controller has 8 channels ...

So glad this isn't TSMC N7, as that means there's probably some potential for power savings. Perhaps I'll wait until a N7 or N6-based controller is released. Until then, I'm sure PCIe 4.0 would suit me just fine.

 

[PlaneInTheSky]

Cooling fans on SSD? Is this some kind of joke. I don't even want one for free if they get that hot.

Only thing I want from SSD is more data for lower prices.

I couldn't care less how many gigabytes the sequential read speed is. It's completely irrelevant for regular consumers.

 

[PlaneInTheSky]

these don't decrease load times of the os/game at all

Yup, game loading time has nothing to do with the speed of SSD anymore. (yes going from HDD to SSD decreased loading times significantly, but it stopped there, faster SSD are irrelevant).

Loading times in games are caused by:

1. decompression, responsible for about 30% of loading times
2. drm like Denuvo, responsible for about 40%-60% of loading times

The first one can be fixed because developers will start to support Nvidia's GDeflate, where the GPU will do hardware decompression instead of the CPU.

The last one is a DRM issue and calls to Denuvo DRM servers during loading won't get any faster regardless of which hardware you have. When Denuvo is stripped from games, they tend to load significantly faster.

Articles claiming that DirectStorage and faster SSD will give PC loading times equal to PS5 are clueless. Loading times in games are caused by decompression and DRM. PS5 fixed both. PC can fix decompression but PC can not fix DRM the way it is implemented now. DRM checks on PS5 are internal with a dedicated chip, on PC it requires calls to online DRM servers, many slow back and forth handshakes and it relies on the CPU and network to do all those checks.

 

[bit_user]

The first one can be fixed because developers will start to support Nvidia's GDeflate, where the GPU will do hardware decompression instead of the CPU.

Or just use a faster compression method, like ZStd.

LOL @ using GPUs for decompression! That's got to be one of the least-efficient things you can do on a GPU!

 

[DavidLejdar]

We don't need more speed we need bigger sizes, these don't decrease load times of the os/game at all so whats the point of them getting faster all the time?

Yup, game loading time has nothing to do with the speed of SSD anymore. (yes going from HDD to SSD decreased loading times significantly, but it stopped there, faster SSD are irrelevant).
...
The first one can be fixed because developers will start to support Nvidia's GDeflate, where the GPU will do hardware decompression instead of the CPU.
...
Articles claiming that DirectStorage and faster SSD will give PC loading times equal to PS5 are clueless. Loading times in games are caused by decompression and DRM. PS5 fixed both. ...

Or just use a faster compression method, like ZStd.

LOL @ using GPUs for decompression! That's got to be one of the least-efficient things you can do on a GPU!

Playstation 5 actually comes with a Gen4 NVMe. And GDeflate is actually part of DirectStorage, which still needs to read that data, with the idea being that the GPU will speed up the loading when the CPU as such is at full load (which may happen in some rigs even without Denuvo taking up a large chunk of the processing power) - and this will require a NVMe drive, as a SATA SSD is clearly a bottleneck when a game wants to load several GBs into the memory.

Personally, I am playing The Witcher 3, where the Next-gen update optimized the game for NVMe usage. And loading time here, with everything set to max at 1440p, is around 2-3 seconds - and no stutter at all, with no asset popping in later.

That may not be a prime example, but I would not get such a loading speed with a SATA SSD. Depending on what one plays, NVMe may sure not make a difference though. E.g. some games may simply not want to load a lot of data into the memory at once, or especially older games are not coded to do that (with the loading still taking 10+ seconds). But in particular modern open-world games (especially with manual saves, where the player wants to load into whichever scenery they feel like right now), that is on a different level than some game on rails, which loads assets into the memory while the player moves towards the next area, and so on.

So as far as I am concerned, I have a slot for a Gen5 NVMe reserved, and looking forward to the next gen of game releases.

EDIT: And to name another example, I just realized that I have not seen a single "loading" in Death Stranding Director's Cut. One can of course argue that the loading is hidden behind cutscenes. But the result would probably not be like that with a SATA SSD:

 

[PlaneInTheSky]

LOL @ using GPUs for decompression! That's got to be one of the least-efficient things you can do on a GPU!

GDEFLATE decompression on the GPU is extremely efficient.

This graph doesn't even do justice how fast it is, because the PCIe 3.0 bus can not even keep up with the GPU's decompression speed.

And GDeflate is actually part of DirectStorage

It became part of the DirectStorage 1.1 update. Developers weren't very interested in DirectStorage before, GDEFLATE makes it worthwhile to implement.

 

[PlaneInTheSky]

Depending on what one plays, NVMe may sure not make a difference though.

HDD to Sata SSD made a huge difference in loading times. But faster SSD make a negligible difference.

The part we can improve on PC is decompression, by doing it on the GPU with GDeflate.

That really won't require a fast PCIe 4.0 or 5.0 drive to see the benefit. As you can see from the graph above, even on a PCIe 3.0 bus, GDeflate should be much faster.

The loading bottleneck will quickly become DRM after that.

 

[kiniku]

What an embarrassment that delivery of the PCIe 5 SSDs are almost a year late. Thankfully most of these SSDs do not require the stupid mile-high heat sinks. Typical low profile mobo supplied heatsinks are reported to work fine and not cause any throttling on the better drives.

and how many motherboards are available with a PCI 5.0 M2 slot? Not many. Most definitely hardly any 1 year ago.

 

[RichardtST]

We don't need more speed we need bigger sizes, these don't decrease load times of the os/game at all so whats the point of them getting faster all the time?

100% wrong. Maybe put a spinny drive back into your system to remind yourself how much of an effect disk I/O has on a system? Heck, even a proper SSD is slow as molasses now. Let's talk about VM restore time, shall we? When I can pause and restore a 16GB (in-memory) VM in just 2 or 3 seconds, that's amazing. That's time saved. I see, and appreciate, the difference in speed every day.

 

[peachpuff]

100% wrong. Maybe put a spinny drive back into your system to remind yourself how much of an effect disk I/O has on a system? Heck, even a proper SSD is slow as molasses now. Let's talk about VM restore time, shall we? When I can pause and restore a 16GB (in-memory) VM in just 2 or 3 seconds, that's amazing. That's time saved. I see, and appreciate, the difference in speed every day.

Who said anything about spinning rust? And what percentage of people restore vm's? Regular sata ssd's are fast enough for majority of people, we just need larger 8tb+ sizes.

 

[RichardtST]

Who said anything about spinning rust? And what percentage of people restore vm's? Regular sata ssd's are fast enough for majority of people, we just need larger 8tb+ sizes.

I can't understand your need for size, and you can't see my need for speed. We have different uses and requirements. Fair enough.

 

[alceryes]

It's the controller that needs cooling. Most NAND memory actually likes it toasty.
Ugh, I hope fan technology has improved from the old whiny tiny chipset fans of yesteryear. I'll never allow one of those in my system again. I'm somewhat sensitive to that high pitched noise. Both the NVMes in my new rig came with heatsinks attached. I knew I wouldn't need them, but the Silicone Power was such a good deal I couldn't pass it up. I had a nervous half hour or so prying and cutting them off with an X-Acto.

@peachpuff Kioxia announced some heavily stacked 3D NAND at CES. You'll be getting your extra storage (albeit at an extra price) soon enough.

 

[apiltch]

@apiltch , I couldn't comment on the main CES blog/article, but I wanted to thank you & your staff for the excellent show coverage!

Thanks so much! The mail blog / article should have allowed comments, but there's always weird stuff going on with the live blog template. Anyway, we don't get a ton of positive feedback so, after having put in a lot of late nights this week so I could support the team on the ground in Vegas (I sadly didn't get to go myself), seeing your comment means a lot to me.

 

[Giroro]

I will never be interested in any storage solution that requires a fan, not even if it were 100x faster than the fastest passively cooled option. I wouldn't notice the faster storage, but I would notice that fan every second of every day, forever.

 

[bit_user]

we don't get a ton of positive feedback so, after having put in a lot of late nights this week so I could support the team on the ground in Vegas (I sadly didn't get to go myself), seeing your comment means a lot to me.

Toms is a special site. Sometimes, I worry us readers are just a spoiled lot.
: (

 

[emike09]

I'm far more interested in affordable 4TB and 8TB NVMe drives. At this time, I could care less about these kinds of speeds. I'm still rocking a couple 2TB PCI-e 3 970 Evo Plus NVMe drives and very satisfied with the speed they offer.

 

[bit_user]

GDEFLATE decompression on the GPU is extremely efficient.

Heh, don't confuse efficiency with speed!

This graph doesn't even do justice how fast it is, because the PCIe 3.0 bus can not even keep up with the GPU's decompression speed.

Spoiler

It became part of the DirectStorage 1.1 update. Developers weren't very interested in DirectStorage before, GDEFLATE makes it worthwhile to implement.

Thanks for sharing. I wonder where you got it, because I'm curious whether the CPU path is single-threaded.

Even if it's not, consider that the ratio between GPU and CPU ranges between 10:1 and only 3:1. Considering how much more raw compute power GPUs have, it's pretty obvious GDeflate is implemented by scalar code on the GPU, rather than SIMD. That's what I was afraid of.

Anyway, I didn't outright say it was bad idea, just that it's not what GPUs are good at. So, if we're at a place where we're having to make them do decompression, perhaps it shows some weakness in the system architecture.

That really won't require a fast PCIe 4.0 or 5.0 drive to see the benefit. As you can see from the graph above, even on a PCIe 3.0 bus, GDeflate should be much faster.

Depends on what you spend most of your time loading. Textures are already compressed using texture compression, and probably won't gain much benefit from GDeflate.

 

[bit_user]

Most NAND memory actually likes it toasty.

IIRC, heat is generally good for reducing wear, but bad for data retention?