Next-Generation NVMe SSD and I/O Technology [PC, PS5, XBSX|S]

[pjbliverpool]

I recall some benchmarks earlier this year and the improvements were minor.

The improvements were actually significant in terms of reducing the file IO overhead. However that simply revealed other bottlenecks that need to be addressed by the developers or a later iteration of Direct Storage in the case of the decompression workload. The devs have already stated their intention to continue working on those other bottlenecks to bring the load times down to around 1 second - on PC.

But again, much depends on someone's expectations. It's not a magic bullet to drastically slash load-times even on modern hardware in the way the changes of this generation of consoles was really quite spectacular when developers put in the effort, e.g. Astro's Playroom, Spider-Man and Miles Morales, Horizon Forbidden West, and Ratchet & Clank.

I may have misinterpreted what you're saying here but there is no magic bullet on the consoles either. That's why games don't automatically load faster there and in many cases still load faster on PC's with faster CPU's. All the new consoles are doing at a high level is:

1. Providing a mutli-GB/s data path between the storage device and video memory
2. Removing the majority of the processing associated with high IO rates from the CPU

That's great but it means nothing if the developer doesn't then address the other CPU related bottlenecks that can slow down load times which have nothing to do with either of the above.

With GPU decompression enabled Direct Storage, the PC will be doing both of those things too. Granted the consoles still probably remove a bit more CPU overhead but by that point we're probably just playing around in the margins anyway.

I think we can be confident that with a similar speed SSD and a fully implemented version of Direct Storage, a well designed game will load on decent PC at a very similar rate to the PS5.

 

[Deleted member 11852]

I may have misinterpreted what you're saying here but there is no magic bullet on the consoles either. That's why games don't automatically load faster there and in many cases still load faster on PC's with faster CPU's.

Console games are not magically faster, most of Sony's teams are clearly putting the effort in on a bunch of games but not all, e.g. Demon Souls is nowhere instant loading. Ubisoft are still managing to release games that take an age.

That's great but it means nothing if the developer doesn't then address the other CPU related bottlenecks that can slow down load times which have nothing to do with either of the above.

I don't think it's CPU being the bottleneck per se, I think it's simply that a bunch of devs have not re-architectured the way their games are designed and stored on disc. On PC, and still on a bunch of console games, there are still clearly games with a chunk of PAK files where assets need to be loaded, unpacked, probably decompressed in memory, before the basic data can be used. And I'd wager a lot of this data is in a format that is just convenient - possibly for cross-platform maintenance - rather than for performance.

Then there are games like Spider-Man and Ratchet & Clank on PS5 which are clearly doing everything they can in terms of storing data, and having it in a readily-usable format without seconds (multiple frames) of CPU work. I don't think it's a coincidence that both of these games are from Insomniac, because as far back as PS2 their games gad pretty trim loading times.

I hope that Insomniac do a deep dive into the changes made for Spider-Man on PS5 and PC, and Ratchet & Clank, just like their did for Spider-man on PS4. They really need to be sharing their approach so others can consider adopting and implementing it.

edit: typos

 

[Shifty Geezer]

I don't think it's CPU being the bottleneck per se, I think it's simply that a bunch of devs have not rearcgitectures the way their games are designed and stored on disc. On PC, and still on a bunch of console games, there is still clearly a chunk of PAK files where assets needed to be loaded, unpacked, probably decompressed in memory, before the basic data can be used. And I'd wager a lot of this data is a format that is just convenient - possibly for cross-platform maintenance - rather than for performance.

Yes. Legacy structures from build tools that work a certain multiplatform way. We're still in fast IO infancy.

 

[arandomguy]

That just means the cost of entry for the technology just becomes cheaper, which is a good thing for everyone.

I'm not understanding your reasoning here given the context of the discussion? I thought the discussion here was in the context of buying NVMe storage in preparation for Direct Storage implementation in gaming?

The cost of entry being lower in the future would be another reason to not buy into it now.

 

[Remij]

That chart doesn't really indicate that. This is basically just talking about how they plan on doing GPU decompression. I've seen nothing to suggest that DS will ever be removing the copy to system RAM step, though.

Yes we have. It's already been posted here. AMD has already come out and said that their implementation will do just that.. they've already elaborated on it, and the fact that it uses SAM (Resizable Bar) Nvidia posted a similar graph ages ago showing RTX/IO bypass the CPU and system RAM completely.. though obviously the CPU will have to play a part in that.

DirectStorage = I/O optimizations, file stack improvements + a standard for GPU based decompression

It's meant to work on a wide range of devices and whether your hardware supports all features or not, it's meant to choose the best path for data to take given your hardware support.

However, when it comes to the GPU decompression aspect, DirectStorage requires hardware vendors support. Each GPU vendor will do their own implementation.. while supporting the compression standard that DirectStorage puts forward. It requires GPUs to have Resizable Bar support, which is what enables the CPU to actually send these massive amounts of data to VRAM quickly.

This is why GPU vendors say they "plug in" to DirectStorage. DirectStorage is its own GPU independent thing... however it's not really complete until the GPU vendors unlock the final piece of the puzzle.

We can assume AMD and Nvidia are ready... but maybe Intel isn't? Maybe that's what's taking long?

 

[pjbliverpool]

Yes we have. It's already been posted here. AMD has already come out and said that their implementation will do just that.. they've already elaborated on it, and the fact that it uses SAM (Resizable Bar) Nvidia posted a similar graph ages ago showing RTX/IO bypass the CPU and system RAM completely.. though obviously the CPU will have to play a part in that.

DirectStorage = I/O optimizations, file stack improvements + a standard for GPU based decompression

It's meant to work on a wide range of devices and whether your hardware supports all features or not, it's meant to choose the best path for data to take given your hardware support.

However, when it comes to the GPU decompression aspect, DirectStorage requires hardware vendors support. Each GPU vendor will do their own implementation.. while supporting the compression standard that DirectStorage puts forward. It requires GPUs to have Resizable Bar support, which is what enables the CPU to actually send these massive amounts of data to VRAM quickly.

This is why GPU vendors say they "plug in" to DirectStorage. DirectStorage is its own GPU independent thing... however it's not really complete until the GPU vendors unlock the final piece of the puzzle.

We can assume AMD and Nvidia are ready... but maybe Intel isn't? Maybe that's what's taking long?

I think most of what you say there is probably correct, however I don't see resizeable bar being a requirement for GPU decompression. I know SAS involves it for increased efficiencies somehow but there's no reason for it to be needed. The hardware decompressor on the consoles would likely have far less memory available to it for the same operation than the 256MB available to non resizable bar enabled systems.

Making that a requirement would massively reduce the compatibility of the solution and make the very wide SM6.0 GPU requirement pointless.

I expect resizable bar (and system memory bypass) to be optional additions to the basic GPU decompression function if your system supports them.

 

[davis.anthony]

I'm not understanding your reasoning here given the context of the discussion? I thought the discussion here was in the context of buying NVMe storage in preparation for Direct Storage implementation in gaming?

The cost of entry being lower in the future would be another reason to not buy into it now.

It's quite simply, you stated this:

What if we find out that having the fastest NVMe drives actually have tangible differences than slower NVMe drives?

In that case people don't need to buy a $200 NVME drive as an $80 drive offers the same performance.

$80 is a much lower price making the prince of entry to that technology cheaper.

$80 is cheap enough for people to buy in to it now in preparation for it.

 

[arandomguy]

It's quite simply, you stated this:



In that case people don't need to buy a $200 NVME drive as an $80 drive offers the same performance.

$80 is a much lower price making the prince of entry to that technology cheaper.

$80 is cheap enough for people to buy in to it now in preparation for it.

You seem to be ignoring the other scenarios presented in that what if we find out the $80 drive simply is not suitable compared to the $200 drive once we have actual data with DirectStorage games. The problem is the uncertainty at this point, buying a $200 drive could be a way overspend, buying a $80 drive could turn out to be not suitable. Right now it's just all theory of how these solutions will actual work in practice.

I'm still having trouble with where you are going with this line of reasoning. This discuss chain was based on the question of whether or not one should be buying into the DirectStorage ecosystem right now at this present time versus waiting.

If you want to buy a NVMe drive for other reasons right now that is another matter. But given that DirectStorage is not implemented (with the sole current known implementation to not be until early 2023, who knows about wider adoption rate), the question marks surrounding it (what to actually buy) and the likely general trend of adoption being cheaper in the future I have to again ask why would someone buy into in advance for the express purpose of preparing for DirectStorage?

This also relates to the broader issue of adoption rate for both consumers and developers. Consumers will be slow to adopt if there is software that is convincing them (including marketing) to upgrade, which is the case at this stage as there still isn't even single DirectStorage game and only 1 announced. Developers will also be slow to adopt if the consumer base is slow to adopt. So both these factors feed into each other.

 

[davis.anthony]

You seem to be ignoring the other scenarios presented in that what if we find out the $80 drive simply is not suitable compared to the $200 drive once we have actual data with DirectStorage games. The problem is the uncertainty at this point, buying a $200 drive could be a way overspend, buying a $80 drive could turn out to be not suitable. Right now it's just all theory of how these solutions will actual work in practice.

You mean I ignored a silly scenario? $80 drives are 3.5GB/s which is the same as the Series consoles, so it's a big reach to assume 3.5GB/s won't be enough as games will be built with that as a potential baseline, again, due to Xbox.

I'm still having trouble with where you are going with this line of reasoning. This discuss chain was based on the question of whether or not one should be buying into the DirectStorage ecosystem right now at this present time versus waiting.

By 'buying into the DirectStorage ecosystem' you mean simply buying an $80 NVME drive? Which people do anyway without even thinking about DS.

If you want to buy a NVMe drive for other reasons right now that is another matter.

NVME's offer the best GB per dollar and the best transfer rate of any other drive per dollar.

So the question is, why should anyone buy an SSD or HDD over an NVME if they're in the market for a new drive?

But given that DirectStorage is not implemented (with the sole current known implementation to not be until early 2023, who knows about wider adoption rate), the question marks surrounding it (what to actually buy) and the likely general trend of adoption being cheaper in the future I have to again ask why would someone buy into in advance for the express purpose of preparing for DirectStorage?

Again, it's an $80 drive, you don't have to build a new $3000 PC to take advantage of DS *smh*

This also relates to the broader issue of adoption rate for both consumers and developers. Consumers will be slow to adopt if there is software that is convincing them (including marketing) to upgrade, which is the case at this stage as there still isn't even single DirectStorage game and only 1 announced. Developers will also be slow to adopt if the consumer base is slow to adopt. So both these factors feed into each other.

You know what will entice developers to use DS? Knowing performant NVME drives are only $80 and thus within reach of pretty much everyone.

Here's something crazy, when my NVME drive arrives from Amazon today my PC will DS compliant and get this, I didn't mean to.

I simply built a PC and DS compliance 'just happened'

 

[Pinstripe]

Well, a HDD is still the most common sense choice for "inert" data such as music, videos, backups, etc.

 

[PSman1700]

I see no reason why anyone would opt for anything else then a NVME drive for gaming at this point in time. Its almost impossible to buy a prebuild without one (or a laptop).

 

[davis.anthony]

Well, a HDD is still the most common sense choice for "inert" data such as music, videos, backups, etc.

As long as people are happy with slow transfer speeds they're OK.

 

[pjbliverpool]

Well, a HDD is still the most common sense choice for "inert" data such as music, videos, backups, etc.

I used to think that but I'm planning to replace by 4TB HDD with a big SATA SDD as soon as possible due the noise it makes. Currently my HDD spinning is the noisiest part of my PC when not gaming.

 

[arandomguy]

You mean I ignored a silly scenario? $80 drives are 3.5GB/s which is the same as the Series consoles, so it's a big reach to assume 3.5GB/s won't be enough as games will be built with that as a potential baseline, again, due to Xbox.

By 'buying into the DirectStorage ecosystem' you mean simply buying an $80 NVME drive? Which people do anyway without even thinking about DS.

NVME's offer the best GB per dollar and the best transfer rate of any other drive per dollar.

So the question is, why should anyone buy an SSD or HDD over an NVME if they're in the market for a new drive?

Again, it's an $80 drive, you don't have to build a new $3000 PC to take advantage of DS *smh*

You know what will entice developers to use DS? Knowing performant NVME drives are only $80 and thus within reach of pretty much everyone.

Here's something crazy, when my NVME drive arrives from Amazon today my PC will DS compliant and get this, I didn't mean to.

I simply built a PC and DS compliance 'just happened'

I'm just going to make this my final post on this as there seems to be some sort of communication issue here that is just going around in circles.

The context is buying a NVMe drive now specifically to prepare in advance for Direct Storage. As in you have no need for a new NVMe drive right at this moment and are only wanting to buy one in preparation for DirectStorage. If you have no need for a NVMe drive right now it does not make sense to buy it in advance simply for DirectStorage as we have no data.

There is no advantage in doing so in advance. There are many unknowns still despite your assertion that simply having advertised sequential read speeds above the Xbox will result in an essentially identical experience regardless of the performance differential between NVMe SSDs (as in the drive doesn't matter).

If you happen to need or want a NVMe SSD in general for other reasons right at this moment that is outside of the scope of what I am trying to communicate here. If I have no other need I would want to put off the purchase as long as possible until I have more data to guide a purchasing a decision. I would not say just preemptively replace a SATA SSD right now with a NVMe.

 

[PSman1700]

I'm just going to make this my final post on this as there seems to be some sort of communication issue here that is just going around in circles.

The context is buying a NVMe drive now specifically to prepare in advance for Direct Storage. As in you have no need for a new NVMe drive right at this moment and are only wanting to buy one in preparation for DirectStorage. If you have no need for a NVMe drive right now it does not make sense to buy it in advance simply for DirectStorage as we have no data.

There is no advantage in doing so in advance. There are many unknowns still despite your assertion that simply having advertised sequential read speeds above the Xbox will result in an essentially identical experience regardless of the performance differential between NVMe SSDs (as in the drive doesn't matter).

If you happen to need or want a NVMe SSD in general for other reasons right at this moment that is outside of the scope of what I am trying to communicate here. If I have no other need I would want to put off the purchase as long as possible until I have more data to guide a purchasing a decision. I would not say just preemptively replace a SATA SSD right now with a NVMe.

When talking deployment you have to consider pre-builds, laptops and custom builds since 2017/2018, how many of those have opted out of nvme's since then? Someone still on a sata 3 drive is most likely also on a older system that wont even support nvme ssd drives and thus cant even upgrade to m2 nvme. These users are looking at a mb/cpu upgrade to begin with, and when they do, they will most likely go with nvme since thats cheaper and faster.

Today nvme is about the only way to go if you arent delibirately opting out on it for some reason. And yes, going from a sata 3 ssd to nvme does improve performance across the board, running windows updates, file transfers and even responstimes. Not the difference from hdd>ssd, but large enough a difference to warrent a measily 80 dollar price tag for the average m2 ssd at 1tb of storage.

 

[davis.anthony]

I'm just going to make this my final post on this as there seems to be some sort of communication issue here that is just going around in circles.

The context is buying a NVMe drive now specifically to prepare in advance for Direct Storage. As in you have no need for a new NVMe drive right at this moment and are only wanting to buy one in preparation for DirectStorage. If you have no need for a NVMe drive right now it does not make sense to buy it in advance simply for DirectStorage as we have no data.

There is no advantage in doing so in advance. There are many unknowns still despite your assertion that simply having advertised sequential read speeds above the Xbox will result in an essentially identical experience regardless of the performance differential between NVMe SSDs (as in the drive doesn't matter).

If you happen to need or want a NVMe SSD in general for other reasons right at this moment that is outside of the scope of what I am trying to communicate here. If I have no other need I would want to put off the purchase as long as possible until I have more data to guide a purchasing a decision. I would not say just preemptively replace a SATA SSD right now with a NVMe.

There's data out there.....

This is from Forposken, the NVME drive is 5x faster than the SATA III SSD yet the loading difference is only 2x.

Those speeds are even with decompression so it's actually a 2.4GB/s NVME drive (Funny how the raw and decompression speeds match the Series consoles SSD speeds perfectly )

Stop thinking you need some magical 14GB/s NVME drive, there's literally zero evidence that you require such a high speed drive and even Microsoft haven't confirmed a minimum speed for DS.

Even bargain basement NVME drives are twice as fast as that SATA III SDD in the Forspoken data.

If you feel you want to wait for benchmarks to come out before you make an NVME purchase that's fine, but one game isn't going to give you enough data to help you make that purchase, it'll be some years yet before there's enough DS enabled games out there for anyone to be able to say you need 'X' transfer speed at minimum.

Are you really going to wait that long?

If you spend $80 on a drive today (Like I have) and it's not fast enough then change the drive, they're one of the cheapest parts in a PC to upgrade and change.

I can't wait to get back home and fit my NVME drive, just having the drive means I can remove:

1. SATA III SSD
2. HDD cage
3. SATA cable
4. SATA power cable from my PSU

There are practical benefits to having NVME over SATA III.

 

[Seanspeed]

You seem to be ignoring the other scenarios presented in that what if we find out the $80 drive simply is not suitable compared to the $200 drive once we have actual data with DirectStorage games. The problem is the uncertainty at this point, buying a $200 drive could be a way overspend, buying a $80 drive could turn out to be not suitable. Right now it's just all theory of how these solutions will actual work in practice.

I mean, while there's guesswork involved, it's not exactly blind guesswork by any means, especially for at least semi-informed people paying attention.

There's very little reason to think that we're gonna need some kind of huge overhead in terms of SSD specs over what's in the Xbox Series storage setup. Those aren't using some proprietary or highly custom SSD tech, it's genuine off the shelf stuff, and the specs themselves are considerably low-to-mid end PCIe 3.0 in terms of capabilities. Given that this should basically be the baseline for multiplatform games, you shouldn't need anything drastically better than that in order to be fine for basically the whole generation.

Even if we're talking theoretical PS5 1st party titles that are somehow pushing that console's storage to its fullest, there would be some pretty simple ways to scale down data sizes to make them playable on something like the Xbox's SSD or some PC equivalent.

The idea that the baseline for an adequate SSD for DirectStorage would be some fairly expensive SSD really doesn't make any sense. Obviously I endorse the idea of 'dont buy anything right now if you have no need for it yet', but it's almost assuredly not going to be the case that people who do buy some middling PCIe 3.0 drive or whatever will find themselves out on their ass with an inadequate product once DirectStorage does roll around in games.

 

[PSman1700]

A 7gb/s nvme isnt even that expensive right now if you stay below 2tb. Its one of the cheapest parts of a system. I got my 1tb 980 pro for 130usd. Its raw speeds are faster than the PS5's, and when GPU decompression probably double that.

 

[Allandor]

I used to think that but I'm planning to replace by 4TB HDD with a big SATA SDD as soon as possible due the noise it makes. Currently my HDD spinning is the noisiest part of my PC when not gaming.

That's why I no longer have a HDD in my PC since a few years (it is so quite now). 2 SATA SSDs and 1 m.2 nvme are in my PC now. Problem is, even with that ~3,5TB it can run out of space. I uninstalled almost everything and now only installing what I really need. Now it is enough space (without all those steam & gog games that I had constantly on my HDD ^^).
But still, really big SSDs (like 4TB) are still much to expensive and there is no way to get around SATA right now. There are still only 1-2 m.2 slots on the new boards (a really expensive board has even 3) but m.2 drives tend to be smaller. There is also almost no application/game that really needs something faster than SATA SSDs right now and when games really start using SSDs (as a requirement) there should be enough headroom (because loaded package data should shrink) even for SATA SSDs in read speeds. E.g. even the matrix demo didn't use that much of data and if more bandwidth is needed, a PC can easily compensate that with more RAM and a larger buffer (and maybe a loading screen at the start of a game).
But what it really needs is a new SATA standard to increase speeds for SSDs not directly screwed to the mainboard. But ... well ... SATA is quite dead. Seems more like USB is taking that job in the future (but has also higher latencies).

A 7gb/s nvme isnt even that expensive right now if you stay below 2tb. Its one of the cheapest parts of a system. I got my 1tb 980 pro for 130usd. Its raw speeds are faster than the PS5's, and when GPU decompression probably double that.

The decompression units are more or less only there to relieve the CPU of that work. PCs nowadays have more than enough processing power to decompress the stuff. The bandwidth between CPU and main memory should also be more than enough (especially with DDR5 and PCIe 5&6 in future PCs). Yes, it is still an extra step that requires resources, but this gives consoles only a small head start as PCs should quite easily compensate that with raw power (which is not that efficient for sure).
I guess we start seeing the first examples of SSD "in mind" games on xbox when xb1(x) is no longer supported.
I really don't expect wonders. Getting graphics quality higher (well that's what the consoles are all about in the last generations) will be quite hard. Diminishing returns are really "visible" in the last years. And that has not only to do with the the support for the last generation. It get's just really hard to create such a visual difference like e.g. FarCry -> Crysis did more than 10 years ago.

 

[orangpelupa]

Btw does ssd goes into read only mode before it went kaput?

Or there's no standard of it and it depends on the controller and/or how the manufacturers set the firmware options?