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

2) What is market penetration goign to be like until that market becomes significant for devs not to factor in weaker IO performance in their game development decisions?
People keep framing this question in terms of developers waiting for the install base for NVMe SSD's to be big enough, but it wont, and cant work like that. As usual, developers have to be the ones leading on this. They have to demonstrate the advantages of getting this better hardware, cuz gamers aren't just gonna all upgrade preemptively for no immediate benefit. Some might, but the market as a whole wont.

But we know that once tangible and notable advantages are demonstrated in available products, that gamers absolutely will start upgrading. A sufficiently fast NVMe drive isn't even expensive or anything. I'd expect that PC users will probably want something at least a bit beyond what's in the Xbox, and the usual overheads for CPU and memory, but nothing outrageous. It shouldn't be a big issue for people to go out and get an NVMe drive in order to play the latest games properly. Cuz I promise that PC gamers will be very interested in doing so once they see what true next gen games are actually like.

Besides, multiplatform devs are gonna have to make the choice to do so soon enough, or just cut off any PC version altogether. Cuz I dont think there's any world in which they gimp their games on console simply cuz PC users cant take advantage of things. For AAA, consoles are still the big money makers.
 
There are probably just as many pc's with nvme drives out there as there are all current gen consoles combined
It doesnt matter. That is still a huge problem. Because you also have countless of users that have mechanical hardrives too. You cant extapolate which NVME and which HDD users are your potential customers. The developers can optimise the NVME drives for better loading times and faster streaming of higher LOD assets. That wont change much the experience for HDD users. They can still run the game, play it and have the same experience more or less.
But for functions that are very specific to innovative NVME utilization like the ones described by Mark Cerny, the developers may eithe have to exclude HDD users that are potential customers by requiring NVME as a mandatory function (basically making the game unplayable/incompatible for the former), or just leave out the new opportunities brought by the technology and just use it for better loading times and streaming which is what we have been getting mostly. The latter is a safer bet in terms of sales.
 
All NVMe are "high performance" relatively speaking. A 2.5GB/s NVMe drive (i.e. PCIe 3.0 speeds) gets you into "optimised territory" for Direct Storage.

Then why do Microsoft say "With a high performance NVMe SSD and the proper drivers, Windows 11 can soon load new games faster than ever thanks to a breakthrough technology called DirectStorage"? If all NVMe drives are sufficiently high performance, just say NVMe. Why use language that feel like a qualifier and introduces doubt?
 
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 agreed, I don't think DS itself will remove the copy to system RAM any time soon as many hardware platforms simply don't support that. And DS relies on a very wide compatibility range (basically any modern PC can use it, if not take full advantage of it). What it will do though is remove the copy on AMD systems that are Smart Access Storage enabled (and possibly RTX-Io compatible systems too). The only question is whether this will be automatic or require specific dev integration. I hope for the former but expect the latter.

But for functions that are very specific to innovative NVME utilization like the ones described by Mark Cerny, the developers may either have to exclude HDD users.

This is exactly what will happen (Nanite requires an SSD) and I don't think it will have a significant impact on the size of the addressable market to be honest. How many PC gamers who are interested in playing the latest top end games today don't already have an SSD available in their system? And of those very few that don't, how many would be unwilling to pick up a cheap one? I imagine the number will be negligible. Making NVMe a requirement might be a more interesting question but I'm honestly pretty dubious if there is anything that could be done on the PS5 that couldn't be scaled down in a reasonable way to run on a SATA SSD on the PC - especially with more RAM as backup.

Then why do Microsoft say "With a high performance NVMe SSD and the proper drivers, Windows 11 can soon load new games faster than ever thanks to a breakthrough technology called DirectStorage"? If all NVMe drives are sufficiently high performance, just say NVMe. Why use language that feel like a qualifier and introduces doubt?

I'm not sure what the purpose of this conversation is. I've already linked to a Microsoft architect specifically saying that 1. Direct Storage supports all drive types, and 2. It needs an NVMe to take advantage of the improvements brought by DS with a speed of 2.5GB/s taking you into the "optimised category". So we already have the answer in black and white (a video actually); 2.5GB/s NVMe is what you need to take full advantage of Direct Storage, but everything below it is still supported. Clearly, Microsoft consider at least 2.5GB/s to be a "high performance SSD". And why would they say otherwise, it's what their own consoles are using!
 
Then why do Microsoft say "With a high performance NVMe SSD and the proper drivers, Windows 11 can soon load new games faster than ever thanks to a breakthrough technology called DirectStorage"? If all NVMe drives are sufficiently high performance, just say NVMe. Why use language that feel like a qualifier and introduces doubt?
Because it's a throwaway adjective? You're probably ascribing FAR more deliberation into their wording than actually went into it.
 
Then why do Microsoft say "With a high performance NVMe SSD and the proper drivers, Windows 11 can soon load new games faster than ever thanks to a breakthrough technology called DirectStorage"? If all NVMe drives are sufficiently high performance, just say NVMe. Why use language that feel like a qualifier and introduces doubt?
Because NVMe is inherently 'high performance' versus SATA SSD etc, I think it's just an ambiguous statement. As PR speak, there's as much pointless flash added as possible as opposed to a technical paper. Compare "This means you’ll get to experience incredibly detailed game worlds rendered at lightning speeds" and "We build Windows to be a magical place" - it's just not technically grounded. I don't think 'any old NVMe' conflicts with what's said so far.
 
Yes agreed, I don't think DS itself will remove the copy to system RAM any time soon as many hardware platforms simply don't support that. And DS relies on a very wide compatibility range (basically any modern PC can use it, if not take full advantage of it). What it will do though is remove the copy on AMD systems that are Smart Access Storage enabled (and possibly RTX-Io compatible systems too). The only question is whether this will be automatic or require specific dev integration. I hope for the former but expect the latter.
Right, so this is the technical discussion I'm interested in. ;) When does the PC achieve the immediacy of the console storage solution (in sufficient numbers to be a viable market, so probably about the same number of PCs of adequate spec as one of the consoles) and will that immediacy every really be needed anyway? In a way, the Matrix Demo showed not, as it never relied on PS5's IO stack to its fullest. But then you have to wonder what using that stack to its fullest actually allows, if anything. But again, I don't want a Console SSD Storage discussion and that's not the emphasis even though many are focussing on that part of my original statement - I just want the technical discussion about the features the consoles have and when these are coming to PC. As I understand it, the hardware requirements are satisfied from 2018-2020 tech. Ergo, would my Ryzen 3900X on a B550 motherboard just be waiting on software? Or would that need an upgrade? And if the latter, how many PCs are hardware ready?
 
It doesnt matter. That is still a huge problem. Because you also have countless of users that have mechanical hardrives too. You cant extapolate which NVME and which HDD users are your potential customers. The developers can optimise the NVME drives for better loading times and faster streaming of higher LOD assets. That wont change much the experience for HDD users. They can still run the game, play it and have the same experience more or less.
But for functions that are very specific to innovative NVME utilization like the ones described by Mark Cerny, the developers may eithe have to exclude HDD users that are potential customers by requiring NVME as a mandatory function (basically making the game unplayable/incompatible for the former), or just leave out the new opportunities brought by the technology and just use it for better loading times and streaming which is what we have been getting mostly. The latter is a safer bet in terms of sales.
If devs aren't gonna properly utilize them this generation outside of 1st party PS5 titles or something, then the whole effort will have been a MASSIVE lost opportunity. It will also mean that we can forget about having PS5 titles come to PC.

Again, as I say in a post right above, all developers have to do is demonstrate the 'next gen' advantages of what this enables and PC gamers will upgrade. NVMe drives aren't expensive products that we cant reasonably expect people to get. You dont even need a large or cutting edge one. A $60 500GB middling 3.0 drive will be fine! Prices will continue to drop as well.

I mean, I feel like we've lost all perspective of where the PC is supposed to be in this market. It's not supposed to be the one dragging its feet and holding things back because users are unwilling to upgrade their hardware. What a sorry situation that would be where console users are complaining that PC gamers are holding things back because they still expect to run everything on their HDD's.
 
I don't know if we can say all NVMe drives are "higher performance." There are some rather low performing NVMe drives especially if you look into cases outside more idealized clean synthetic test situations. Not just some rare off brand drives either. For example the Crucial P2 500GB, especially with the new QLC rework, I'm pretty sure will load existing games slower than SATA drives like Crucials own MX500 going by the FFXIV benchmark load test results.

Not to mention my general concern of what happens once data remotely ages at all and ECC has to kick in due to QLC. Which no review really tests for (not really practical to do so).

I think people tend to conflate peak rates with sustained rates. When most people see PS5 having a data rate of 5.5GB/s then they often seem to expect 5.5GB/s to be the sustained rate and I don’t really believe that.

But 550MB read in 0.1s is also a read rate of 5.5GB/s. And I think we’ll see a lot more of those scenarios. The main question is how slower SSDs will handle that since 0.1s vs 0.2s isn’t really a whole lot of perceptible difference.

A single frame at 60 fps is only ~0.017s. 30 fps is ~0.034s. 15 fps is ~0.068s. Yet people will say that 60 to 30 fps much less 15 fps drops for even a single frame are very noticeable and even 1 frame of delay for something like input lag at 60 fps is rather noticeable.

I don't think any of that matters much when looking for absolute lowest bounds. Any plain SATA SSD should be capable of hitting 500 MB/s. The maximum theoretical speed of SATA 2 is around 600 MB/s. Yes, it's barely anything at all but there can be so much done with that as most of the drastic improvements is around latency. I think this should serve as the absolute lowest bounds instead of supporting HDDs for games.

Yes, Things get interesting and more varied once you get into Gen 3 NVME M.2, where the ceiling is around 3500 MB/s and even lower latency. Then you have even higher tiers.

The jump from HDD to SSD (SATA) is drastic enough to make a world of difference. It's the going from 20 MB/s to 500 MB/s that gives an incredible 25x lift. Anything above that is less measurable, 7x (Gen 3 NVME) to 11x (PS5 NVME).

I don't think that is really the case. In terms of sustained transfer rates HDDs are much faster than 20 MB/s (typically over 100 mb/s even for 5400 rpm drives). The huge gain is from the several orders magnitude better access latency (something like 10ms vs. 0.1ms) and 4k read speeds (something like 0.3mb/s vs 30 mb/s for low end SATA SSDs). This is also a reason why NVMe drives did not provide the same boost in typical usage (and gaming), while sequential speeds and high QD random speeds have increased something like 10x now over SATA, latency and burst 4k have only 2x or so.

Supposedly, although I might wrong with this, one of the aspects of DirectStorage was to change the type of I/O in games to basically effectively enable higher QD and more sustained workloads, which in theory if implemented would then be able to leverage the actual hardware differences of todays faster NVMe SSDs over SATA SSDs.
 
Again, as I say in a post right above, all developers have to do is demonstrate the 'next gen' advantages of what this enables and PC gamers will upgrade. NVMe drives aren't expensive products that we cant reasonably expect people to get. You dont even need a large or cutting edge one. A $60 500GB middling 3.0 drive will be fine! Prices will continue to drop as well.

I'm just going to offer a different perspective on this. I don't fall into this camp but you'd be surprised at how many people feel they need to "all" their games installed all the time and don't want to manage and swap drives. This means from their perspective they aren't comparing a $60 NVMe drive vs. $60 HDD, but say a 2TB HDD vs. 2TB NVMe drive (or even higher capacities).

For people that only play a couple if not 1 game at a time, such as myself, the bar of entry is now low price wise. But not for the above.
 
It doesnt matter. That is still a huge problem. Because you also have countless of users that have mechanical hardrives too. You cant extapolate which NVME and which HDD users are your potential customers. The developers can optimise the NVME drives for better loading times and faster streaming of higher LOD assets. That wont change much the experience for HDD users. They can still run the game, play it and have the same experience more or less.
But for functions that are very specific to innovative NVME utilization like the ones described by Mark Cerny, the developers may eithe have to exclude HDD users that are potential customers by requiring NVME as a mandatory function (basically making the game unplayable/incompatible for the former), or just leave out the new opportunities brought by the technology and just use it for better loading times and streaming which is what we have been getting mostly. The latter is a safer bet in terms of sales.

Games designed around current generation of hardware (2018 and beyond) arent that intresting to users who are still rocking mechanical hdd's for gaming. They are 12+ years behind the curve.
Just like today with Spiderman, they can just put ssd's in the hardware requirements recommendations.

Right, so this is the technical discussion I'm interested in. ;) When does the PC achieve the immediacy of the console storage solution (in sufficient numbers to be a viable market, so probably about the same number of PCs of adequate spec as one of the consoles) and will that immediacy every really be needed anyway? In a way, the Matrix Demo showed not, as it never relied on PS5's IO stack to its fullest. But then you have to wonder what using that stack to its fullest actually allows, if anything. But again, I don't want a Console SSD Storage discussion and that's not the emphasis even though many are focussing on that part of my original statement - I just want the technical discussion about the features the consoles have and when these are coming to PC. As I understand it, the hardware requirements are satisfied from 2018-2020 tech. Ergo, would my Ryzen 3900X on a B550 motherboard just be waiting on software? Or would that need an upgrade? And if the latter, how many PCs are hardware ready?

We can try and interpolate that from statistics perhaps? We know that there are atleast as many PS5/XSX matching and exceeding GPU's sold in the discrete GPU space (RTX2060S and up) as the consoles have sold. Thats not counting mobile (laptop) gpu's. We can gurantee that none of those systems are running their games on mechanical hdd's. I assume most if not all of them some sort of Nvme m2 drive. Already in 2020 it was impossible to source a prebuild system without nvme drives installed.

Im also (still) on a 3900X, RTX2080Ti, but on a X570 mb which supports PCIe4.0 (pcie4 m2 installed). This hardware should support DS in full i think. B550 if im correct supports PCIe4.0 for the CPU/GPU but M2 slots are PCIe3.0. Should still suffice for DS and any current gen game though.

Another thing to note, PCIe3 drives might support 'only' 3.5gb/s, though could and probably will reach much higher than that when decompression is done using the GPU.
 
I'm just going to offer a different perspective on this. I don't fall into this camp but you'd be surprised at how many people feel they need to "all" their games installed all the time and don't want to manage and swap drives. This means from their perspective they aren't comparing a $60 NVMe drive vs. $60 HDD, but say a 2TB HDD vs. 2TB NVMe drive (or even higher capacities).

For people that only play a couple if not 1 game at a time, such as myself, the bar of entry is now low price wise. But not for the above.

I'm the same, I have a 480Gb SSD and store the rest on an external HDD.

Failing that I'm on a Gigabit internet connection so download games takes mere minutes, aside from COD....that still takes me 4 days :runaway:
 
Because NVMe is inherently 'high performance' versus SATA SSD etc, I think it's just an ambiguous statement.

Remember that NVMe is just a bus protocol. It doesn't mean you can't some some slow-arse solid state, or even a tape drive connected on the other end. You may have missed it, and I wouldn't be surprised, but Seagate make a HDD NVMe drive, with an deliverable performance lower than SATA's bandwidth. This may not even be the slowest/stupidest thing out there, just the one I particularly remember. ¯\_(ツ)_/¯
 
Remember that NVMe is just a bus protocol. It doesn't mean you can't some some slow-arse solid state, or even a tape drive connected on the other end. You may have missed it, and I wouldn't be surprised, but Seagate make a HDD NVMe drive, with an deliverable performance lower than SATA's bandwidth. This may not even be the slowest/stupidest thing out there, just the one I particularly remember. ¯\_(ツ)_/¯
Well I think your clarification highlights the communication objective - I as an ignorant consumer think NVMe means a small circuit-board flash-storage module with fast, low latency storage. I expect the copywriter thought the same. Those words quoted just really aren't technical at all and you need to go to MS's technical sources to understand the DS requirements. But the waters are also muddied in the same way NVMe is a bus protocol in that DS is a storage protocol and API which doesn't necessitate minimum performance metrics. You can run DS on an NVMe floppy drive. What we need to be clear on is the subset of NVMe devices in a DS enabled system that provide adequate IO for whatever our target performance is. Though in the absence of a real-world case maxxing out the console IO, I guess this conversation has really come to a head. High performance IO to a degree exists now, in a significant number of gaming PCs. Whether it's enough to go toe-to-toe with consoles can't really be known until we have a game that's dependent on the lowest latencies consoles provide. Until then, the court is out on whether the consoles over-engineered their IO systems or not.
 
Right, so this is the technical discussion I'm interested in. ;) When does the PC achieve the immediacy of the console storage solution (in sufficient numbers to be a viable market, so probably about the same number of PCs of adequate spec as one of the consoles) and will that immediacy every really be needed anyway?

I think we probably need to break that question down to "is it of advantage?" - yes because otherwise something like Smart Access Storage wouldn't have been developed. And "is it needed". Then the is it needed can be further broke down into "for parity with consoles" and "to enable console games that take advantage of it on the consoles, to be ported to PC". The last of those 2 seems like a clear no to me. The additional latency of a copy to and from system memory should be tiny compared to the latency of reading from the SSD in the first place and system memory doesn't introduce a bandwidth bottleneck either. I guess it probably removes more load from the CPU (less copies), which is definitely nice to have, but with an already fairly low overhead via "standard" Direct Storage (with GPU decompression), and obviously much faster CPU's, then it would be hard to argue it's needed even for parity with consoles - albeit probably needing a slightly faster CPU to make up the difference.

I just want the technical discussion about the features the consoles have and when these are coming to PC. As I understand it, the hardware requirements are satisfied from 2018-2020 tech. Ergo, would my Ryzen 3900X on a B550 motherboard just be waiting on software? Or would that need an upgrade? And if the latter, how many PCs are hardware ready?

I'm obviously no expert but with my limited knowledge I'd broadly break it down something like this:

  • A 5.5GB/s NVMe SSD on a PCIe 4.0 4x interface - this one is already available and we can take a guess at the market penetration
  • Offloading of the decompression work from the CPU - Direct Storage will take care of the GPU data portion of this, but not as I understand the CPU data portion. So the PC CPU will still have to deal with an extra overhead here but it will be significantly reduced compared with today. This should work on any DX12 enabled system (OS and GPU) so over 90% of the PC gaming market.
  • An API that reduces the file IO overhead - Direct Storage is already available for this, although it doesn't seem to be unreasonable to expect the PS5 API to be even more efficient. Past a certain point though this probably isn't too relevant. Forespoken benchmarks this as 3% of CPU usage using Direct Storage for example. Again over 90% of PC gaming market.
  • Reduce unneeded data copies between the SSD and final memory destination. This is the unclear one as it looks like Smart Access Storage is probably doing this by enabling a direct peer to peer route between the SSD and VRAM (obviously the route to system RAM is already direct). If that's the case, and if RTX-IO is also doing the same then it really comes down to what platforms that tech is supported on. This is likely to be the biggest sticking point for across the board functional parity. For example for SAS you need an all AMD system and that might be locked into just the 7xxx series CPU/GPU's. Hopefully it will extend back further than that though. Nvidia would seem to have an even more difficult compatibility path to navigate.
 
Remember that NVMe is just a bus protocol. It doesn't mean you can't some some slow-arse solid state, or even a tape drive connected on the other end. You may have missed it, and I wouldn't be surprised, but Seagate make a HDD NVMe drive, with an deliverable performance lower than SATA's bandwidth. This may not even be the slowest/stupidest thing out there, just the one I particularly remember. ¯\_(ツ)_/¯

you previous quote form msft
"With a high performance NVMe SSD and the proper drivers..."

I think you are trying too hard, at this point its should be clear what DS is and what is trying to solve. It may work with basically anything sdd,hdd.floppy etc but this is not why it was created.
 
Those words quoted just really aren't technical at all and you need to go to MS's technical sources to understand the DS requirements. But the waters are also muddied in the same way NVMe is a bus protocol in that DS is a storage protocol and API which doesn't necessitate minimum performance metrics.
You might be interested in Microsoft's own technical overview of DirectStorage on Xbox Series consoles - where they control the software and hardware stack top-to-bottom. Whilst Xbox Series's architecture is considerably different from your average PC, there are many references to the situation on Windows and you can infer where the additional challenges are with reaching I/O parity with consoles by factoring in the differences of the PC architecture. On a Windows PC, the APIs will generally have less direct control over several hardware features like PCI controllers and the degree for which its possible for the OS to micromanage I/O across the bus and when read/writem to the drive. Assuming a NVMHCIS connected M2, which is the best case scenario for having solid state storage as immediately addressable by the CPU, most drive controllers will abstract the I/O from the host operating system - if only to implement wear-levelling and other performance enhancing techniques - and generally cannot be overridden.

The controllers in both consoles - I believe - are completely subordinate. This is the case on PS5 and I assume its the case on Xbox Series as well.
 
I think you are trying too hard, at this point its should be clear what DS is and what is trying to solve. It may work with basically anything sdd,hdd.floppy etc but this is not why it was created.
I think a great chunk of this thread demonstrates it's not clear. As Shifty said above (his ignorant consumer comment) much of this is abut what a person's understanding of something maybe be. I don't know why DirectStorage wouldn't work with any type of drive, it's just a different way of managing I/O, but it working and it delivering measurable benefits are two very different things.

Microsoft proffered DirectStorage in the context of Xbox Series I/O and I think many would expect performance somewhere in that ballpark. The $64,000,000 question is what combinations of hardware is required to reach that - assuming games begin to adopt the APIs. As we've just seen, Marvel's Spider-Man, released a week ago, does not. Would it's loading times be closer to PS5 if it did? Unless they patch it in, we'll never know.
 
Last edited by a moderator:
Back
Top