Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
Will the next generation of consoles bring consumer benefits to using NVMe rather than SATA?
As you probably know, Sony and Microsoft are due to launch new game consoles later this year. In a lot of ways, the PlayStation 5 and Xbox $\sum x$ (I refuse to use Microsoft's official name because it's stupid) are just beefier versions of the PlayStation 4 Pro and Xbox One X, which are also very similar to each other.
One critical difference is the storage system. NAND flash has gotten cheap enough that it's viable to use it as your main storage in game consoles, which it wasn't in previous generations. As such, both consoles will use NVMe SSDs. The PS5 will have a 0.75 TB SSD (no, Sony, I'm not willing to call 3 * 2^37 bytes "825 GB"), while the next Xbox will have a 1 TB SSD.
As the cost of the SSD is mostly in the NAND flash, not the controller, going with NVMe makes a lot of sense. AMD will put however many PCI-E lanes the vendor wants in the SoC, so that's not a major restriction. SSD controller vendors may tend to charge more for a higher end controller, but SSD controllers aren't that expensive to build on a per unit basis, and Microsoft and Sony have sufficient volume to negotiate better deals. For that matter, Sony is getting a custom SSD built specifically for the PS5.
Thus far in the consumer space, NVMe SSDs haven't been much better than SATA. They're much faster in synthetic benchmarks, but that doesn't translate to much of a real-world advantage. Usually even a SATA SSD is plenty fast enough that something else is the bottleneck, so getting an SSD that is faster yet doesn't help.
That's largely because consumer applications pretty much have to be usable on a hard drive, as much of the potential market only has a hard drive and not an SSD. The problem is that hard drives are massively slower than SSDs. A SATA SSD might offer 100k 4K random read IOPS, while an NVMe one might offer 500k IOPS. But a hard drive might offer about 200 (not thousand), and if 100k is slow enough to feel a little sluggish, the program is going to be unusable on a hard drive.
If developers continue to structure games the way they have in the past, with an eye toward making sure that it's playable on a hard drive, then NVMe SSDs still won't offer much benefit over SATA. But console games won't need to do that anymore, as they'll know that everyone with a PS5 or the next Xbox has a very fast SSD. Microsoft is claiming 2.4 GB/s sequential reads, while Sony is claiming 5.5 GB/s. For comparison, the theoretical cap on SATA is 0.6 GB/s. Microsoft's SSD is basically a budget NVMe one, while Sony's would be the fastest on the market at sequential reads if it were to launch today, but is unlikely to still be by the time the PS5 launches, as all of the major SSD controller vendors are working on a PCI Express 4.0 controller that should launch soon.
A lot of PCs still won't have an SSD, of course. But if a console game has done the work to optimize for loading things from an SSD, why not still use that at least as an option when making a PC version of the game? For that matter, it wouldn't be that surprising if we start seeing PC versions of console games that require an SSD, and possibly even an NVMe SSD, in the system requirements.
The real question is what consoles can do to take advantage of the extra speed. If they continue to structure loading games about how they do now, then the NVMe SSD won't offer much benefit over SATA. Just how much benefit they can squeeze out of NVMe remains to be seen, as it will probably involve structuring how games are loaded differently.
For example, suppose that by some miracle, it suddenly became tremendously cheap to get enormous amounts of DRAM. Nothing else about PCs changed, but game developers could suddenly assume that all of their customers had many terabytes of system memory and also of graphics memory. If they were to continue loading games how they do now, that wouldn't offer any benefit. They wouldn't even be able to jump to super high resolution textures, as that would massively bloat the game installation size.
But if you had many terabytes of RAM, there would be no need for zoning. You could just load the whole game into RAM, fully decompress everything, and have it sitting there when you need it. No need to unload one zone to free up memory for the next to use. Just leave everything sitting in RAM all the time until the game closes entirely.
SSDs in consoles aren't going to have quite that dramatic of an effect. But they might mean a lot shorter loading screens, a lot fewer of them, or both. We'll have to see what developers can come up with. People who insist that they really want a seamless world should be excited about this, as SSDs being used universally is the key technology to make it practical without having some severe drawbacks.
One critical difference is the storage system. NAND flash has gotten cheap enough that it's viable to use it as your main storage in game consoles, which it wasn't in previous generations. As such, both consoles will use NVMe SSDs. The PS5 will have a 0.75 TB SSD (no, Sony, I'm not willing to call 3 * 2^37 bytes "825 GB"), while the next Xbox will have a 1 TB SSD.
As the cost of the SSD is mostly in the NAND flash, not the controller, going with NVMe makes a lot of sense. AMD will put however many PCI-E lanes the vendor wants in the SoC, so that's not a major restriction. SSD controller vendors may tend to charge more for a higher end controller, but SSD controllers aren't that expensive to build on a per unit basis, and Microsoft and Sony have sufficient volume to negotiate better deals. For that matter, Sony is getting a custom SSD built specifically for the PS5.
Thus far in the consumer space, NVMe SSDs haven't been much better than SATA. They're much faster in synthetic benchmarks, but that doesn't translate to much of a real-world advantage. Usually even a SATA SSD is plenty fast enough that something else is the bottleneck, so getting an SSD that is faster yet doesn't help.
That's largely because consumer applications pretty much have to be usable on a hard drive, as much of the potential market only has a hard drive and not an SSD. The problem is that hard drives are massively slower than SSDs. A SATA SSD might offer 100k 4K random read IOPS, while an NVMe one might offer 500k IOPS. But a hard drive might offer about 200 (not thousand), and if 100k is slow enough to feel a little sluggish, the program is going to be unusable on a hard drive.
If developers continue to structure games the way they have in the past, with an eye toward making sure that it's playable on a hard drive, then NVMe SSDs still won't offer much benefit over SATA. But console games won't need to do that anymore, as they'll know that everyone with a PS5 or the next Xbox has a very fast SSD. Microsoft is claiming 2.4 GB/s sequential reads, while Sony is claiming 5.5 GB/s. For comparison, the theoretical cap on SATA is 0.6 GB/s. Microsoft's SSD is basically a budget NVMe one, while Sony's would be the fastest on the market at sequential reads if it were to launch today, but is unlikely to still be by the time the PS5 launches, as all of the major SSD controller vendors are working on a PCI Express 4.0 controller that should launch soon.
A lot of PCs still won't have an SSD, of course. But if a console game has done the work to optimize for loading things from an SSD, why not still use that at least as an option when making a PC version of the game? For that matter, it wouldn't be that surprising if we start seeing PC versions of console games that require an SSD, and possibly even an NVMe SSD, in the system requirements.
The real question is what consoles can do to take advantage of the extra speed. If they continue to structure loading games about how they do now, then the NVMe SSD won't offer much benefit over SATA. Just how much benefit they can squeeze out of NVMe remains to be seen, as it will probably involve structuring how games are loaded differently.
For example, suppose that by some miracle, it suddenly became tremendously cheap to get enormous amounts of DRAM. Nothing else about PCs changed, but game developers could suddenly assume that all of their customers had many terabytes of system memory and also of graphics memory. If they were to continue loading games how they do now, that wouldn't offer any benefit. They wouldn't even be able to jump to super high resolution textures, as that would massively bloat the game installation size.
But if you had many terabytes of RAM, there would be no need for zoning. You could just load the whole game into RAM, fully decompress everything, and have it sitting there when you need it. No need to unload one zone to free up memory for the next to use. Just leave everything sitting in RAM all the time until the game closes entirely.
SSDs in consoles aren't going to have quite that dramatic of an effect. But they might mean a lot shorter loading screens, a lot fewer of them, or both. We'll have to see what developers can come up with. People who insist that they really want a seamless world should be excited about this, as SSDs being used universally is the key technology to make it practical without having some severe drawbacks.
Comments
“Microtransactions? In a single player role-playing game? Are you nuts?”
― CD PROJEKT RED
NVMe is an interface, not a form factor. You can actually have that interface with either PCI Express or SATA as the protocol. The usual form factor is M.2 2280, though there is also M.2 2240, M.2 22110, U.2, and others. The choice of the form factor (probably M.2 2280, though I'm not sure if this has been confirmed) is surely due to size as you said, as the 2.5" form factor was driven by the need to fit existing hardware by matching laptop hard drive sizes, and wasn't something you'd do if hard drives had never existed.
But it's still interesting that both vendors went with NVMe over PCI Express rather than over SATA. That's what my analysis is geared toward explaining.
May be a reason to go for the NVMe now that we are not aware of, that offsets the extra expense. But i thought that even though they are getting better...there is still a extra price to pay for NVMe over SATA.
But even if you're going NVMe over PCI Express, some ways to build it are more expensive than others. The next Xbox seems to have gone the cheap route, likely using a simple PCI Express 3.0 x4 SSD with QLC NAND and possibly only a 4-channel controller. The PS5 is probably using a more expensive PCI Express 4.0 x4 connection, and is known to use a custom 12-channel SSD controller. That adds cost, and Sony apparently thought it was worthwhile to get a relatively high end SSD, while Microsoft went with a cheaper one.
You've misunderstood something. Most AAA games can't be loaded entirely into RAM.
https://www.newegg.com/g-skill-256gb-288-pin-ddr4-sdram/p/N82E16820232945
Of course, that won't reduce decompression times, and you'd have to do it all over again every time you reboot.
Will this trickle down to PC games requiring an SSD, preferably an NvME? Probably, depending on the game. A game designed for the Xbox sucky-name should not be too difficult to port to Win10 - I say that from the mouth of a guy that has very little clue what is involved in programming, just going by what I have read.
The world is going to the dogs, which is just how I planned it!