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

Load times are incredibly fast on PC. Seemingly faster than PS5 with the right hardware. How much of that is down to DirectStorage and how much is simply down to good game design and fast hardware is anyones guess. It's certainly a great showing for the first DirectStorage game regardless though.
That's honestly what I am concerned about to be honest...
 
Yes, it's fast, but how much data is it loading? The scenes look pretty bland, she moves slowly, slows down even more before cutscenes, and often there's the strategic positioning of the camera at the start to fill some seconds while loading the rest.
How much is design and how much technology?
 
CPU plays a large factor in loading times once IO gets out of the way.
That's the shift that needs to happen, if it's not already. Rather than shove data together so that the CPU has to separate it out, or convert data so that it is ready to be used by the engine, devs need to be storing more data - potentially at the expensive of storage efficiency - in formats that you can load it into memory and use it without the need for CPU to do something - removing that CPU bottleneck.

For years, decades even, bottlenecks have shifted from one place to another. When I/O is slow and CPU is fast, it may be faster to compress data so it's loaded faster and decompresses on the CPU in realtime. When I/O is fast and CPU relatively slow, it may be faster to not compress/organise data in ways that requires the CPU to prep the data before it is ready to use.

This shift must already be happening on consoles because some PS5 games load in a few seconds and the CPU isn't that good.
 
Last edited by a moderator:
The ideal shift there is too expensive in storage to be viable - don't store assets to be constructed in RAM, but store the RAM dump of the game elements and just copy them into RAM. Hmmm, even if we didn't have finite storage capacity, the need to link elements across memory spaces is still going to be slow.
 
CPU plays a large factor in loading times once IO gets out of the way.
What I meant was I am concerned direct storage will not give same benefit in every title because of the engine limitation. That's the only way we are gonna see games widespread using it along with console solutions. but if it's not something that can just be used and get the benefit it's an issue.

Every dev should be making sure their pipeline is totally fit for the future hw architectures
 
Finally tried the demo and as I suspected before, the rendering distance for the grass and foliage is insane.

I've loaded up Days Gone, Witcher 3, Horizon:ZD, Elden Ring and none them draw grass out in to the distance any where near as far as what Forespoken does.

Don't get me wrong, it's actually nice to play a game with loads of dense foliage on screen and not see an obvious line 25 metres in front of me where the foliage render distance stops, but it comes at a cost as all that alpha is surely killing off memory bandwidth.
 
The ideal shift there is too expensive in storage to be viable - don't store assets to be constructed in RAM, but store the RAM dump of the game elements and just copy them into RAM. Hmmm, even if we didn't have finite storage capacity, the need to link elements across memory spaces is still going to be slow.
That isn’t actually the ideal shift — for a texture you might want every pixel (or, in reality, some compressed chunk of pixels) in memory so that you can access memory in an organized and performant way. However, you may have a much smaller on disk representation which you can decompress _faster_ than you could read the uncompressed texture in the first place. This is already not that rare for texture formats.
 
That isn’t actually the ideal shift — for a texture you might want every pixel (or, in reality, some compressed chunk of pixels) in memory so that you can access memory in an organized and performant way. However, you may have a much smaller on disk representation which you can decompress _faster_ than you could read the uncompressed texture in the first place. This is already not that rare for texture formats.
That's just textures and not the problem in loading times as I understand it. If you have an Orc model and textures, you load these once and then repeatedly instantiate 10,000 orc objects into your game world - that's the slow bit. Alternatively you could already create the world in RAM, and then dump that to storage. Then the load avoids all the object-building and can operate as fast as your storage can.
 
I thought this was worth posting in here!

 
Can someone try with a SATA SSD, someone complain about long loading time with SATA Evo 870. After maybe this is just a few seconds but seeing instantaneous it is the placebo effect. Someone try with the HDD, this is a nightmare for example open the map take 30 seconds because the game is optimized for load data just in time.

 
The game Forespoken is quite weird from what I saw of streamers running on a 64 GB RAM system and 1 TB NVMEs with midrange GPU of 6700 XTs. Many spots where the game fades to black, loads the cutscene and plays it, fades to black, loads the game world. Those fade to black sections were 2 to 3 seconds in length at the quickest. It looked very much like something just wasn't working right. Nearly the entire game should be in memory already with that much main ram.
 
The game Forespoken is quite weird from what I saw of streamers running on a 64 GB RAM system and 1 TB NVMEs with midrange GPU of 6700 XTs. Many spots where the game fades to black, loads the cutscene and plays it, fades to black, loads the game world. Those fade to black sections were 2 to 3 seconds in length at the quickest. It looked very much like something just wasn't working right. Nearly the entire game should be in memory already with that much main ram.

I suppose they didn't change how the game is coded compared to console. This is like the guy who tried to run it on HDD, each time he load the map it takes 30 seconds to load it. A better solution would be to keep the map in RAM but they didn't want to develop two different version.
 
Can someone try with a SATA SSD, someone complain about long loading time with SATA Evo 870. After maybe this is just a few seconds but seeing instantaneous it is the placebo effect. Someone try with the HDD, this is a nightmare for example open the map take 30 seconds because the game is optimized for load data just in time.

Here's a video of the demo. Not my video, btw, but a SATA3 7200 RPM HDD and a SATA3 SSD.
 
The game Forespoken is quite weird from what I saw of streamers running on a 64 GB RAM system and 1 TB NVMEs with midrange GPU of 6700 XTs. Many spots where the game fades to black, loads the cutscene and plays it, fades to black, loads the game world. Those fade to black sections were 2 to 3 seconds in length at the quickest. It looked very much like something just wasn't working right. Nearly the entire game should be in memory already with that much main ram.
There are a lot of weird behaviors in this demo that are still unaccounted for. The game reads through 65GB in a few minutes while standing still, for example.


I thought it might be loading/evicting data in that scene so I set up a RAM cache to see how large it was. But I had to stop at 24GB because it seems to read through the entire 40GB data files over and over.
 
Here's a video of the demo. Not my video, btw, but a SATA3 7200 RPM HDD and a SATA3 SSD.

Probably a problem with the guy config but I am sure if Forspoken was tailored around HDD it would load faster here they probably do like with SSD and load everything in RAM at once.
 
The game Forespoken is quite weird from what I saw of streamers running on a 64 GB RAM system and 1 TB NVMEs with midrange GPU of 6700 XTs. Many spots where the game fades to black, loads the cutscene and plays it, fades to black, loads the game world. Those fade to black sections were 2 to 3 seconds in length at the quickest. It looked very much like something just wasn't working right. Nearly the entire game should be in memory already with that much main ram.
Only if the developers programmed the game to cache as much RAM as possible which is something I wish games did on PC as so much RAM is just sat there doing nothing.

Message to developers..............If my RAM is there doing nothing then please use it.
 
There are a lot of weird behaviors in this demo that are still unaccounted for. The game reads through 65GB in a few minutes while standing still, for example.


I thought it might be loading/evicting data in that scene so I set up a RAM cache to see how large it was. But I had to stop at 24GB because it seems to read through the entire 40GB data files over and over.
I have also seen this behaviour in the final game. Just sitting still seeing a constant 500 mb/s on the NVme for no reason. This happens for mintues on end. This screenshot was taken with me having sat still here for about 4 minutes and it was doing 500-530 mb/s the entire time.

ssd.00_02_02_25.stillrrcof.png
 
Back
Top