Current Generation Games Analysis Technical Discussion [2020-2021] [XBSX|S, PS5, PC]
- Thread starter Shortbread
- Start date
- Status
If all you are doing is moving data from the SSD to system memory, and that data is properly configured to take advantage of the interface and compression hardware, then yeah. Well, maybe. But when you are running a game, it's not that simple. Just because you can move 5GB of models and texture into main memory doesn't mean they don't have to be placed into the game state. That stuff isn't handled by the I/O stack. Think of the I/O stack like a dump truck hauling things. If you haul some pre-fab'd concrete slabs, you can drop them off and make a concrete platform quick. If you are hauling crushed stone, you can move it just as fast but it's going to require some work to make anything except a pile of stones when it gets to it's destination.
Some context from someone who knows a bit about this (large twitter thread):
Silent_Buddha
Legend
Pretty sure there is no shipping game that is utilizing SFS nor the Velocity Architecture, yet.
So, I/O on XBS systems are still mostly just brute force loading assets from the drive.
Regards,
SB
Random thought/question:
How will SFS affect Quick Resume?
Presumably, if you're playing a game that utilizes SFS and then exit that game, whatever is in memory gets caches to the SSD. But all the texture data that's getting cached will be in bits and pieces since that's what SFS does. Is Quick Resume able to just load that cached data and it picks up where it left off or are they going to have to load the full texture data in addition to what is cached so that SFS can kick in again since it needs rendered frames to work?
That got me interested, so I had a look.
Heres my game from a cold boot, I think its mostly IO (Of course its gonna be a combination of everything)
I think I can knock a couple of seconds off it, I'm gonna look more into it. I do have a faster SSD so its prolly gonna be slower on other PCs (and quicker on others), none of my PCs have a HDD so can't see what happens there, I suppose you can't emulate a HDD with a SDD?
Last edited:
- The game runs in backward compatibility mode on all 3 consoles.
- On S Series, the maximum resolution has been increased from 900p to 1080p (on Xbox One it was limited to 900p).
- Series X and PS5 use dynamic resolutions just like PS4 Pro and One X. The resolution in Series X usually reaches 2160p, while PS5 reaches 1620p.
- Series X has a more unstable framerate due to resolution. At times of greatest stress, it can drop 10FPS below the PS5/Series S.
- Loading times are faster on Xbox Series as is always the case with backward compatible titles.
XSX and ps5 are using dynamic res scaling XSX (1692-2160p) and PS5 (1242-1620p). During normal gameplay fps seems to be very solid on all platforms.
During the heavy battle scenes all consoles drops frames with xsx dropping noticeably lower than ps5 (low 50).
Those loading times are really irritating on PS5. So I guess the cuts to the CPU are in a place that are heavily used for decompression, as PS5 titles don't need those on the CPU. Everything else really makes no sense. Yes the CPU in the series x has a bit higher clocks, but just a bit. While loading the PS5 CPU should also always reach their highest clockrates, as the GPU has not much to do.
Funny that they upped the Series S maximum resolution (this was missing in the Tomb Raider patch) but didn't do the same thing on the PS5. As this is a dynamic resolution, I really don't get why they don't increase it. PS5 shouldn't have a problem to get more pixels out, even in back-compat more.
That the consoles couldn't hold a stable framerate all the time was clear from the beginning, as even really high end PCs struggle to run AC Odyssey at 60fps all the time. The engine seems not really well optimized to run at 60fps all the time. There seem to be some things in that game that don't easily scale that well with the performance, even if double the performance is available.
There shouldn't be any downclocking going on while loading. If not the GPU is doing some tasks we dont know about. Anyway, bad optimized title to begin with on all platforms.
I guess the question was more what is possible. E.g. it would be possible to give the game some kind of feedback "prepare to be hibernated in the next few seconds" and the game would than release resources (like raw textures etc) from it's memory to reduce it), than after resuming it could reload those parts (e.g. what SF tells is needed). But overall this makes things much more complicated and is more the "savegame" kind of solution. Quick Resume is more the brute force method so it can work with almost everything.
Just because you need less memory for a scene (when using SFS) it doesn't mean the game really uses less. Game will just fill the memory with other stuff. Not really randomly, but than developers have the room to actually use more memory than before, so that might make some features that weren't possible before possible now.
E.g. I guess this is why MS thought that 8GB for games is enough on the Series S, because if SF(S) is used the game just does not need those large amounts of memory. SFS has also the advantage that it reduces bandwidth usage. So even the much lower bandwidth shouldn't be a problem in the future.
Doesn't PS5 cap frequency in BC mode? That's probably what's happening here.
Not for newer and patched games
Flappy Pannus
Veteran
I suspect what may be at least part of the reason why he sees such drops is his CPU though, on the PC this title has a pretty large CPU bottleneck with frequent CPU spikes - it can't maintain 60fps at 1080p on my 1660 with periodic stutters that coincide with single-core CPU spikes of 80%+ when they occur on my I5-9400. I can get mid 40's at 1440p better than console settings inside Meridian/Mothers Heart too, but that's on a 1660 - his 2070 should actually be quite a bit better, but that Zen-1 architecture CPU could be holding him back.
He also states that this title 'needs' PCI-E 4.0, and while he may be correct in that this title was problematic to port due to the lack of shared memory architecture, not sure that's entirely the culprit for the less than optimal performance here. This is somewhat tough to determine though, as the cards that support PCI-E 4.0 are newer generation cards that have enough excess vram, it would be interesting to see if a 6GB card would show a difference if we had a version PCI 3.0 vs 4.0 to compare, but TechReport testing this on a 3080 and found no substantial change in performance:
I mean it may indeed be based in reality, a 3080 has so much brute force (but from youtube videos I've seen may still even have small drops below 60fps at 4k with everything maxxed) but like I said before his tendency to state things authoritatively when they're more hypothetical in nature is kind of my quibble with some of his videos.
All that said, some instructive concrete details in the video too - I especially liked that he showed side by side comparisons between checkerboarding on the Pro vs PS5 - the PS5 looks to resolve fine detail noticeably better, I remember playing this on my Pro and the hair shimmering was far more noticeable to my recollection. I've gone back and forth between this and the PC version at true 4k (it's hard to make a fully valid comparison as anything above 1080p on my 1660 screws up the texture detail though), in terms of fine detail they're remarkably close, as he said you would have to go to 1800p or above to really match how 4K checkerboarding looks in this title at 60fps. Checkerboarding was often dismissed by PC gamers as 'fake' 4K but really, at least in PS5 titles at 60fps with two titles I can directly compare on the same set - Days Gone and HZD - it holds up very well to native 4K imo.
- Status
Similar threads
- Locked
