On GeForces, not on Radeons. Don't know about Arcs.
DirectStorage GPU Decompression, RTX IO, Smart Access Storage
- Thread starter DavidGraham
- Start date
Interesting, I suspect it probably will be on XSX also though. Probably helping with the fast loading it has.
The install size for Indy is much smaller on XSS, but with it using SFS would have expected bit higher texture quality.
VRS having lower performance impact the lower the resolution, you'd think they would use that on XSX but not XSS where will also have bigger visual impact.
Not sure which has biggest texture impact on XSS Indy, VRS or texture quality
@GLS28R @Kaotik
I'm quite happy to report that there's no stuttering at all going in and out of cinematics or camera cuts... also what's great to see is that there's absolutely no issue with DirectStorage tanking frametimes on Nvidia hardware if you spin the camera around wildly with the mouse. Not sure if this game uses GPU decompression though. I suppose I could check with SpecialK next time.
But whatever they've done or are doing with this game... it's MUCH better than their results with the Insomniac engine.
Overall, I'm fairly content with this quality of this port thus far. It's miles better than TLOUP1.. although there's always some issues I hope they work out.. as well as a few nitpicky things I have.
I'm quite happy to report that there's no stuttering at all going in and out of cinematics or camera cuts... also what's great to see is that there's absolutely no issue with DirectStorage tanking frametimes on Nvidia hardware if you spin the camera around wildly with the mouse. Not sure if this game uses GPU decompression though. I suppose I could check with SpecialK next time.
But whatever they've done or are doing with this game... it's MUCH better than their results with the Insomniac engine.
Overall, I'm fairly content with this quality of this port thus far. It's miles better than TLOUP1.. although there's always some issues I hope they work out.. as well as a few nitpicky things I have.
Today I got my answer to the question above. They're only using CPU based DirectStorage. Makes sense..
This comes from Digital Foundry's Tech interview with Naughty Dog and Nixxes posted today
www.eurogamer.net
I don't think GDeflate GPU based decompression is long for this world... but it also makes me worry about neural texture compression, since the same fundamental issue applies to it as well. The technology is great, but the hit to performance and frametime consistency isn't worth it. These types of decompression need to happen off of the GPU on a dedicated decompression block.
This comes from Digital Foundry's Tech interview with Naughty Dog and Nixxes posted today
The Last of Us Part 2 Remastered: how the PlayStation game was ported to PC
Alex Battaglia interviews the developers of The Last of Us Part 2 for PC. How did the port progress, what tech does it use and what remains to fix?
www.eurogamer.net
I don't think GDeflate GPU based decompression is long for this world... but it also makes me worry about neural texture compression, since the same fundamental issue applies to it as well. The technology is great, but the hit to performance and frametime consistency isn't worth it. These types of decompression need to happen off of the GPU on a dedicated decompression block.
DegustatoR
Legend
Or how about developers would follow what MS has explicitly recommended in DS SDK and provide the users with a UI option to choose what h/w to use for decompression?
The problem is that PCs are all different (and evolve constantly!) which means that the bottlenecks are different which means that you can't just assume that it is better to decompress the assets on CPU or GPU as on different systems either can be optimal. If a game is 100% GPU limited then putting additional work on the GPU will always result in a performance loss, and the same will happen when using CPU decompression in a CPU limited scenario. The best option in this case would be to let users decide where to perform decompression as the best option is 100% system specific.
The "dedicated decompression block" sounds like an awful idea on PC IMO.
Edit: In fact the same logic can be applied to shader compilation. It is also almost 100% system specific which would be the best option to perform shader compilation on PC - as a one "precompile" step at game's launch (or in the menus prior to actually launching the gameplay) or as a background compilation process. The latter can be completely fine on many core CPUs while lower end ones would in fact struggle with it quite a bit.
Last edited:
Again, the issues with GPU decompressions are only known to apply to NVIDIA. At least AMD is free of any such issues with GPU decompression. Due lack of proper testing we have no clue on what's the situation at Intels side.
The world doesn't need to and shouldn't bend over because NVIDIA can't fix their issue.
How do developers allow users to decide which hardware to use for decompression when they require that the assets be compressed using different algorithms which may only work on either the CPU or GPU?
No, they don't only apply to Nvidia. Nobody wants this type of shit when they simply turn the camera in their games...
Last edited:
DegustatoR
Legend
You've said this many times now without providing any sort of proof - care to do so?
From what I've seen "the issues" (if you could call performance degradation "an issue") are present on all GPUs and do not in fact "apply only to Nvidia".
Granted that Radeons generally has higher async compute ceiling (due to lower h/w utilization in the absence of it) which could help them with hiding the cost more efficiently.
You can decompress GDeflate on CPU. And if that's the problem then there can be a different algorithm introduced - GDeflate is what Nvidia has supplied MS with essentially, if people aren't happy with it nothing stops them from providing a better alternative.
Care to elaborate?
ComputerBase did do the testing and found that Radeons have no issues with DS GPU Decompression, while NVIDIA does.
Ratchet & Clank: Rift Apart: Ohne DirectStorage gibt es mehr FPS auf GeForce-GPUs
Ratchet & Clank: Rift Apart unterstützt als eines der ersten Spiele DirectStorage. Offenbar gibt es aber auf Nvidia-Grafikkarten Probleme.
www.computerbase.de
I'm pretty sure I've provided the link above several times
DegustatoR
Legend
That's just one game and just one test. AFAIK the "issue" there goes away completely if you drop texture quality by one point suggesting that this can be an implementation issue.
Somewhat interestingly.. Monster Hunter Wilds recent patch came out supposedly improving VRAM utilization and a number of players reported issues with stuttering.
Someone released a tool to *decompress the textures and repackage them* and then suddenly people were able to play with the highest textures on lower end GPUs without stuttering..
Here's a link to the tool:
www.nexusmods.com
Someone released a tool to *decompress the textures and repackage them* and then suddenly people were able to play with the highest textures on lower end GPUs without stuttering..
Here's a link to the tool:
MHWs Tex Decompressor Tool
A tool to create a patch pak with uncompressed textures. It may improve frame rate stability, but will takes up more disk space.
www.nexusmods.com
Yes, I did already elaborate. I showed a game where AMD definitely does have the same issues.Care to elaborate?
ComputerBase did do the testing and found that Radeons have no issues with DS GPU Decompression, while NVIDIA does.
View attachment 13396Ratchet & Clank: Rift Apart: Ohne DirectStorage gibt es mehr FPS auf GeForce-GPUs
Ratchet & Clank: Rift Apart unterstützt als eines der ersten Spiele DirectStorage. Offenbar gibt es aber auf Nvidia-Grafikkarten Probleme.
I'm pretty sure I've provided the link above several times
I think fundamentally the issue here is that at the cost of a bit more storage space, you can provide something which we know works better across all GPU vendors... and developers are actively choosing to forego GPU Decompression and doing it on the CPU instead. Currently, I agree with their decision... because the results are simply night and day. Also in Alex's interview there with Nixxes about TLOUP2 port they specifically say the PC version can load much faster than the PS5 version, but it's the shader compilation which happens during loading that often prevents it.
Apparently you edited the message while I was replying.
Considering the mess MHW is, can you be sure those issues are caused by GPU Decompression?
You can't just assume it's the cause without some sort of test to show it is caused by it. Supposedly "Special K" mod allows one to disable GPU Decompression in MHW but can't find any benchmarks comparing it.
If someone provides me with the game I'm more than happy to test it myself, but I'm not paying for that steaming pile of unoptimized crap.
DegustatoR
Legend
It's not just the storage space though. GDeflate allows streaming to the GPU VRAM in a compressed format which gives you 2-4X the effective PCIE bandwidth. With CPU decompression you're reading the compressed data but uploading it to the GPU in a decompressed format which means that you lose half of bandwidth savings.
And again it is completely system configuration specific whether you'd want to prefer one or the other.
Yes, but that bandwidth savings is for nothing when game frametimes go to shit just by turning the camera.
For things such as decompression it shouldn't be the player choosing. Games should be doing this stuff invisibly to the player. One of the points MS was touting when they were talking up DirectStorage before it released was how it (Windows) would always choose the best path given the hardware that was being run. Games need to do the same thing. If a player has an overabundance of RAM and VRAM, a game should saturate all available memory and improve performance and load times. If they have fast storage, but less memory, it should stress the storage bandwidth. If a person has a fast CPU but a slow GPU, it should decompress on the CPU.. or vice versa with a fast GPU and slow CPU. Games should allow players to optionally pre-compile all PSOs if they choose, or do it on background threads. If the CPU doesn't have many cores... behavior should be to pre-compile more PSOs upfront.. if there's lots of cores, it should do more background compilation... and so on and so forth.
DegustatoR
Legend
The current situation is what you get with this philosophy.
Consider that PCs evolve and that in the future a gaming CPU can have 4X more cores making it an ideal candidate for any form of decompression.
Can a game has some heuristic which would suggest the optimal settings for a particular configuration? Sure. Should a game also has an explicit option in case that fails? Definitely.
Not having options is a console mindset and it works there only because a console is a 100% known fixed h/w.
Again, you have not shown this to be the case. Your example is from a game that is known to be horribly unoptimized in many ways without any proof whatsoever that GPU Decompression let alone DirectStorage is the culprit for said frametime dips.
You would need to show it doesn't happen without GPU Decompression to make your argument valid.
Other games with GPU Decompression do not exhibit similar issues on Radeons.
We know it to be the case because when you decompress the textures using the tool I linked above... there's no issue on either GPUs.
Furthermore, Monster Hunter Wilds is one of the only other points of reference using GPU decompression which isn't a port done by Nixxes utilizing Insomniac's engine.. which could just as easily be an explanation as to why Nvidia has problems where AMD doesn't in said titles... which helps us rule out a specific engine or developer issues being the cause.
Last edited:
Similar threads
