Reason for Insane CPU usage.

[MisterEdd]

This is a link from the yuzu emulator that explains that consoles have dedicated hardware to decode ASTC textures. Even high quality textures being decoded by a Nintendo switch would need a 24 core CPU to be decoded without noticeable issues like stuttering.

https://www.patreon.com/posts/upcoming-changes-47169294

[episoder]

nonsense. sry. consoles don't use astc. this is not mobile phones or arm vr. it's bc1-bc7. even the switch runs a custom tegra from nvidia and i think should use the same compression.

[Bagel]

The reason is that $on¥ is using our CPUs to mine crypto currencies so they can hoard even more money..

[MisterEdd]

Originally posted by episoder:

nonsense. sry. consoles don't use astc. this is not mobile phones or arm vr. it's bc1-bc7. even the switch runs a custom tegra from nvidia and i think should use the same compression.

Still the same basic principle they could use cuda to decompress instead of hammering the processor. Its almost like they have API's for such things like nvida rtx io or direct storage decompression so you wouldnt have to destabilize the frame pacing by juggling these huge textures back and forth between cpu, ram and vram.

[episoder]

i dunno if cuda could manage general purpose decompression like zip or oodle. direct storage does not yet use graphics hardware decompression either, afaik, since today. it's also using the cpu. direct storage on pc is just an optimization of the loading procedure from the ssd and still using ram cache.

either way. the decompression is not the issue of the game. it's pretty damn fast. it's not hammering the cpu at all when you stream assets. this is damn sh!t DF threw out there. they have not tested or verified that information. they're hobby journalists and clickbaiting. you should unfollow them. yes

[Medusa]

Originally posted by episoder:

i dunno if cuda could manage general purpose decompression like zip or oodle. direct storage does not yet use graphics hardware decompression either, afaik, since today. it's also using the cpu. direct storage on pc is just an optimization of the loading procedure from the ssd and still using ram cache.

either way. the decompression is not the issue of the game. it's pretty damn fast. it's not hammering the cpu at all when you stream assets. this is damn sh!t DF threw out there. they have not tested or verified that information. they're hobby journalists and clickbaiting. you should unfollow them. yes

Are you allright ?
I am pretty sure DF knows what they are saying.

[episoder]

Originally posted by Medusa:

Are you allright ?
I am pretty sure DF knows what they are saying.

defo not. they got a big thumb down from me on this one. bs testing and mouth running without facts. not sure what else they "report" on. i don't wanna know. unbiased opinion.

[「 Boss 」]

High CPU usage is an nvidia driver bug, I never see more than 10% usage in game even during stressful sections.

[MisterEdd]

That's kind of my point though why are they using oodle at all. its literally destroying the port. I am not extremely knowledgeable about this subject so maybe there is some good reason other than them just being lazy but I dough't it. Yes its fast from a theoretical level but hammering the IO and juggling all that information around is not efficient and even in cases were the CPU isn't maxed at 100% CPU usage its still taking up a massive amount of cycles and increasing latency between everything. This is going to directly correlate with a drop in 1%/.01% low frame rate when other cpu related things happen like NPC's and combat happen. Its also going to ruin the the frame pacing. Just play a game that runs perfectly fine and run a cpu benchmark tat about 50-70% of the cpu and watch how choppy everything gets. Also it absolutely is hammering the CPU its literally in the patch notes "updated texture streaming to reduce CPU usage". I Have a 5900x and its using 40 to 60% which is insane. People that are at console level parity or even well above that are actually going to be CPU limited which will make all these issues much worst.

Decompression is the main issue and is going to continue to be an issue in upcoming games that leverage the IO speed of consoles with unified memory. Developers need to implement direct storage or even do something proprietary. Sure fast load times are great but most people don't build there own PC to shave seconds off a loading screen. The reason these API's exist is to streamline the IO because of all the issues I described above. Take a look at these links from Microsoft and NVIDIA. PC's have been able to brute force there way out of a lot of stuff over the years but this issue will likely not be fixed even with faster CPU's and memory. Texture sizes are going up exponentially while moore's law dies for non parallel workloads. Sure you could throw more cpu threads at it but that would again just destabilize everything and make processing uneven.

https://www.nvidia.com/en-us/geforce/news/rtx-io-gpu-accelerated-storage-technology/
https://devblogs.microsoft.com/directx/directstorage-1-1-coming-soon/

[Nightwulf1974]

Originally posted by episoder:

i dunno if cuda could manage general purpose decompression like zip or oodle. direct storage does not yet use graphics hardware decompression either, afaik, since today. it's also using the cpu. direct storage on pc is just an optimization of the loading procedure from the ssd and still using ram cache.

either way. the decompression is not the issue of the game. it's pretty damn fast. it's not hammering the cpu at all when you stream assets. this is damn sh!t DF threw out there. they have not tested or verified that information. they're hobby journalists and clickbaiting. you should unfollow them. yes

This is what happens when you do too much drugs. Don't be this guy.

[Metro]

I just don't get why this game uses more than 80% of a Ryzen 5600x, not even the witcher 3 next gen, Marvels Spiderman, marvels spiderman Miles morales use so much Cpu even with RT On, The last of us is almost a corridor game, it's not an open world, there aren't that many characters on screen or a very complex AI, it doesn't even have RT or an exaggerated LOD.

[MisterEdd]

Yes i'm sure these multi billion dollar companies spent massive amounts of time and money to create these API's just for laughs. It's not like they would try to be proactive in alleviating a massive bottleneck they knew would be an huge issue. Lets just ignore it and hope it goes away like shader compilation did. Am I right? Crazy to think a game associated with two different studios who are both renown for there ability to optimize games could over come this hurdle. There is simply no solution at all to running this game smoothly on hardware that is literally 10x as fast with more memory. ITS IMPOSSIBLE! You know what they say if you cant attack the argument attack the person that you don't know on the internet.

[MisterEdd]

Originally posted by Metroplexmaster:

I just don't get why this game uses more than 80% of a Ryzen 5600x, not even the witcher 3 next gen, Marvels Spiderman, marvels spiderman Miles morales use so much Cpu even with RT On, The last of us is almost a corridor game, it's not an open world, there aren't that many characters on screen or a very complex AI, it doesn't even have RT or an exaggerated LOD.

Yea witcher 3 runs horrible because of the translation layer but again its still a CPU issue. As much as people rag on it red engine is actually pretty solid when it comes to frame time consistency if your hardware is up for the task. Not sure if you are being sarcastic but its decoding the compressed textures which are quite a bit more detailed than the witcher 3. It doesn't really matter that it is an open world its all about the LOD vs texture size and how efficient it is. Spider man is probably doing something similar to the last of us just in a more efficient way. Its atleast somewhat justified because its open world and you are streaming stuff in and out extremely quickly due to how fast the spiderman moves through the world. Last of us on the other hand is literally just decompressing textures with the CPU. Its really the only thing it could be as most cpu workloads in games would choke if you split it up across 12+ threads. Compression/decompression on the other hand would scale with as many threads as possible.

[episoder]

i ran the game completely uncompressed. other people tested initial level loading times. they got half. i tested only the prologue asset streaming. i saw no improvement in there. it's not the compression hogging the cpu while streaming assets.

just now tested dstorage v1.1 samples from microsoft's github. i saw a video earlier that was from 1.0 that did not have the gdeflate function yet. in v1.1 it actually works tho. still not a cuda thing. it's a plain hlsl compute shader.

[azogthedefiler25]

Originally posted by 「 Boss 」:

High CPU usage is an nvidia driver bug, I never see more than 10% usage in game even during stressful sections.

The in game monitoring isn’t accurate. Pull up MSI Afterburner to see the true cpu usage.

[Sustain]

Originally posted by MisterEdd:

This is a link from the yuzu emulator that explains that consoles have dedicated hardware to decode ASTC textures. Even high quality textures being decoded by a Nintendo switch would need a 24 core CPU to be decoded without noticeable issues like stuttering.

https://www.patreon.com/posts/upcoming-changes-47169294

No the ASTC texture compression format is only really used on the switch and other mobile devices and not on all consoles it was specifically made for mobile gaming so that's wrong and TLOU PC doesn't use it.
There is no reason to use it on any other console or pc since there are better options for texture compression. Besides the Gpus in the PS5/Xbox and pretty much every PC GPU (only Intel Iris chips support it but not their discrete GPUs) lack the support for this format which is the main reason why the emulator needs to use more CPU power to decode it for switch games.