Posted December 22, 2020
SilentSnake86: Wow, so much theorycrafty bullshit in this thread.
Came here thinking I have a problem with my RTX 3080 and 5900X because CP2077 "only" uses 93-97% GPU.
My first instinct - I haven't swapped from Be quiet 600W to Fractal Ion+ 860W yet and Be quiet is not providing enough juice (not even in GPU stress benchmarks GPU goes above 98%), but maybe there is another problem (like NVIDIA drivers).
Thank you goguser9913, I suspected latest NVIDIA drivers and will try installing 457.51.
Let me clarify some stuff as a helpless and utter computer geek of 25+ years (and a software dev of way less):
1. Hyperthreading, AMD, CDPR "not liking" AMD - not true.
People don't realize how hyperthreading works - a totally random example explaining why it may hurt performance (this applies to games where you want as many fps as possible) - can't post link but you can search for stackoverflow "hyperthreading-makes-my-code-run-slower"
Basically, you allow 1 core to work on 2 threads. Imagine cooking - you do multitasking, peeling potatoes, pouring water into the kettle, preparing meat, cutting vegetables. You can't do all these things at the same time, but if you have potatoes boiling in a pot and you have to wait for them to get ready, and you can do something else meanwhile.
How does it relate to games - if a game can use, for example, up to 6 cores, then in case of a 6 core/12 thread cpu it should/will work better if you only assign the physical (6) cores to the game - this way you will have all your actual cores dedicated to the game, without using the "I have some spare time just now" technology. The difference isn't large, but just like in cooking - if you had another person with you, the dinner would be ready faster, even if only by 5-10%, because the workload you have to finish before the potatoes being boiled may span over their boiling time (e.g., potatoes will be ready, but you're still finishing the veggies, so dinner will be 1-3 minutes later than the optimal time).
Therefore, for my 5900X, it would PROBABLY (since it's hard to say that it's guaranteed) be better to set only physical cores to the game as it can't use 24 threads anyway. This is why 1.05 patch adds back hyperthreading support for AMD CPUs that have 6 cores/12 threads or less but keeps it off with CPUs that have more (like in my case).
2. Regarding the original post and GPU/CPU usage - huge topic, not even gonna try to cover it all.
a) TL;DR version - 100% CPU/low GPU usage = CPU bottleneck. With hyperthreading you may be CPU bottlenecked without having 100% CPU - CPU will bottleneck GPU because physical cores are at 100% while "I have some spare time just now" cores are, let's say, 50%, resulting in a seemingly stable and perfectly fine ~70-80% CPU usage. In reality, newer/better CPU will give you an FPS boost.
My own real life example - Destiny 2, 4790k, RTX 3080. 4 core/8 thread cpu - can't bottleneck Destiny 2, a game from 2017, right??? CPU usage would oscillate around 70%, and yet, after switching to 5900X, I'm getting up to +200 FPS depending on location - this game uses up to 6 cores.
b) 100% GPU usage/low CPU usage = GPU bottleneck. This basically means everything runs as it should in a situation where your FPS are unlocked. Typical scenario - GPU benchmarking using 4K, where GPU is taxed way more than CPU.
c) low, medium, high usage of CPU and GPU but none reach highest values (preferably 100% on GPU) = my (very little as I see) problem. plethora of reasons, could be PSU not providing enough power, could be RAM problem (hardware faults), could be GPU drivers/other drivers/Windows being Windows/moon in the wrong phase/banana peel. With this we can toss a coin.
With the above in mind, my (very short and limited so far) experience with CP - I overclocked my 3080 - Destiny 2 runs fine, benchmarks runs fine, other games run fine, GPU stress tests run fine - CP flatlines after ~1 minute in.
Lowered clock speeds on GPU - game 100% stable.
It's nice and easy to complain about CDPR and I can't blame people not to be PC troubleshooting experts, but it's not just the game that can be at fault here. The game seems to be very sensitive to any problems related to your PC, so if it flatlines often - try to use default settings on CPU, GPU, and RAM and see if it's still happening, use some benchmarking tools to test stability of your system etc. (just google, this post is long enough ;) ).
For my case ( c ) - like mentioned previously, I'll try with NVIDIA drivers first, if that doesn't change anything I'll check with new PSU. If the problem persists - my struggle and search continues.
What's the point of theorizing? o.O Came here thinking I have a problem with my RTX 3080 and 5900X because CP2077 "only" uses 93-97% GPU.
My first instinct - I haven't swapped from Be quiet 600W to Fractal Ion+ 860W yet and Be quiet is not providing enough juice (not even in GPU stress benchmarks GPU goes above 98%), but maybe there is another problem (like NVIDIA drivers).
Thank you goguser9913, I suspected latest NVIDIA drivers and will try installing 457.51.
Let me clarify some stuff as a helpless and utter computer geek of 25+ years (and a software dev of way less):
1. Hyperthreading, AMD, CDPR "not liking" AMD - not true.
People don't realize how hyperthreading works - a totally random example explaining why it may hurt performance (this applies to games where you want as many fps as possible) - can't post link but you can search for stackoverflow "hyperthreading-makes-my-code-run-slower"
Basically, you allow 1 core to work on 2 threads. Imagine cooking - you do multitasking, peeling potatoes, pouring water into the kettle, preparing meat, cutting vegetables. You can't do all these things at the same time, but if you have potatoes boiling in a pot and you have to wait for them to get ready, and you can do something else meanwhile.
How does it relate to games - if a game can use, for example, up to 6 cores, then in case of a 6 core/12 thread cpu it should/will work better if you only assign the physical (6) cores to the game - this way you will have all your actual cores dedicated to the game, without using the "I have some spare time just now" technology. The difference isn't large, but just like in cooking - if you had another person with you, the dinner would be ready faster, even if only by 5-10%, because the workload you have to finish before the potatoes being boiled may span over their boiling time (e.g., potatoes will be ready, but you're still finishing the veggies, so dinner will be 1-3 minutes later than the optimal time).
Therefore, for my 5900X, it would PROBABLY (since it's hard to say that it's guaranteed) be better to set only physical cores to the game as it can't use 24 threads anyway. This is why 1.05 patch adds back hyperthreading support for AMD CPUs that have 6 cores/12 threads or less but keeps it off with CPUs that have more (like in my case).
2. Regarding the original post and GPU/CPU usage - huge topic, not even gonna try to cover it all.
a) TL;DR version - 100% CPU/low GPU usage = CPU bottleneck. With hyperthreading you may be CPU bottlenecked without having 100% CPU - CPU will bottleneck GPU because physical cores are at 100% while "I have some spare time just now" cores are, let's say, 50%, resulting in a seemingly stable and perfectly fine ~70-80% CPU usage. In reality, newer/better CPU will give you an FPS boost.
My own real life example - Destiny 2, 4790k, RTX 3080. 4 core/8 thread cpu - can't bottleneck Destiny 2, a game from 2017, right??? CPU usage would oscillate around 70%, and yet, after switching to 5900X, I'm getting up to +200 FPS depending on location - this game uses up to 6 cores.
b) 100% GPU usage/low CPU usage = GPU bottleneck. This basically means everything runs as it should in a situation where your FPS are unlocked. Typical scenario - GPU benchmarking using 4K, where GPU is taxed way more than CPU.
c) low, medium, high usage of CPU and GPU but none reach highest values (preferably 100% on GPU) = my (very little as I see) problem. plethora of reasons, could be PSU not providing enough power, could be RAM problem (hardware faults), could be GPU drivers/other drivers/Windows being Windows/moon in the wrong phase/banana peel. With this we can toss a coin.
With the above in mind, my (very short and limited so far) experience with CP - I overclocked my 3080 - Destiny 2 runs fine, benchmarks runs fine, other games run fine, GPU stress tests run fine - CP flatlines after ~1 minute in.
Lowered clock speeds on GPU - game 100% stable.
It's nice and easy to complain about CDPR and I can't blame people not to be PC troubleshooting experts, but it's not just the game that can be at fault here. The game seems to be very sensitive to any problems related to your PC, so if it flatlines often - try to use default settings on CPU, GPU, and RAM and see if it's still happening, use some benchmarking tools to test stability of your system etc. (just google, this post is long enough ;) ).
For my case ( c ) - like mentioned previously, I'll try with NVIDIA drivers first, if that doesn't change anything I'll check with new PSU. If the problem persists - my struggle and search continues.
As you can see problem is solved with windows in-upgrade or format. End of story and no bottleneck ;)