Cpu and GPU not fully used but framerate decrease

Well, from your point of view sure not much could be upgraded but from Poumpoum point of view, there's a lot to gain, his config is definitely held up somehow by his RAM kit.

Just try for yourself to disable DOCP on your PC to get JDEC timings and run an ACC replay file measuring the average frame time, you'd be surprised by the results.

Not saying optimizations aren't welcome, but there are also integration constraints to be considered between components to reach peak performance.

 

I bought DDR4-4000MHz CL14 (and won the price for the compulsive buyer of the month...) with the hope of running them at low timings @2666MHz and surprisingly, I was able to run them @2666MHz CL11 to achieve the famous "Sweet spot".

Those are the results with my test scenario : Quick Race / 29 opponents @90% / Nurburgring GP / Clear weather:
- 5780*1080 - Mid Settings (+shadow epic and mirror 40m) / 2666MHz @CL16 : avg=98fps - low1%=70fps - low0.1%=58fps
- 5780*1080 - Mid Settings (+shadow epic and mirror 40m) / 2666MHz @CL11 : avg=99fps - low1%=71fps - low0.1%=63fps
- 5780*1080 - Epic Settings / 2666MHz @CL16 : avg=74fps - low1%=53fps - low0.1%=48fps
- 5780*1080 - Epic Settings /2666MHz @CL11 : avg=79fps - low1%=58fps - low0.1%=50fps

Honestly it changes nothing @Mid Settings, and I can't even say than I have a +5fps @Epic Settings because my low0.1% is only +2fps. Perhaps it's more a ram frequency bottleneck than a ram true latency bottleneck....

However, I noticed that the big difference I have in fps between Mid and Epic settings comes from mainly 1 parameter "Opponents Visibility". If I set this parameter @1 (and only see 1car in front of me) I can reach an avg 150fps in Mid and 100 fps in Epic. I begin to think that this game doesn't like triple screens. I'm pretty sure that playing with triple screens (3K) on this game cost more CPU than playing it with single screen 4K because you see the same amount of cars in main monitor + cars on your right / left...

I'm a little disappointed right now and I want to burn this damn PC !

 

It's a bit of both tbh, on EPIC you gain almost 7% avg and 10% on 1%low on RAM latency only, nothing to sneeze at.

Sure wider field of view means more assets to load/unload, more car to simultaneously display, in VR I use 15 cars and unless massive pile up at the start, I see no car popping in and out, so between 15 and 29 you may could find a trade off to regain some performance. Car LOD could help too.

Also, with such a high resolution and field of view, try to see what could a click or two down on texture quality could bring you, in VR at least that's a pretty significant cost.

Have you tried on your 3080ti to increase VRAM speed ? I pushed mine a bit to try up to +1000 in MSI AfterBurner, but I found more balanced 0.1% and 1% low @ +500

Really hope you'll find settings that get you to the desired performance.

 

the scheduler always tries to switch to different cores for the same process

 

Coming back to this thread, ACC is CPU bound and my best gains were all with CPU upgrades



bottom lows bar charts were with 1080ti already published in this thread here
upper ones with 3080ti, first keeping the 5800X (ok still good gains in 1% and 0,2% lows, but meh in avg) then drop in of the 5800X3D.... pretty massive uplift...

All figures above are from the same replay file, at 1440p at almost lowest graphics settings you can set, same personal graphics preset was used.

I could even run a 35 AI race at Spa locked at smooth 90 fps in my VR headset. That was purely impossible with my former 5800X.

Why 5800X3D is working so much better? Because it has a lot of L3 cache, and L3 cache is several times faster than your system RAM, it means a lot more physics computations can be run with data never leaving the CPU and processed in one of the fastest memory available on the system.

ACC is very CPU and RAM intensive to run all the physics of the game (tyre simulation, suspension simulation, aero simulation, sounds, AI when playing solo, etc...) all of this makes a lot of calculations, a lot of data to process, and ACC physics engine runs at 400 Hz...

The slowest your memory subsystem is, the more CPU cycles are wasted waiting for results to be used and move on. For that reason I've always recommended to get good RAM kits and RAM tuning to maximise your results when CPU upgrade couldn't be an option, now a new king is in town : 5800X3D and whatever the CPU you're looking at, take the one in the lineup with the largest L3 cache (from Intel or from AMD)

ACC is CPU/RAM bound period.