August 3, 2025Aug 3 Debug says 'limited by mainthread'. Yet, as you can see from attached HWINFO screenshot, neither GPU nor CPU is fully utilized. What in the hell is going on here. 7800x3d and rtx5090 64gb expo ram. Run Cinebench to test performance and stability, all flying colors. Also Cyberpunk 2077 all maxed out, ray tracing path tracing and whatnot utilizes GPU at 100% and CPU at 80% and runs flawlessly smooth without any hiccups. Any ideas?
August 3, 2025Aug 3 I think your stats are showing that the CPU overall is experiencing approx 75% usage overall. However, that's only a general picture. Within that, you'll likely have one or two cores operating at max while the others aren't fully utilised. Hence limited by mainthread. Kevin Firth - AMD 9800X3D; Asus Prime X670E; 64Gb Cas30 6000 DDR5; RTX5090; AutoFPS
August 3, 2025Aug 3 Author That is what I initially thought, but looking at individual cores, not even one was near full utilization
August 3, 2025Aug 3 1 hour ago, SimianFlu said: That is what I initially thought, but looking at individual cores, not even one was near full utilization Are you actually looking at *cores*, or at virtual CPUs? If you have SMT enabled, then each processor being reported is actually a vCPU, which are paired two to each physical core. So utilization on each will be lower, because it's sharing a physical core with it's companion vCPU. Bob Scott | President and CEO, AVSIM Inc ATP Gulfstream II-III-IV-V Sys1 (MSFS20+24/XPlane12+11): AMD 9800X3D, water 2x240mm, MSI MPG X670E Carbon, 64GB GSkill 6000/30, nVidia RTX4090FE Alienware AW3821DW 38" 21:9 GSync, 2x4TB Crucial T705 PCIe5 + 2x2TB Samsung 990 SSD, EVGA 1000P2 PSU, 12.9" iPad Pro Thrustmaster TCA Boeing Yoke, TCA Airbus Sidestick, Twin TCA Airbus Throttle quads, PFC Cirrus Pedals, Coolermaster HAF932 case Sys2 (P3Dv5/v4): i9-13900KS, water 2x360mm, ASUS Z790 Hero, 32GB GSkill 7800MHz CAS36, ASUS RTX4090 Samsung 55" JS8500 4K TV@60Hz, 3x 2TB WD SN850X 1x 4TB Crucial P3 M.2 NVME SSD, EVGA 1600T2 PSU Fiber link to Yamaha RX-V467 Home Theater Receiver, Polk/Klipsch 6" bookshelf speakers, Polk 12" subwoofer, 12.9" iPad Pro PFC yoke/throttle quad/pedals with custom Hall sensor retrofit, Thermaltake View 71 case, Stream Deck XL button box Sys3 (DCS/P3Dv4/ATS/ETS): AMD 7800X3D, MSI MPG X870E Carbon, Noctua NH-D15S, 64GB GSkill 6000/30, EVGA RTX3090 Alienware AW3420DW 34" 21:9 GSync, Corsair HX1000i PSU, 4TB Crucial T705 PCIe5 + 2TB Samsung 970Evo Plus, TM TCA Officer Pack, Saitek combat pedals, TM Warthog, TM RS300 FF wheel/pedals, Coolermaster HAF XB case
August 3, 2025Aug 3 Nowadays, CPU utilization will never tell you the full story due to rapid thread shuffling by the software itself and even the OS. Only reliable readout due to the fact it is a single ccore is the GPU: if it is below 90-95%, you are CPU limited, no matter what any CPU readout tells you. With a 5090 it is no surprise, you will be CPU limited even using a 9950X3D in some scenarios with this fast GPU. Greetings, Chris AMD Ryzen 7 9800X3D, 2x32GB DDR5 6000MT/s RAM, MSI RTX 4090 Ventus 3X, Windows 11 Home, MSFS2024
August 3, 2025Aug 3 Ah. Easy fix. Take out that 400 base TLOD and put a more reasonable number like 50 or 60, change automethod to FPS variance 2-3% and keep the upper tlod near 350 as you're barely going to notice a difference. Have higher numbers can take more time for the app to work in the changes, causing stutters. Elevation to start TLOD can be between 1500-2000feet. Your main thread is getting hammeres because your TLOD is crazy high. Turn it down with autofps. I just tried your settings and go similar results: terrible frames. My settings have no stutters and keeps smooth. I have an fps cap for 50 in thr cfg file and let FG take care of the rest. I let autofps find its own fps during flight, keeps it smooth. At KJFK, fenix, beyond atc and traffic....smooth 7800+4090+64ram Just Flight RJ, 146 and F28, Piper Arrows ---A2A Aerostar and Comanche---Black Square Starship, Duke(s), TBM, Bonanza/BaronV2, KingAir---FSReborn FSR500---COWS Da42---FX P180, HJet & VJet---FlySimWare Chancellor and LearJet---FlightSimStudio EMB175 &P2006T---Fenix 320---PMDG DC6, 737(700+900), 777---C22J---Milviz Cessna 310 & Porter---SimWorksStudios Kodiak, PC12, Zenith & RV14---BigRadials Goose---IndiaFoxEcho MB3339+F35.
August 3, 2025Aug 3 One of the breakthroughs of 2024 is it no longer has the mainthread issue. dd Edited August 3, 2025Aug 3 by Sky_Pilot071
August 3, 2025Aug 3 Author 31 minutes ago, Sonosusto said: Ah. Easy fix. Take out that 400 base TLOD and put a more reasonable number like 50 or 60, change automethod to FPS variance 2-3% and keep the upper tlod near 350 as you're barely going to notice a difference. Have higher numbers can take more time for the app to work in the changes, causing stutters. Elevation to start TLOD can be between 1500-2000feet. Your main thread is getting hammeres because your TLOD is crazy high. Turn it down with autofps. I just tried your settings and go similar results: terrible frames. My settings have no stutters and keeps smooth. I have an fps cap for 50 in thr cfg file and let FG take care of the rest. I let autofps find its own fps during flight, keeps it smooth. At KJFK, fenix, beyond atc and traffic....smooth That would be a good advice with my previous GPU, but now TLOD seems to have little to no impact whatsoever (100 vs 400 on the ground 🤷♂️)
August 3, 2025Aug 3 Author 30 minutes ago, Sky_Pilot071 said: One of the breakthroughs of 2024 is it no longer has the mainthread issue. dd What am I looking at then? 🤔
August 4, 2025Aug 4 5 hours ago, SimianFlu said: That would be a good advice with my previous GPU, but now TLOD seems to have little to no impact whatsoever (100 vs 400 on the ground 🤷♂️) Contrary to popular belief, the cpu hit comes from terrain level of detail. Lower that down and turn off expert settings as well. Tlod of 400 will hammer the cpu. I tried your initial settings they hammered my cpu (we have the same). My 4090 is also a monster and barely takes a hit with my settings. Could be the plane.... do you gsx installed? and if so, visible passengers? I notice the fenix update caused stutters initially but they subsided. 7800+4090+64ram Just Flight RJ, 146 and F28, Piper Arrows ---A2A Aerostar and Comanche---Black Square Starship, Duke(s), TBM, Bonanza/BaronV2, KingAir---FSReborn FSR500---COWS Da42---FX P180, HJet & VJet---FlySimWare Chancellor and LearJet---FlightSimStudio EMB175 &P2006T---Fenix 320---PMDG DC6, 737(700+900), 777---C22J---Milviz Cessna 310 & Porter---SimWorksStudios Kodiak, PC12, Zenith & RV14---BigRadials Goose---IndiaFoxEcho MB3339+F35.
August 4, 2025Aug 4 Author I found the culprit! Some evil goblin enabled vsync, and because of my three 60hz popped out instrument disolays this capped the frame rate 🤦♂️ 🤡
August 5, 2025Aug 5 On 8/3/2025 at 2:04 PM, Sky_Pilot071 said: One of the breakthroughs of 2024 is it no longer has the mainthread issue. dd From the Oracle, and certainly seems to be the case in my experience because I can have ample VRAM unused, CPU headroom bigtime, GPU the same, and yet frames get limited: ✅ Summary Yes, MSFS 2024 still has a main-thread bottleneck. Despite some improvements in multithreading, the main thread remains the key limiter of performance in most real-world scenarios. Reducing CPU-heavy settings like displacement mapping, photogrammetry, and LODs can help, but this limitation remains inherent to how the sim processes frames. Noel System: 9900X3D Noctua NH-D15 G2, MSI Pro 650-P WiFi, G.SKILL 64GB (2 x 32GB) 288-Pin PC RAM DDR5 6000, WD NVMe 2Tb x 1, Sabrent NVMe 2Tb x 1, RTX 4090 FE, Corsair RM1000W PSU, Win11 Home, LG Ultra Curved Gsync Ultimate 3440x1440, Phanteks Enthoo Pro Case, TCA Boeing Edition Yoke & TQ, Cessna Trim Wheel, RTSS Framerate Limiter w/ Front Edge Sync. Aircraft used in MSFS 2024: Fenix A320, Aerosoft CRJ, FBW, WT 787X, I-Fly 737 MAX 8, Citation Longitude.
Create an account or sign in to comment