I recently bought a used gaming PC with an RTX 5070, Ryzen 5 9600X, ASRock B650M-H/M.2+ motherboard, 32 GB of DDR5-5600, a Corsair AIO, and a 650 W Seasonic PSU. It worked well when I tested it at the seller's house, but after bringing it home I started getting severe stuttering in GTA V Enhanced and CS2. The reported FPS looks reasonable, but frametimes are consistently around 20–50 ms with occasional spikes near 300 ms, and 1% lows can drop to zero.
I have already tried reinstalling GPU and chipset drivers, using DDU, testing each RAM stick separately, resetting the BIOS and CMOS, trying both EXPO and JEDEC memory speeds, disabling the integrated GPU, checking the SSD, resetting Windows, and confirming that Resizable BAR is enabled. GPU and CPU stress tests complete without reported errors, although the BIOS has occasionally frozen during startup.
LatencyMon shows sustained latency around 1500–2500 microseconds, with occasional spikes near 16,000 microseconds. The drivers most often listed include nvlddmkm.sys, Wdf01000.sys, dxgkrnl.sys, and ntoskrnl.exe.
The most suspicious issue is in HWiNFO: shortly after startup, the CPU initially runs around 4.5–5 GHz, but later drops to about 2 GHz or even 600 MHz. The CPU die temperature remains around 40–50°C, while the CPU VDD_MISC VRM sensor gradually climbs to roughly 99°C. CPU frequency drops when that sensor reaches about 99°C, and HWiNFO reports HTC Thermal Throttling and PROCHOT EXT Thermal Throttling.
Could the motherboard VRM be overheating or triggering protection, especially since this board has limited VRM cooling? Should I remount the AIO and reapply thermal paste, improve case airflow, update the BIOS, or replace the motherboard?
4 Answers
The CPU core temperature is probably not the real problem here. The VRM sensor reaching roughly 99°C at the exact moment the CPU drops clocks strongly suggests VRM thermal or power protection. That motherboard is a fairly basic model with little or no VRM heatsinking, and an AIO does not provide airflow over the motherboard power circuitry the way a tower cooler would.
Try removing the side panel and directing a household fan at the VRM area while repeating the OCCT test. Also set the case fans to a more aggressive curve and check whether the CPU stops throttling. If the behavior improves, you have confirmed that the VRM area is overheating or operating too close to its protection limit.
The LatencyMon results may be a consequence rather than the root cause. nvlddmkm.sys and dxgkrnl.sys can appear when the GPU is waiting for a CPU that is repeatedly dropping clocks, and Wdf01000.sys is a general Windows hardware-driver component rather than a specific failed device. Once the CPU throttling is fixed, retest latency before spending more time chasing individual drivers.
If latency remains high afterward, temporarily disconnect or disable extra USB, audio, networking, and other hardware devices one at a time and update their drivers.
Update the motherboard BIOS to the newest stable version and load default settings afterward. Then test with normal JEDEC memory speed before enabling EXPO. You can also temporarily increase airflow across the VRM area and use CPU Eco Mode. If the board still reaches its protection threshold while the CPU itself stays cool, replacing the motherboard with a model that has properly heatsinked VRMs would be more sensible than replacing the CPU or GPU.
It is still worth checking the physical cooling installation. Make sure the AIO pump is connected to the correct header and that it is actually running. Remove and reinstall the cooler with fresh thermal paste, especially since the system was transported after the original test. A poorly mounted cooler usually causes high CPU temperatures, though, so the separate 99°C VRM reading remains the more convincing explanation here.
Reducing CPU power with Eco Mode or a mild negative PBO offset can also be a useful diagnostic. If that prevents the throttling, it further points toward the motherboard’s power delivery rather than the CPU cooler.