7/26/2026

A Cooler Is Not Better Just Because It Is Tighter: Tracing an Unexpected Restart

After raising PBO, the system did not throttle or show a blue screen; it went black and restarted. I backed all four cooler screws off evenly, and the same settings completed the test. This is the pitfall I hit—and a clear separation between what I measured and what remains an inference.

How this started

After the previous test ended with the X53 FC and C65 effectively tied, I still had one cooler I had never properly tortured: the SilverStone XE02-AM5. It is a 65mm, five-heatpipe product originally sold as a server cooler, with an official 170W rating. The lesson from the previous article was that official TDP numbers are not especially trustworthy, so I decided to raise PBO and find its real ceiling.

The 9950X3D is not exactly honest out of the box either. It carries a 170W TDP, yet its default PBO limits are PPT 200W, TDC 160A, and EDC 225A—the factory has already exceeded its own headline number. I kept raising PPT. At PPT 230W, TDC 175A, and EDC 230A, the screen went black and the system restarted.

No blue screen—and that mattered

My first instinct was that I had pushed current too far. Maybe TDC or EDC had spiked and triggered VRM or power-supply protection, cutting the system off at the knees. The direction of that guess was reasonable. I had simply caught the wrong suspect.

I had missed one clue: a hard restart and a crash are not the same thing. When a CPU genuinely overheats or the system becomes unstable, Windows usually struggles first—a freeze, a blue screen, or clocks falling while it tries to hang on. This time there was no warning at all. The display vanished and the system restarted as if someone had yanked the power cable out and plugged it straight back in. That clean, immediate cutoff looked more like a protection mechanism stepping in than a system slowly losing the fight.

The real turning point: the mount was too tight

With nothing to lose, I backed all four cooler screws off by the same tiny amount, then reran the exact same PPT 230W, TDC 175A, and EDC 230A settings.

Stable. The same settings completed R23.

T005 scored 41,196 in R23 multi-core and sustained 218.5W package power over the last five minutes, with about 11 minutes 48 seconds of high load. The 230W figure is the PPT setting, not actual sustained CPU power. AIDA64 ran for only 1 minute 14 seconds, so this record cannot support a claim of more than 230W sustained heat dissipation.

Most people follow a simple mounting intuition: tighter means better contact, more pressure, and therefore better cooling. But the contact between a cooler base and the CPU needs even pressure, not maximum pressure. If an overtightened mount distributes force unevenly, the base or CPU substrate can flex by a tiny amount. A broad contact patch can turn into a few pressure points: much higher local force, but worse overall heat transfer. More awkwardly, the AM5 socket was never designed to absorb unlimited bending force. Too much pressure may affect pin contact, which could explain why protection cut power immediately instead of the CPU merely throttling.

That mechanism is my inference, not a conclusion proven by dissecting the board. The dramatic difference in stability before and after changing the screw pressure is what I actually reproduced.

Troubleshooting: first find out which wall you hit

This episode gave me a troubleshooting habit I will use next time:

Open HWiNFO64 and look at Limit%. PPT Limit at 100% is the power wall. TDC or EDC Limit at 100% is the current wall. Thermal Throttling means the temperature wall. You respond to those three walls by lowering the corresponding number. But if there is not even a trace of throttling and the system simply cuts out and restarts, those three walls are not the whole story. Check contact pressure instead of endlessly trimming PBO numbers.

Most guides teach you how to adjust PPT, TDC, and EDC. Nobody warns you that “power disappears instantly” and “the CPU throttles while hanging on” are two different failures pointing to two different troubleshooting paths.

So what should you do?

If overclocking or a PBO adjustment gives you an unexpected restart, do not immediately start cutting numbers. Check whether the mount has been overtightened. Back all four screws off evenly by a very small amount—do not loosen just one corner and tilt the base—then rerun the exact same settings.

If you are trying to find the real ceiling of a server-class cooler: the XE02-AM5 is officially rated at 170W. In my test, with PPT set to 230W, R23 sustained 218.5W over its last five minutes. Mounting pressure made a bigger difference to whether this test completed than buying another expensive fan.

Afterword

The actual sustained figure over the last five minutes of this R23 run was 218.5W. I have not found the true ceiling yet, and AIDA64 still needs a proper long-duration record. When I have time, I will keep raising the limit and see how far the screws can be backed off—and how much power it takes—before I hit the next wall. The full record is on the T005 database page. This article is just the story of the hole I stepped into.

TK Lab