How this started

To fit everything inside the NFC Skyreach 5 MINI and its 66mm CPU-cooler limit, I have spent the past six months collecting a pile of low-profile coolers: the AXP90-X53 FC, AXP-100 C65 Full Copper, IS-60 EVO, Black Ridge, AN400 ... By the end, even I thought I had a problem.

The two most expensive pieces were Thermalright full-copper models: the AXP90-X53 FC (53mm, four heatpipes) and the AXP-100 C65 (65mm, six heatpipes, with a 620g full-copper heatsink). My logic when I bought the C65 was simple: one class taller, two more heatpipes, discontinued, and full copper. How could it not be the ceiling of my collection?

Then the measurements slapped me in the face.

Test conditions

First, the test method, so you can decide whether any of this is useful to you:

• Platform: 9950X3D on a 28°C open bench
• Fan: the same 4070 GPU mod fan fitted to both coolers, removing the fan as a variable
• Load: a sustained AIDA64 FPU run, measuring package power at the 95°C thermal limit
• Secondary result: Cinebench R23 multi-core

I use “sustainable watts” instead of “how many degrees cooler” because a modern AMD CPU lowers its clock once it reaches the thermal limit. At that point, comparing temperature is meaningless. What matters is how much power the cooler can keep feeding the CPU inside that wall.

Results

R23 was practically identical: a 90-point gap, or about 0.23%, with the larger cooler ahead this time. But in AIDA64, the smaller cooler dissipated 8W more.

The extra two heatpipes, the larger body, and the extra money failed to produce a consistent lead. My first thought was that I had mounted it badly. I removed it, checked the paste impression, mounted it again, and retested. The result still pointed the same way.

Still unconvinced, I used another X53 FULL BLACK sample for a same-heatsink fan swap. Replacing the 4070 fan (T015) with a PF120 (T016) only moved AIDA64 from 181.3W to 183.2W—a gain of 1.9W. R23 improved by just 43 points.

Why this happens

Once you think it through, it is not that strange. The heat path is CPU → base → heatpipes → fins → air. A heatpipe only moves heat from the base to the fins. It is the courier, not the warehouse. Whether the heat actually leaves after delivery depends on fin area and the air moving through those fins. These two coolers live in roughly the same fin-stack class, both trapped under the height limit of a low-profile design. Four heatpipes are already delivering fast enough; the fifth and sixth are extra couriers arriving at a warehouse whose door is still jammed with parcels.

The bigger fan adding only 1.9W is the other side of the same problem. The airflow was already sufficient; the fin stack could not hand much more heat to it. Whichever side saturates first—airflow or fin area—throwing more hardware at the other side is wasted effort.

Floating Point Studio published a Bilibili test that lines up perfectly with my conclusion. They pushed a 9700K to 220W inside a closed case, and the entire 53mm class collapsed. The full-copper AXP90 hit 115°C and throttled, while the aluminium-fin AK90 held at 99°C. When intake is restricted inside a closed case, the thermal-capacity advantage of copper cannot be fully cashed in. Full copper is not invincible. It needs airflow, and when you cannot feed it enough, it can behave no better than aluminium—or even worse.

So where should the money go?

Let us make this practical.

Choosing between copper and aluminium, or four and six heatpipes, within 53–66mm: stop agonising over it. Within the same size class, much of that difference can land near the error margin. Pick the cooler that fits, is actually available, and does not come with an annoying fan.

Want an upgrade you can actually feel: jump to another class. Moving from 53mm to 67mm—the AXP120-X67 class—gets you a 120mm-scale fin stack. That is a different battlefield. One size-class jump matters more than piling extra material into the same class.

Still want the full-copper one: go ahead, but buy it for the right reason. Transient-load buffering, chasing an open-bench limit, or simply collecting a discontinued cooler are all valid. “It must sustain more watts” is not a reason supported by my testing.

Afterword

Thermalright replied that the X53 Pro should arrive in the second half of the year with five heatpipes and a new mount. Following the logic here, if the fin area does not change, the improvement may be exactly what their email called it: “slight.” I will buy one when it launches and come back to see whether this article gets vindicated—or slapped in the face instead.

The complete results, including every cooler, fan, and Test ID, live in the TK Lab test database. This article is about the story behind the numbers.

TK Lab