I would respectfully have to strenuously disagree with the idea that the Rock64 needs only a small heatsink OR NO HEATSINK AT ALL.
I have actually done paid thermal design consulting work, and I have never heard such silly foolishness.
I have a HUGE 28mm x 28mm x 20mm heatsink on my Rock64 CPU and it still hits 80C to 85C on stressful benchmarks.
This heatsink is about as large as can be mounted directly to the chip with thermal tape (so large that it requires a 2mm high copper shim to lift it up far enough to clear other components on the board).
Heatsinks of this size are rated by themselves at about 12 degrees C per Watt, but when we add in the thermal interface material (Arctic Alumina Thermal-Epoxy plus Thermal Tape, total about 2 °C/W) , and the 5 °C/W thermal resistance between the junction and the top of the CPU package, you end up with about 19 °C/W.
Coincidentally, this thermal resistance number is roughly EQUAL to the published default no-heatsink-at-all thermal resistance of the CPU when mounted on a reference 90mm square PCB.
This no heatsink 19 °C/W number is a result of the natural heatsinking effect that a BGA package like the one used by the Rock64's RK3328 CPU derives from simply being soldered to a multi-layer PCB with internal copper ground and power planes.
Thermal resistances in parallel add just like electrical resistances, by proportionally LOWERING the total resistance, so the effective 19 °C/W junction to ambient of that big giant 28mm x 28mm x 20mm heatsink, added in parallel with the PCB derived 19 °C/W cuts the TOTAL thermal resistance IN HALF to about 9.5 °C/W .
Let's look at how this translates into an actual peak CPU temperature by looking at some actual numbers.
First, run a stressful benchmark that fully loads ALL FOUR CPU CORES:
>> openssl speed -multi 4
The CPU load crated by this test varies a bit, so I let it run a few minutes to see the peak current load and CPU temperature.
The input current to the Rock64 when running this stressful benchmark peaks at between 1.2 and 1.4 Amps corresponding to between 6 and 7 watts of input power at 5 volts, with temperatures peaking over 100°C without a heatsink installed.
When we take into account the published efficiencies of the RK805 (which does the input power conversion and conditioning for the RK3328), and the other loads on the PCB, we can estimate that the CPU TDP is hitting a maximum of about 4 Watts.
To see if this agrees with the high CPU temperatures we are seeing, multiply that 4 Watt CPU TDP times the 19 °C/W of baseline PCB thermal resistance that Rockchip list in their spec sheet, and we are looking at a punishing 76 °C temperature rise above ambient, which could push your CPU temperatures above 100°C very quickly (which corresponds exactly to what I am seeing in free air without a heatsink). Over 100 °C is NOT a healthy operating temperature for a 28nm geometry SOC to start with, and If you place the Rock64 in a case, this situation gets worse, and it will quickly either throttle to protect itself, or ramp up to the 120°C destruction point and shut down. THIS IS WHY I SAID THAT THE IDEA THAT THE ROCK64 DOESN'T NEED A HEATSINK IS SILLY.
With my large heatsink cutting the CPU junction to ambient thermal resistance to roughly HALF, the temperature rise is also cut in half, so instead of temperature spiking to over 100C and throttling, I am seeing a much more reasonable 40°C to 55°C temperature rise above ambiant (depending on airflow), which with an ambient temperature of 30°C gives me a maximum CPU temp of 85°C. Technically I should be seeing even better results, but the effectiveness of the heatsink is reduced quite a bit by the stagnant lack of airflow and heat radiating up from the PCB. With even a tiny amount of airflow, I see temperatures in the low 60°C range, even at peak CPU loads.
I have actually done paid thermal design consulting work, and I have never heard such silly foolishness.
I have a HUGE 28mm x 28mm x 20mm heatsink on my Rock64 CPU and it still hits 80C to 85C on stressful benchmarks.
This heatsink is about as large as can be mounted directly to the chip with thermal tape (so large that it requires a 2mm high copper shim to lift it up far enough to clear other components on the board).
Heatsinks of this size are rated by themselves at about 12 degrees C per Watt, but when we add in the thermal interface material (Arctic Alumina Thermal-Epoxy plus Thermal Tape, total about 2 °C/W) , and the 5 °C/W thermal resistance between the junction and the top of the CPU package, you end up with about 19 °C/W.
Coincidentally, this thermal resistance number is roughly EQUAL to the published default no-heatsink-at-all thermal resistance of the CPU when mounted on a reference 90mm square PCB.
This no heatsink 19 °C/W number is a result of the natural heatsinking effect that a BGA package like the one used by the Rock64's RK3328 CPU derives from simply being soldered to a multi-layer PCB with internal copper ground and power planes.
Thermal resistances in parallel add just like electrical resistances, by proportionally LOWERING the total resistance, so the effective 19 °C/W junction to ambient of that big giant 28mm x 28mm x 20mm heatsink, added in parallel with the PCB derived 19 °C/W cuts the TOTAL thermal resistance IN HALF to about 9.5 °C/W .
Let's look at how this translates into an actual peak CPU temperature by looking at some actual numbers.
First, run a stressful benchmark that fully loads ALL FOUR CPU CORES:
>> openssl speed -multi 4
The CPU load crated by this test varies a bit, so I let it run a few minutes to see the peak current load and CPU temperature.
The input current to the Rock64 when running this stressful benchmark peaks at between 1.2 and 1.4 Amps corresponding to between 6 and 7 watts of input power at 5 volts, with temperatures peaking over 100°C without a heatsink installed.
When we take into account the published efficiencies of the RK805 (which does the input power conversion and conditioning for the RK3328), and the other loads on the PCB, we can estimate that the CPU TDP is hitting a maximum of about 4 Watts.
To see if this agrees with the high CPU temperatures we are seeing, multiply that 4 Watt CPU TDP times the 19 °C/W of baseline PCB thermal resistance that Rockchip list in their spec sheet, and we are looking at a punishing 76 °C temperature rise above ambient, which could push your CPU temperatures above 100°C very quickly (which corresponds exactly to what I am seeing in free air without a heatsink). Over 100 °C is NOT a healthy operating temperature for a 28nm geometry SOC to start with, and If you place the Rock64 in a case, this situation gets worse, and it will quickly either throttle to protect itself, or ramp up to the 120°C destruction point and shut down. THIS IS WHY I SAID THAT THE IDEA THAT THE ROCK64 DOESN'T NEED A HEATSINK IS SILLY.
With my large heatsink cutting the CPU junction to ambient thermal resistance to roughly HALF, the temperature rise is also cut in half, so instead of temperature spiking to over 100C and throttling, I am seeing a much more reasonable 40°C to 55°C temperature rise above ambiant (depending on airflow), which with an ambient temperature of 30°C gives me a maximum CPU temp of 85°C. Technically I should be seeing even better results, but the effectiveness of the heatsink is reduced quite a bit by the stagnant lack of airflow and heat radiating up from the PCB. With even a tiny amount of airflow, I see temperatures in the low 60°C range, even at peak CPU loads.
