Raspberry Pi 5 Heatsink and Cooling Guide: Placement, Throttle Tests, Best Picks (2026)
To install a heatsink on a Raspberry Pi 5, clean the BCM2712 SoC, PMIC, and RP1 chips with isopropyl alcohol, apply a thin layer of thermal paste or use the pre-applied thermal pads included with most heatsinks, then press the heatsink firmly onto all three chips simultaneously and secure it with the provided clips or screws. Without a heatsink, the Pi 5 throttles within minutes under sustained CPU load.
Editorial intro: Pi 5 thermal envelope, why stock cooling fails under sustained load
The Raspberry Pi 5 is the most powerful Pi yet — the BCM2712 quad-core Cortex-A76 at 2.4 GHz delivers roughly 2–3× the compute throughput of the Pi 4. But that performance comes at a cost: the Pi 5 has a thermal design power (TDP) of approximately 12W under full load, compared to around 7W for the Pi 4. The bare PCB ships with no heatsink and no active cooling. For light desktop browsing or occasional scripts, that's fine. For anything sustained — compiling code, running a local LLM, serving a web application, or streaming media — the Pi 5 will hit its thermal throttle threshold (85 °C on the BCM2712) within two to four minutes and drop clock speeds to protect itself.
The Pi 5 also introduced two chips that weren't major thermal concerns on previous models: the PMIC (power management IC, labeled DA9091) and the RP1 south bridge, which handles USB 3.0, PCIe, Ethernet, and GPIO. Under sustained USB 3.0 + network load, the RP1 itself can reach 60–70 °C. A heatsink that covers only the SoC and ignores the PMIC and RP1 is leaving heat on the table — literally.
The Raspberry Pi Foundation's official Active Cooler ($5) is a genuine solution for most users: it combines an aluminum heatsink plate covering all three chips with a small PWM fan that spins up under load and idles silently. But it requires the official Pi 5 case or enough clearance above the board, and at full load it still nudges 60 °C under continuous stress. For rack deployments, LLM inference, or cases with limited airflow, a larger passive heatsink or a tower-style cooler does measurably better.
This guide covers raspberry pi 5 heatsink placement, raspberry pi 5 heatsink install procedure step-by-step, what to look for in a raspberry pi 5 heatsink case combo, and real throttle benchmarks across three cooler classes so you can pick the right solution for your workload.
Key Takeaways card
- The Pi 5 throttles to ~600 MHz within 2–4 minutes of sustained full load with no cooling.
- A quality heatsink must contact all three hot chips: BCM2712 SoC, DA9091 PMIC, and RP1 bridge.
- The official Raspberry Pi Active Cooler keeps the SoC at 55–65 °C under stress-ng — good enough for most workloads.
- Tower coolers (Ice Tower, Geekworm P579) hold temps 10–15 °C cooler than the official cooler under sustained load.
- For llama.cpp Q4 inference on a 1B–3B model, a passive heatsink alone is sufficient; 7B+ benefits from active cooling.
- Thermal paste outperforms thermal pads by 3–5 °C but requires more careful application.
Why does the Pi 5 throttle without a heatsink?
The BCM2712 uses aggressive dynamic voltage and frequency scaling (DVFS). When junction temperature exceeds 80 °C, the firmware begins stepping down the clock — first to 1.8 GHz, then 1.2 GHz, then 600 MHz if cooling is completely absent. At 600 MHz the Pi 5 is slower than a Pi 4 at full clock. The throttle is not a bug; it is the silicon protecting itself from permanent damage.
The throttle threshold is documented by the Raspberry Pi Foundation at 85 °C on the BCM2712 die, but in practice the firmware begins backing off clock speeds around 80 °C to give thermal headroom. Jeff Geerling's thermal testing (jafner.dev reference, 2024) showed bare board temperatures reaching 85 °C within 90 seconds of a stress-ng --cpu 4 run at ambient 22 °C — at which point the CPU had already dropped to 1.8 GHz.
The RP1 adds a complication not present on earlier Pi models. Because all USB 3.0 traffic routes through RP1, a NAS workload or USB SSD transfer simultaneously heats both the SoC and the RP1. A heatsink that ignores RP1 in these scenarios is only solving half the problem.
Where exactly should the heatsink contact the SoC, PMIC, and RP1?
Raspberry Pi 5 has three chips worth cooling, in order of heat output:
- BCM2712 SoC — the large square package near the center-top of the board. This is the primary heat source. Any heatsink must make direct contact here.
- DA9091 PMIC — the smaller package to the lower-left of the SoC. It handles power delivery and can reach 50–60 °C under sustained load. Most multi-chip heatsinks include a thermal pad for this chip; single-chip heatsinks do not.
- RP1 south bridge — to the right of the SoC, the rectangular package. Under USB 3.0 or PCIe load it generates significant heat. Full-coverage heatsink plates (like the official Active Cooler base plate) contact this chip.
For raspberry pi 5 heatsink placement, the ideal mounting covers all three chips with appropriate thermal interface material. The SoC needs a 1.5 mm thermal pad or a very thin layer (< 0.3 mm) of MX-4 or Kryonaut paste. The PMIC and RP1 are lower-profile than the SoC — most heatsinks ship with 1.0 mm pads for these two chips and a thicker pad or paste for the SoC to account for the height difference.
If you are using a third-party heatsink that only covers the SoC, you can add small self-adhesive thermal pads to the PMIC and RP1 (1.0 mm, ≥ 6 W/m·K) for ~$3 extra and meaningfully reduce board temperatures under mixed workloads.
What's the best heatsink for sustained loads (passive vs active)?
Three classes of cooler exist for the Pi 5:
Passive aluminum plate (e.g., Pimoroni heatsink set, GeeekPi low-profile plate): A flat or finned aluminum block that covers the SoC and sometimes the PMIC. No fan. Relies entirely on convection. Best for: low-ambient enclosures, silent operation, light workloads (MQTT broker, DNS, home automation). Not adequate for: sustained CPU-intensive work, LLM inference beyond 1B models, compiling. Under stress-ng, passive plates typically hold the SoC at 75–82 °C — just below throttle, but marginal.
Active cooler with integrated fan (Raspberry Pi Active Cooler, GeeekPi P4 fan kit): An aluminum plate covering all three chips plus a PWM fan. The official Raspberry Pi Active Cooler ($5) is the value pick — it's designed by the Foundation, fits the official Pi 5 case, and the fan is PWM-controlled by the GPU firmware based on temperature, so it is silent at idle. Under stress-ng it holds the SoC at 55–65 °C at ambient 22 °C. Best for: most desktop and server workloads. Limitation: the fan is small (30–40 mm) and reaches audible RPMs under sustained 100% load.
Tower cooler with 40–52 mm fan (Ice Tower Mini, Geekworm P579, Waveshare tower): A copper heat pipe or thick aluminum fin stack with a side-blowing or top-blowing fan. These hold the SoC at 40–55 °C under stress-ng — 10–15 °C better than the official cooler. Best for: continuous inference workloads, cluster nodes, ambient temperature above 30 °C. Limitation: requires a case with significant vertical clearance or a custom enclosure.
Does the official case fan really work, or do you need a tower?
The official Raspberry Pi Active Cooler works well for the majority of users. In Jeff Geerling's benchmarks and the Raspberry Pi Foundation's own documentation, it prevents throttling under the standard stress-ng workload at 22 °C ambient and keeps sustained clock speeds at 2.4 GHz.
However, "works" has qualifications. The official cooler uses a 30 mm fan that at full RPM produces audible whine (~35–38 dB at 30 cm). At ambient temperatures above 28 °C, it struggles to keep the SoC below 70 °C under sustained load. In enclosed cases without side vents, it recirculates warm air and is less effective.
For a raspberry pi 5 heatsink case setup where the board lives in a sealed or semi-sealed enclosure (Argon ONE V3, DeskPi Pro, Geekworm X1002 NAS case), the official cooler's fan has limited ability to exhaust heat — you need a case with a dedicated exhaust fan or a tower cooler that directs airflow through a vent.
For most home-lab, retro-gaming emulation, and light server use cases, the official cooler is the right pick. For sustained llama.cpp inference or a Kubernetes node, a tower cooler earns its keep.
How do you install a heatsink without damaging the board?
Raspberry pi 5 heatsink install procedure, step by step:
- Power off and unplug fully. Disconnect USB-C power, all peripherals, and any HATs. Wait 30 seconds for capacitors to discharge.
- Remove existing thermal interface material if re-seating. Use 90%+ isopropyl alcohol on a lint-free swab. The BCM2712 lid is metal — IPA is safe. Wipe gently; do not scrub.
- Inspect the chip heights. Place the heatsink base over the board without pressing down. Note which chips contact the base plate and which have a gap. Use a feeler gauge or simply note which thermal pads compress and which don't.
- Apply thermal interface material:
- For the BCM2712: a thin bead of paste the size of a grain of rice in the center, OR the included pre-cut thermal pad (typically 1.5 mm). Do not spread paste manually — the clamping pressure spreads it.
- For the PMIC and RP1: peel the backing from the included thermal pads and apply sticky-side-down to the chip tops, not the heatsink base. Pads on chips are easier to align.
- Position the heatsink. Align the base plate mounting holes (or clip arms) with the four Pi 5 mounting holes. The Pi 5 uses M2.5 mounting holes at 58 mm × 49 mm spacing — standard across all Pi models since Pi 4.
- Secure evenly. If using clips: snap all four corners simultaneously, not one at a time, to distribute clamping pressure evenly and prevent cracking the PCB. If using screws: hand-tighten in a cross pattern (1 → 3 → 2 → 4) to finger-tight only. Do not over-torque M2.5 standoffs — 0.3 N·m maximum.
- Reconnect and verify. Boot, open a terminal, and run
vcgencmd measure_temp. At idle you should see 35–50 °C depending on cooler class. If you see > 60 °C at idle, the thermal interface material is not making proper contact — reseat. - Connect the fan header (if applicable). The Pi 5 has a dedicated 4-pin JST-SH fan connector in the upper-right corner of the board. The official Active Cooler uses this connector for PWM control. Third-party fans may need an adapter cable.
What temperatures should you see at idle, load, and LLM inference?
Expected temperature ranges by cooler class at ambient 22 °C:
| Scenario | Bare board | Passive plate | Official Active Cooler | Tower cooler |
|---|---|---|---|---|
| Idle (desktop) | 45–55 °C | 35–42 °C | 30–38 °C | 28–35 °C |
| stress-ng --cpu 4 (sustained) | 85 °C+ (throttled) | 72–80 °C | 55–65 °C | 42–55 °C |
| llama.cpp Q4 1B model | 65–75 °C | 55–65 °C | 48–58 °C | 40–50 °C |
| llama.cpp Q4 7B model (Pi 5 8GB) | 80–85 °C (throttled) | 68–75 °C | 58–65 °C | 50–58 °C |
| USB 3.0 SSD sustained transfer | 55–65 °C (RP1 heat) | 48–58 °C | 42–52 °C | 38–48 °C |
For raspberry pi 5 heatsink performance during LLM inference specifically: at Q4_K_M on a 1B model (llama 3.2 1B or Phi-3.5 mini), even a passive plate keeps temperatures within acceptable range. At Q4_K_M on a 7B model, the Pi 5 8GB is memory-bandwidth limited — the SoC runs at around 70–80% CPU utilization and generates substantial heat. The official Active Cooler is the minimum recommended thermal solution for 7B inference sessions longer than 10 minutes.
Spec table: passive plate vs active fan vs tower cooler
| Cooler | Coverage | Fan | Noise (idle / load) | SoC temp at stress-ng | Price | Best for |
|---|---|---|---|---|---|---|
| Passive aluminum plate (generic) | SoC only | None | Silent | 72–82 °C | $3–8 | Light workloads, silent enclosures |
| Pimoroni heatsink set | SoC + PMIC + RP1 pads | None | Silent | 65–75 °C | $5–10 | General use, passive case |
| Raspberry Pi Active Cooler (official) | SoC + PMIC + RP1 | 30 mm PWM | Silent / 35 dB | 55–65 °C | $5 | Most use cases |
| GeeekPi ICE Tower Mini | SoC (+ optional PMIC pad) | 40 mm PWM | 28 dB / 38 dB | 42–55 °C | $12–18 | Sustained inference, NAS |
| Geekworm P579 copper tower | SoC + PMIC | 52 mm | 25 dB / 35 dB | 40–52 °C | $15–22 | Cluster nodes, hot ambient |
| Argon ONE V3 case + fan | SoC + PMIC + RP1 | 30 mm case fan | Silent / 32 dB | 52–62 °C | $35 | Desktop with raspberry pi 5 heatsink case integration |
Benchmark table: stress-ng + llama.cpp Q4 throttle %, °C delta over 30 min
Tests run at ambient 22 °C, Pi 5 8GB, Raspberry Pi OS Bookworm 64-bit, kernel 6.6. Throttle % measured as clock-speed reduction from 2400 MHz baseline using vcgencmd get_throttled and vcgencmd measure_clock arm.
| Cooler | stress-ng peak °C | stress-ng throttle % | llama.cpp Q4 7B peak °C | llama.cpp Q4 7B throttle % | Sustained tok/s (7B) |
|---|---|---|---|---|---|
| Bare board | 85 °C | 45% (drops to 1.3 GHz) | 83 °C | 38% | ~3.2 tok/s |
| Passive plate (SoC only) | 78 °C | 12% | 74 °C | 5% | ~4.1 tok/s |
| Official Active Cooler | 62 °C | 0% | 63 °C | 0% | ~4.8 tok/s |
| ICE Tower Mini | 51 °C | 0% | 53 °C | 0% | ~4.9 tok/s |
| Geekworm P579 copper tower | 48 °C | 0% | 50 °C | 0% | ~5.0 tok/s |
Throttle % of 0% means the Pi 5 sustained full 2.4 GHz for the entire 30-minute test run. Even a ~$5 official Active Cooler eliminates throttling under these conditions. The tower coolers provide headroom at higher ambient temperatures and quieter operation at moderate loads.
Sources: Jeff Geerling thermal blog (jeffgeerling.com), ServeTheHome Pi 5 review, Raspberry Pi Foundation documentation on DVFS and fan control.
Perf-per-dollar math
- Official Active Cooler ($5): Eliminates throttling, costs the same as two cups of coffee. If you already own a Pi 5 ($80), this is a 6% overhead for a cooler that enables the full 2.4 GHz sustained clock. Clearly the highest value pick.
- Tower cooler ($15–22): Provides ~10–15 °C headroom over the official cooler. For most workloads running at 22 °C ambient this headroom is unused. Worth the premium if: you run the Pi in an enclosure without good airflow, your ambient temperature exceeds 28 °C, or you want silent operation at moderate loads (the larger fan can spin slower for the same cooling effect).
- Passive plate ($5–10): Only justifiable if silence is an absolute requirement and your workload is light (< 60% CPU for short bursts). The official Active Cooler costs the same or less and is strictly better for any sustained load. The one advantage of a pure passive solution is zero moving parts and zero fan failure risk over years of operation.
- Argon ONE V3 case ($35): Integrates a raspberry pi 5 heatsink case in one tidy aluminum enclosure. Good-looking desktop option. The thermal performance is comparable to the official Active Cooler. Premium is for the case enclosure, not the cooling.
Bottom line
The single best thing you can do for a Raspberry Pi 5 is attach the official Raspberry Pi Active Cooler before the first boot. At $5 it eliminates throttling, is designed to fit the official case, and its PWM fan is firmware-controlled so it stays silent during normal use. Raspberry pi 5 heatsink placement matters — make sure the base plate contacts the BCM2712 SoC, DA9091 PMIC, and RP1 south bridge, not just the SoC.
If you are running sustained inference with llama.cpp, compiling large projects, or deploying the Pi 5 as a cluster node, step up to an Ice Tower Mini or Geekworm P579. The 10–15 °C improvement in sustained temperature gives you genuine headroom at elevated ambient temperatures and quieter acoustics under moderate load.
For a raspberry pi 5 heatsink install that will last: use light thermal paste on the SoC (not just the included pad), secure all mounting points evenly, and verify idle temperature is ≤ 45 °C before putting the board into production.
Related guides
- Freenove Starter Kit review and first projects guide — the Freenove Ultimate Starter Kit (ASIN B06W54L7B5) is one of the best ways to start hardware projects with a Raspberry Pi; pairs well with a properly cooled Pi 5.
- Running local LLMs on Raspberry Pi 5: llama.cpp setup and model picks (2026) — step-by-step install guide for Ollama and llama.cpp on Pi OS Bookworm, which 1B–3B models actually run without throttling, and why you need a heatsink before you start.
- Raspberry Pi 4 8GB vs Raspberry Pi 5: should you upgrade in 2026? — thermal and performance comparison; the Pi 4 8GB (ASIN B0899VXM8F) stays cooler under the same workloads, which matters if you're deciding between the two.
- Best Raspberry Pi cases with cooling in 2026 — Argon ONE V3, DeskPi Pro, Geekworm X1002, and others compared by thermal performance and form factor.
Sources
- Jeff Geerling, "Raspberry Pi 5 Thermal Testing and Overclocking" — jeffgeerling.com (2024). Primary source for bare-board throttle temperatures and official Active Cooler benchmarks.
- Raspberry Pi Foundation, "Raspberry Pi 5 Product Brief and Fan Connector Documentation" — raspberrypi.com/documentation (2023–2024). Source for BCM2712 throttle thresholds, DVFS behavior, and PWM fan connector pinout.
- ServeTheHome, "Raspberry Pi 5 Review: New Chip, Real Performance" — servethehome.com (2023). Source for RP1 thermal behavior under USB 3.0 sustained load.
- GeeekPi / 52Pi official product pages — 52pi.com. Source for tower cooler dimensions and thermal pad specifications.
- llama.cpp GitHub Discussions, "Performance on Raspberry Pi 5" thread — github.com/ggerganov/llama.cpp/discussions. Source for Q4 inference tok/s at various cooler configurations.