MSI Crosshair 15 // Cooling Mod Guide

This document contains the deep-dive build guides, complete parts lists, and risk mitigation strategies for all four major cooling modifications. As your MSI Crosshair 15 (i9-12900H, RTX 3070 Ti) is being used as a desktop replacement, these guides assume you are comfortable permanently altering the chassis.

#Part 0: System Baseline (Ubuntu, 2026-09-06)

Before any mod, these are the numbers this specific machine produced under light desktop load (a browser and a DisplayLink dock, load average 0.8). Everything below was read from the running Ubuntu system, not from Windows tools.

##The Machine

##Live Thermal Snapshot

##What The Numbers Say

• The thermal interface has failed or the heatsink is not seated evenly. Ninety-six degrees at idle with both fans at full speed and a 29 °C gap between cores on the same die is the signature of pumped-out or dried factory paste, or a heatsink that only touches part of the die. More airflow (Mod 1) cannot fix this. Do Mod 4 (or at minimum Mod 0 step 1, the dust clean) before anything else.
• The 30 W / 45 W power limits are a symptom, not a setting. No MSI performance mode on this platform is that low: Notebookcheck measured a 121 W PL2 on the sister Crosshair 15 R6E in every mode [S1]. The EC has clamped power in response to its own thermal telemetry. Raising the limits in software before fixing the cooling will only push temperatures higher; raise them after Mod 4 (Part 5b) and expect the EC to keep re-asserting them until the sensors calm down.
• Check for bi-directional PROCHOT. MSI firmware can throttle the CPU because the GPU or VRM sensor is hot, even when the CPU die is not [S6][S18]. The throttle counter climbing at idle with a 45 °C GPU makes a CPU-side cause more likely here, but the combined soak in Part 6b is what settles it.
• The GPU is fine right now because it is idle. Its 115 W limit and the shared heatpipes mean it will push the CPU hotter under gaming load, so validate both together (Part 6b).
• Re-capture after every change with tools/thermal_baseline.sh 2 <label> (Part 6b) so every mod has a before and after CSV.

#Mod 0: The Zero-Hardware Software Path

Everything here is free, reversible and done from a terminal on Ubuntu. Do it in this order, re-logging with thermal_baseline.sh after each step, and stop as soon as the idle package temperature is sane. If step 1 alone drops idle by 10 °C or more, the problem was dust and Mod 4 can wait.

##Options

• Option A: Dust clean (do this first, $0): Bottom cover off, compressed air out through the fin stacks from the inside, fans held still with a finger. Generic guides put dust alone at 10 to 15 °C on a 2 to 4 year old gaming laptop; combined with a repaste 20 to 30 °C [S27]. Low-medium confidence on the numbers, high confidence it costs nothing.
• Option B: Set the EC modes from Linux: This box runs the out-of-tree msi-ec DKMS driver, version 0.13 [L1]. Its config for firmware 1583EMS1 knows three shift modes (eco, comfort, turbo) and three fan modes (auto, silent, advanced); there is no custom fan-curve attribute in any msi-ec version [L3]. The unknown (192) reading is EC value 0xC0, which is "sport" in the driver's other config tables but missing from this model's table, so the EC is most likely in its default sport mode and the driver simply cannot name it (medium confidence, read from the driver source).

  cat /sys/devices/platform/msi-ec/available_shift_modes      # eco comfort turbo
  echo turbo   | sudo tee /sys/devices/platform/msi-ec/shift_mode    # after Mod 4; use comfort until then
  echo silent  | sudo tee /sys/devices/platform/msi-ec/fan_mode      # auto | silent | advanced
  echo off     | sudo tee /sys/devices/platform/msi-ec/cooler_boost  # it is stuck on right now

Revert: write the previous value back. msi_wmi_platform (in-tree, gives the fan RPM) shares the same EC memory [L7]; leave it loaded for the RPM readings, but do not run two tools that write the EC at once.

• Option C: Cap the CPU power and calm the governor: Alder Lake-H keeps most of its multi-core performance up to about 65 to 75 W; above that the extra watts are mostly heat [S7]. This machine is already at 30 / 45 W by the EC's choice, so today the useful move is the opposite: stop the sawtooth, not cut further. EPP is already performance here, which fights the powersave governor; set it to balanced until the mount is fixed.

  echo balance_performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
  echo 1 | sudo tee /sys/devices/system/cpu/intel_pstate/no_turbo      # kill turbo spikes; revert with 0
  sudo cpupower frequency-set -u 3.0GHz                                # or cap the ceiling; revert with -u 4.9GHz

[L39] [L35]

• Option D: Frame-rate cap in games: CPU and GPU power scale with rendered frames; one Overclock.net report saw roughly 50 W less CPU package power from a cap in a CPU-bound title [S29]. Set it per game or install MangoHud (fps_limit=); it is not installed here yet.
• Option E: Drive the desktop from the Intel iGPU: Right now Xorg lists NVIDIA-0 as the source output, so the 3070 Ti draws the desktop and idles at 25 W. Switching to PRIME render offload (iGPU primary, NVIDIA as an offload GPU, runtime D3 already enabled on this driver) removes that heat and the GPU's share of the heatpipes [L26] [L27]. This is a display-stack change: on MSI it usually also means picking hybrid mode in the BIOS. It does not touch the CPU idle problem, so it is last on the list.
• Option F: Pin the browser to E-cores: taskset -c 12-19 brave keeps a chatty browser off the P-cores that are hitting 96 °C. Anecdotal benefit, zero risk.

##What not to expect

• CPU voltage undervolting does not exist on this CPU. Non-HX 12th-gen mobile parts ship with the voltage-offset MSR locked; Notebookcheck's summary is that Intel and the OEMs killed it [S3] [L33]. The hidden-BIOS unlock in Part 5 is unverified on BIOS E1583IMS.112 and may show the toggles without them working.
• Cooling pads do not fix idle temperatures. Every pad and vacuum-cooler result found is a load-temperature result [H9]. If idle is 96 °C, no fan under the laptop changes that.

#Alternatives to Modding

Not every fix is a mod. These are the routes that avoid cutting plastic, ranked by how much of the problem they actually address on this machine.

When each beats the mods: the shop repaste beats Mod 4 only on time, not result. Replacement beats Mods 2 and 3 on cost once you count the tools and the risk of a cracked die. Nothing in this table beats Mod 1 for a desktop-replacement machine that is already repasted, because a sealed blower is cheap, reversible and needs no disassembly.

#Tiers and Decision Matrix

Every option in this guide, ordered by how much you have to commit. Expected gains are for sustained load on this chassis after a good mount; "idle fix" says whether the option addresses the 96 °C idle problem from Part 0. Costs are 2026 street prices, unverified for your region.

##The Ladder

##Decision Matrix

Recommended path for this machine: T0 dust clean and re-log; T1 repaste with a mount-pressure check (Mod 4 A + E) and re-log; then, because it lives on a desk, T1 vacuum cooler or T2 sealed pad; then Part 5b to lift the power limits and Part 6b to prove it. Only if the combined soak still throttles does T4 make sense.

#Part 5b: Linux Power Tuning

Part 5 assumes Windows. This is the same tuning on Ubuntu 24.04 with what actually works on this hardware, verified locally where marked. Every command needs root and has a revert line; nothing here persists across a reboot unless you add the systemd unit at the end.

##Tool Map

##1. Lift the CPU power limits (after Mod 4 only)

The kernel exposes PL1 as constraint_0 (long term, 30 W now), PL2 as constraint_1 (short term, 45 W, 2.4 ms window) and a peak limit constraint_2 (215 W). Writes take effect immediately and are lost on reboot.

R=/sys/class/powercap/intel-rapl:0
echo 45000000  | sudo tee $R/constraint_0_power_limit_uw    # PL1 45 W (Intel spec)
echo 90000000  | sudo tee $R/constraint_1_power_limit_uw    # PL2 90 W; the sister R6E chassis sustains 121 W with a healthy cooler [S1]
# revert: 30000000 and 45000000

Then watch whether it sticks. Two things can pull it back down:

• thermald. It is active here, with a custom /etc/thermald/thermal-cpu-cdev-order.xml (backup dated 2026-09-05), and RAPL is one of its cooling devices, so it will re-clamp PL1 when its trip points fire [L36]. Either raise the trip in thermal-conf.xml or, for testing only, sudo systemctl stop thermald and compare.
• The EC. No source documents the MSI EC re-asserting RAPL by itself; on Ubuntu that job is thermald's via DPTF. If the limit still drops with thermald stopped, the EC is doing it and the throttled daemon, which rewrites the limits on a timer, is the workaround [S8]. RAPL throttling does not show in package_throttle_count; watch PkgWatt in turbostat instead [L46].

Persist with a oneshot unit once the values are proven:

# /etc/systemd/system/rapl-limits.service
[Unit]
Description=Set RAPL PL1/PL2
After=multi-user.target thermald.service
[Service]
Type=oneshot
ExecStart=/bin/sh -c 'echo 45000000 > /sys/class/powercap/intel-rapl:0/constraint_0_power_limit_uw; echo 90000000 > /sys/class/powercap/intel-rapl:0/constraint_1_power_limit_uw'
[Install]
WantedBy=multi-user.target

sudo systemctl enable --now rapl-limits. Add a copy under /usr/lib/systemd/system-sleep/ if the limits reset after suspend. Do not install TLP and auto-cpufreq together; they fight over the same knobs [L38].

##2. Governor and EPP

intel_pstate is in active mode with HWP, so scaling_governor is really an EPP hint, not a clamp [L39]. For a repasted desktop-replacement:

sudo cpupower frequency-set -g performance
echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
# revert: -g powersave, balance_performance

##3. GPU

• Power limit: nvidia-smi -pl 140 returns "not supported" on RTX 30 laptop GPUs on current drivers; the 140 W "max" the driver reports is Dynamic Boost headroom, not a user setting [L18]. Dynamic Boost needs nvidia-powerd, which is not installed here (systemctl status nvidia-powerd says not found). Installing it is the only Linux route to the extra 25 W, and it is forum-grade evidence that it delivers.
• Clocks: nvidia-smi -lgc is also commonly ignored on laptops. What works is the Coolbits offset already configured on this box:

  nvidia-settings -a '[gpu:0]/GPUGraphicsClockOffsetAllPerformanceLevels=150'
  nvidia-settings -a '[gpu:0]/GPUMemoryTransferRateOffsetAllPerformanceLevels=1000'   # +500 MHz effective
  # revert: set both to 0

There is no voltage offset on Ampere under the Linux driver; "undervolting" a laptop GPU on Linux means a positive clock offset so the same clock lands at a lower voltage point on the curve, or nothing [L23].

• Hotspot: nvidia-smi --query-gpu=temperature.gpu.tlimit is a valid key on driver 595.84 but returns N/A on this GPU (checked). Only the core temperature is available on Linux, so the Part 6 hotspot-delta check becomes the core-spread check in Part 6b.
• VBIOS flashing: nvflash for Linux exists but the higher-TGP ROM advice in Part 5 is Windows-community lore with no Linux reports; unverified and unchanged in risk.

##4. Fan control per mod

msi-ec offers auto, silent and advanced; no curve. Mod 1 with a sealed blower: advanced gives the internal fans a fixed floor so the pad cannot over-spin them. Mod 3: silent. If you need a real curve, the unmaintained msi-fanctl project claims to write the EC curve table directly [L12]; single source, unverified, and it writes the same EC the two kernel drivers use, so test it with both unloaded.

##Unverified on this firmware

• The hidden BIOS menu and its OC Lock / CFG Lock / Undervolt Protection toggles on E1583IMS.112.
• Any undervolt tool on Alder Lake-H after that unlock.
• Whether the EC re-asserts PL1/PL2 independently of thermald.
• Write conflicts between msi-ec and msi_wmi_platform on this board.
• LACT's power-limit and offset features on a mobile GA104 with driver 595.84.

#Part 6b: Linux Validation Suite

The Windows suite in Part 6 (HWiNFO64, Cinebench, 3DMark) does not exist on Ubuntu. This is the equivalent, built around the thermal_baseline.sh logger that ships next to this report. Every mod gets a labelled CSV before and after, so the decision matrix in this guide can be checked against your own numbers rather than someone else's.

Do not run the stress protocol on this machine until Mod 4 (or at least the Mod 0 dust clean) is done. The baseline already shows 96 °C at idle. A 10-minute all-core load in that state will hit the 100 °C trip point and the EC will hard-shut the laptop.

##Install

sudo apt install stress-ng s-tui lm-sensors linux-tools-common   # turbostat lives in linux-tools
# already present on this box: nvtop, glmark2, sensors, turbostat

##The Logger: thermal_baseline.sh

Samples every N seconds into logs/thermal_<label>_<timestamp>.csv and prints nothing until you press Ctrl-C, when it prints a one-line summary. Columns: package °C, min/max core °C, core spread (the Linux stand-in for HWiNFO's hotspot delta), average CPU MHz, both fan RPMs, package watts, the kernel's package throttle counter, GPU °C / W / SM MHz / utilisation.

tools/thermal_baseline.sh 2 stock-idle          # 2 s interval, label "stock-idle"; Ctrl-C to stop
sudo tools/thermal_baseline.sh 2 after-mod4     # sudo unlocks RAPL package watts (energy_uj is root-only)

Sample output on this laptop, stock, seven seconds at the desktop:

samples=7  pkg avg=75.0C max=90C  worst core spread=18C  throttle delta=40  gpu max=44C

A throttle delta of 40 in seven seconds at idle is the headline problem; after a good repaste it should be zero for the whole run.

##The Protocol

Run each stage with the logger in a second terminal, one label per stage. Keep the room temperature and the laptop position the same between before and after runs.

1. Idle, 5 min. Nothing open but the terminal. Establishes the floor.
2. CPU soak, 10 min. All 20 threads, a heavy floating-point kernel, so turbo and PL2 are fully exercised:

   stress-ng --cpu 0 --cpu-method matrixprod --tz --metrics-brief --timeout 10m   # --cpu 0 = one worker per thread; --tz prints thermal zones

Jarrod'sTech's published method is three 10-minute passes averaged, so a single pass understates the steady state [S12].

3. GPU soak, 10 min. OpenGL loop that pins the 3070 Ti at its power limit:

   glmark2 --run-forever --fullscreen      # light load; vkmark is the Vulkan twin
   # heavier and closer to 3DMark: Unigine Superposition, or FurMark 2's native Linux build

Watch clocks and power with nvtop or nvidia-smi dmon -s pucT alongside.

4. Combined, 10 min. Stages 2 and 3 at once. This is the desktop-replacement worst case and the only stage that reveals shared-heatpipe problems.
5. Live view while it runs: sudo s-tui (temps, clocks, power, throttle flags in one screen) or sudo turbostat --show Package,Busy%,Bzy_MHz,PkgTmp,PkgWatt --interval 5.

##Pass Criteria

##Reading Failures

• Low temps but clocks drop under combined load only: VRM or VRAM heat. The putty in Mod 4 was too thin, or nothing is blowing across the board. Add a fan or re-pad.
• Package fine, one core 20 °C hotter: The heatsink is tilted. Redo the mount in a cross pattern, or add the washer mod from Mod 4.
• Watts capped below PL1 with cool temps: Software, not hardware. Check thermald, the EC shift mode, and RAPL as described in Part 5b.
• Temps climb slowly for 10 minutes and never plateau: Not enough heatsink mass or airflow for the sustained load. That is what Mods 1 to 3 solve; Mod 4 alone cannot.

#Sources

Citations in the text use the prefix of the research pass that found them: H hardware alternatives, L Linux tooling, S software-only path and validation. The full digests, with per-claim confidence ratings and the lists of claims that were dropped as unsupported, are in research/ next to this report.

##Hardware alternatives

• [H1] Amazon — IETS GT600 product listing: https://www.amazon.com/IETS-Equipped-Turbo-Fan%EF%BC%885-5inch-Diameter%EF%BC%89-14-1-19-3/dp/B0CDC3KBC1
• [H2] shantoreview.com — IETS GT600 RGB review: https://shantoreview.com/iets-gt600-rgb-laptop-cooling-pad-review/
• [H3] tensorscience.com — IETS GT600 hands-on notes: https://www.tensorscience.com/cooling/my-thoughts-on-using-the-iets-gt600-turbo-fan-laptop-cooling-pad-with-usb-hub
• [H4] techreviewz.blog — IETS GT500 review (2026): https://techreviewz.blog/iets-gt500-review/
• [H5] Amazon — Llano V12 (RGB) listing: https://www.amazon.com/llano-V12-Gaming-Laptop-Cooling/dp/B0C69BVWGB
• [H6] craftingworlds.com — Llano V12 Ultra review: https://craftingworlds.com/llano-v12-ultra-laptop-cooling-pad-review-a-laptop-cooler-that-actually-works/
• [H7] medium.com (Muteeb Hussain) — KLIM Ultimate review: https://medium.com/@muteeb.com/klim-ultimate-rgb-laptop-cooling-pad-with-led-rim-review-17c98b4b00f7
• [H8] Amazon — KLIM Ultimate listing: https://www.amazon.com/KLIM-Gaming-Compatible-17-inch-Laptops/dp/B07NNQXTQT
• [H9] Notebookcheck — "How well does a laptop cooling pad work?" (2020): https://www.notebookcheck.net/How-well-does-a-laptop-cooling-pad-work-We-Amazon-d-one-ourselves-to-find-out.464231.0.html
• [H10] Amazon — IETS GT202UB vacuum cooler listing: https://www.amazon.com/IETS-temperature-intelligent-measurement-2600-5000RPM/dp/B0991V82S3
• [H11] Tom's Guide forum — notebook exhaust vacuum fan thread: https://forums.tomsguide.com/threads/notebook-exhaust-vacuum-fan-any-thoughts.364354/latest
• [H12] Apex Tech — Honeywell PTM7950 product/spec page: https://apextech101.com/product/ptm-7950/
• [H13] Overclock.net — "PTM7950 or liquid metal on direct die?" thread: https://www.overclock.net/threads/ptm7950-or-liquid-metal-on-direct-die.1811663/
• [H14] Framework — PTM7958 thermal pad store page: https://frame.work/products/ptm7958-thermal-pad?v=FRAMCZ0001
• [H15] MODDIY — Honeywell PTM7958 SP paste, 1kg listing: https://www.moddiy.com/products/6702/Honeywell-PTM7958-SP-Super-Highly-Thermally-Conductive-PCM-Paste-1KG.html
• [H16] Igor'sLab — Thermal Grizzly KryoSheet lab/field test: https://www.igorslab.de/en/thermal-grizzly-cryosheet-in-lab-and-field-test-durable-all-purpose-weapon-with-minor-limitations/6/
• [H17] Tom's Hardware forum — KryoSheet graphene cooling pad discussion: https://forums.tomshardware.com/threads/thermal-grizzly-kryosheet-graphene-cooling-pad.3808893/
• [H18] Thermal Grizzly — official KryoSheet product page: https://www.thermal-grizzly.com/en/kryosheet/s-tg-ks-24-12
• [H19] Overclock.net — "PTM7950 vs Liquid metal vs Kryosheet" thread: https://www.overclock.net/threads/ptm7950-vs-liquid-metal-vs-kryosheet.1808499/
• [H20] Igor'sLab — 5-way phase-change pad roundup (PTM7950/7950SP/PCM5000/PCM8500/Heilos), Aug 2024: https://www.igorslab.de/en/5-phasenwechsel-pads-im-test-honeywell-ptm7950-ptm7950sp-vs-pcm5000-pcm8500-und-thermalright-heilos/
• [H21] Overclock.net — Thermalright Heilos thread: https://www.overclock.net/threads/thermalright-heilos-thermal-pads.1807207/
• [H22] e-catalog.com — Conductonaut Extreme vs Conductonaut spec comparison: https://e-catalog.com/cmp/142454/conductonaut-extreme-1g-vs-conductonaut-1g/
• [H23] LinusTechTips forum — "Thermal Grizzly PTM vs Conductonaut Extreme" thread: https://linustechtips.com/topic/1586126-thermal-grizzly-ptm-vs-conductonaut-extreme-liquid-metal/
• [H24] Overclock.net — "Beware of misleading information ... regarding Thermal Putty" thread: https://www.overclock.net/threads/beware-of-misleading-information-from-computer-systems-gr-regarding-thermal-putty.1809275/
• [H25] Overclock.net — Thermal Putty vs Thermal Pads comparison thread: https://www.overclock.net/threads/thermal-putty-vs-thermal-pads-comparison-chart.1810430/
• [H26] Computer Systems (Greece) — Upsiren U6 Pro product page: https://computer-systems.gr/products/upsiren-u6-pro-10g/
• [H27] LaptopMedia — "Inside MSI Crosshair 15 (B12Ux)" disassembly article (fetch blocked by site, cited via search snippet — re-verify manually): https://laptopmedia.com/highlights/inside-msi-crosshair-15-b12ux-disassembly-and-upgrade-options/
• [H28] Amazon — replacement dual fan assembly for Crosshair 15 B12U/B12UX/B12UGSZ/B12UGZ/B12UEZ: https://www.amazon.com/SYW%C2%B7pcparts-Dual-Cooling-Fans-Crosshair/dp/B0FWYQNDS7
• [H29] eBay — Crosshair 15 R6E B12UEZ/B12UGZ fan+heatsink part listing: https://www.ebay.com/itm/256862163018
• [H30] Amazon — heatsink part E32-2500871-HH7, listed compatible across Katana GF66/Pulse GL66/WF66/Creator M16/Crosshair 15: https://www.amazon.com/Heatsink-E32-2500871-HH7-Compatible-Replacement-Crosshair/dp/B0DNNK9V9X
• [H31] eBay (DE) — Pulse GL66/GL76, Katana GF66/GF76 heatsink+fan part listing: https://www.ebay.de/itm/116496256442
• [H32] nanoreview.net — MSI GE66 Raider vs Crosshair 15 comparison: https://nanoreview.net/en/laptop-compare/msi-ge66-raider-vs-msi-crosshair-15
• [H33] MSI Global Forum — "2nd M.2 screw hole not available and DC-in connector has a lot of play," Crosshair 15 C12VE, Sept 2023: https://forum-en.msi.com/index.php?threads/2nd-m-2-screw-hole-not-available-and-dc-in-connector-has-a-lot-of-play-msi-crosshair-15-c12ve.389373/
• [H34] MSI — MS-15P2 (Crosshair/GL) official service guide PDF: https://storage-asset.msi.com/global/pdf/ServiceGuide_15P2_Crosshair+(GL)_v1.0_English_0614.pdf
• [H35] Overclock.net — "Extra or thicker washers to increase heatsink GPU pressure" thread: https://www.overclock.net/threads/extra-or-thicker-washers-to-increase-heatsink-gpu-pressure.1775188/
• [H36] Tom's Hardware forum — "Increasing mounting pressure of a laptop's heatsink" thread: https://forums.tomshardware.com/threads/increasing-mounting-pressure-of-a-laptops-heatsink-what-do-you-think.2662481/
• [H37] Wikipedia — CPU shim: https://en.wikipedia.org/wiki/CPU_shim
• [H38] Tom's Guide forum — "General Copper Mod (Shim) Question": https://forums.tomsguide.com/threads/general-copper-mod-shim-question.276618/latest
• [H39] Tom's Hardware forum — "DIY Heatsink with copper tube for Laptop cooling?" thread: https://forums.tomshardware.com/threads/diy-heatsink-with-copper-tube-for-laptop-cooling.2619529/
• [H40] Overclockers UK forum — "DIY sub-ambient water chiller with dew point control" build log: https://forums.overclockers.co.uk/threads/i-built-a-diy-sub-ambient-water-chiller-with-dew-point-control.19011177/
• [H41] Overclock.net — "Sub Ambient condensation prevention" thread: https://www.overclock.net/threads/advice-sub-ambient-condesnation-prevention.1583708/
• [H42] Google Patents — US6205796B1, "Sub-dew point cooling of electronic systems": https://patents.google.com/patent/US6205796B1
• [H43] YouTube — "I Bolted a Desktop CPU Cooler to a GTX 1060 Laptop": https://www.youtube.com/watch?v=slLSCf4WP7g
• [H44] Overclockers.com — "Laptop Cooler Mod" article: https://www.overclockers.com/laptop-cooler-mod/
• [H45] Overclock.net — "Laptop cooling with peltier" thread: https://www.overclock.net/threads/laptop-cooling-with-peltier.1401395/
• [H46] Overclock.net — "Condensation on peltier" thread: https://www.overclock.net/threads/condensation-on-peltier.506120/
• [H47] PCWorld — Cooler Master thermoelectric closed-loop desktop CPU cooler coverage (2018, dated): https://www.pcworld.com/article/402079/cooler-master-thermoelectric-cooler.html

##Linux tooling

• [L1] https://github.com/BeardOverflow/msi-ec (repo)
• [L2] https://raw.githubusercontent.com/BeardOverflow/msi-ec/main/msi-ec.c (source, fetched directly and grepped for 1583EMS1)
• [L3] https://github.com/BeardOverflow/msi-ec/blob/main/README.md
• [L4] https://github.com/BeardOverflow/msi-ec/issues/398
• [L5] https://github.com/BeardOverflow/msi-ec/issues/108
• [L6] https://github.com/BeardOverflow/msi-ec/discussions/277
• [L7] https://docs.kernel.org/7.0/wmi/devices/msi-wmi-platform.html
• [L8] https://lkml.org/lkml/2025/5/19/69 (Armin Wolf, msi-wmi-platform fan-curve/platform-profile/tdp/battery patch series)
• [L9] https://lkml.iu.edu/2511.1/01820.html (msi-wmi-platform DMI/GUID matching discussion)
• [L10] https://www.phoronix.com/news/MSI-WMI-Platform-Driver-Linux
• [L11] https://github.com/dmitry-s93/MControlCenter
• [L12] https://github.com/YaxOFF/msi-fanctl
• [L13] https://www.igorslab.de/en/lact-0-10-0-nvidia-tuning-linux-voltage-boost-gddr7-temperatures/
• [L14] https://www.gamingonlinux.com/2026/08/linux-gpu-configuration-and-monitoring-tool-lact-adds-more-nvidia-gpu-overclocking-options/
• [L15] https://github.com/ilya-zlobintsev/LACT/issues/1114
• [L16] https://gitlab.com/leinardi/gwe/-/merge_requests/80
• [L17] https://gitlab.com/leinardi/gwe/-/issues/125
• [L18] https://forums.developer.nvidia.com/t/rtx-3080-mobile-power-limit-shows-n-a-but-worked-before/254925
• [L19] https://forums.developer.nvidia.com/t/changing-power-management-limit-is-not-supported-for-gpu-pascal-gtx1060-laptop-mobile-linux-555-nvidia-smi-powerlimit/298873
• [L20] https://forums.developer.nvidia.com/t/power-limit-on-3000-mobile-series/193443
• [L21] https://bbs.archlinux.org/viewtopic.php?id=302133
• [L22] https://forums.developer.nvidia.com/t/how-to-force-lock-sm-and-memory-clocks-on-rtx-5090-headless-linux/348794
• [L23] https://forums.developer.nvidia.com/t/nvidia-rt3070-eth-oc-in-ubuntu20-04-nvidia-settings-a-gpugraphicsclockoffset-gpumemorytransferrateoffset-not-working/178687
• [L24] https://forums.developer.nvidia.com/t/option-coolbits-is-not-used-optimus-enabled-laptop-running-an-rtx-2070-manjaro-linux/111771
• [L25] https://github.com/NVIDIA/open-gpu-kernel-modules/discussions/236
• [L26] https://download.nvidia.com/XFree86/Linux-x86_64/450.57/README/primerenderoffload.html
• [L27] https://wiki.archlinux.org/title/PRIME
• [L28] https://bbs.archlinux.org/viewtopic.php?id=250844
• [L29] https://forums.tomshardware.com/threads/voltage-control-suddenly-disabled-on-msi-laptop.3822367/
• [L30] https://forums.linuxmint.com/viewtopic.php?t=456082
• [L31] https://www.notebookcheck.net/XMG-reintroduces-BIOS-based-undervolting-on-laptops-with-Intel-Raptor-Lake-CPUs.708045.0.html
• [L32] https://github.com/georgewhewell/undervolt
• [L33] https://code-dev.fixnum.org/2024-11-16-intel-cpu-undervolt-unlock/
• [L34] https://github.com/wesmar/UnderVolter
• [L35] https://man.archlinux.org/man/extra/thermald/thermal-conf.xml.5.en
• [L36] https://github.com/intel/thermal_daemon
• [L37] https://bugs.launchpad.net/ubuntu/+source/thermald/+bug/1940485
• [L38] https://linrunner.de/tlp/settings/processor.html
• [L39] https://wiki.archlinux.org/title/CPU_frequency_scaling
• [L40] https://github.com/AdnanHodzic/auto-cpufreq/issues/9
• [L41] https://manpages.debian.org/testing/stress-ng/stress-ng.1.en.html
• [L42] https://wiki.ubuntu.com/Kernel/Reference/stress-ng
• [L43] https://www.pcsuggest.com/gpu-benchmarking-and-stress-testing-in-linux/
• [L44] https://ubuntuhandbook.org/index.php/2024/11/benchmark-stress-test-gpu/
• [L45] https://linuxvox.com/blog/gpu-stress-test-linux/
• [L46] https://docs.kernel.org/admin-guide/thermal/intel_thermal_throttle.html

##Software-only path and validation

• [S1] Notebookcheck — MSI Crosshair 15 R6E review (i7-12700H + RTX 3070): https://www.notebookcheck.net/MSI-Crosshair-15-R6E-in-review-Core-i7-12700H-and-RTX-3070-combo-augurs-well-for-QHD-gaming.675959.0.html
• [S2] Notebookcheck (DE) — same review: https://www.notebookcheck.com/MSI-Crosshair-15-R6E-im-Test-Vielversprechender-QHD-Gamer-mit-Core-i7-12700H-und-RTX-3070.676534.0.html
• [S3] Notebookcheck — "Intel and OEMs have killed undervolting...": https://www.notebookcheck.net/Intel-and-OEMs-have-killed-undervolting-and-there-is-little-that-you-can-do-about-it.477330.0.html
• [S4] Notebookcheck — throttling-in-reviews opinion piece: https://www.notebookcheck.net/Opinion-It-s-time-we-talked-about-throttling-in-reviews.234232.0.html
• [S5] Notebookcheck — Core Ultra 7 155H throttling behavior (methodology context): https://www.notebookcheck.net/Core-Ultra-7-155H-exhibiting-throttling-behavior-by-up-to-25-percent-on-smaller-laptop-models.977246.0.html
• [S6] Notebookcheck — ThrottleStop guide (PROCHOT/throttling explainer): https://www.notebookcheck.net/How-to-Lower-Temperatures-Stop-Throttling-and-Increase-Battery-Life-The-ThrottleStop-Guide-2017.213140.0.html
• [S7] TechSpot — Intel Core i9-12900HK review (75W parity with 12700H): https://www.techspot.com/review/2425-intel-core-i9-12900hk/
• [S8] GitHub — erpalma/throttled (Linux EC power-limit reset daemon): https://github.com/erpalma/throttled
• [S9] GitHub — horshack-dpreview/setPL (Linux PL1/PL2 setter): https://github.com/horshack-dpreview/setPL
• [S10] AnandTech Forums — setPL utility thread: https://forums.anandtech.com/threads/fyi-my-utility-for-setting-the-pl1-pl2-power-levels-under-linux-setpl.2606729/
• [S11] kernel-internals.org — Power Capping and RAPL: https://kernel-internals.org/power/power-capping/
• [S12] Jarrod's Tech — Laptop CPU Performance in Cinebench 2024 (methodology): https://jarrods.tech/laptop-cpu-performance-in-cinebench-2024/
• [S13] Jarrod's Tech — site/methodology overview: https://jarrods.tech/
• [S14] Tom's Hardware Forum — "How to stop i7 12700H from Power limit throttling?": https://forums.tomshardware.com/threads/how-to-stop-i7-12700h-from-power-limit-throttling.3769453/
• [S15] Tom's Hardware Forum — "Uneven cpu core temps after repasting": https://forums.tomshardware.com/threads/uneven-cpu-core-temps-after-repasting.3752580/
• [S16] Intel Community — "Temperature difference between cores": https://community.intel.com/t5/Mobile-and-Desktop-Processors/Temperature-difference-between-cores/td-p/202375
• [S17] AMD Community — "Are uneven CPU core temps due to bad thermal paste application?": https://community.amd.com/t5/pc-processors/are-uneven-cpu-core-temps-due-to-bad-thermal-paste-application/m-p/595043
• [S18] ROG Forum — BD PROCHOT Red/Yellow in ThrottleStop: https://rog-forum.asus.com/t5/overclocking-tweaking/bd-prochot-red-yellow-in-throttlestop-g731gw-solved/td-p/866355
• [S19] Acer Community — BD PROCHOT causing throttling on Helios 300 i7-8750H: https://community.acer.com/en/discussion/544991/bd-prochot-causes-cpu-throttling-to-803-mhz-helios-300-i7-8750h
• [S20] Tom's Hardware Forum — "What does the PROCHOT 95°C do?": https://forums.tomshardware.com/threads/what-does-the-prochot-95%C2%B0c-do.3301894/
• [S21] Ultrabookreview — The ThrottleStop Guide (2026): https://www.ultrabookreview.com/31385-the-throttlestop-guide/
• [S22] KitGuru — MSI Crosshair 15 R6E review (ports/specs, no Thunderbolt): https://www.kitguru.net/lifestyle/mobile/laptops/luke-hill/msi-crosshair-15-r6e-laptop-review-core-i7-12700h-rtx-3070/
• [S23] Best Buy Q&A — Crosshair 15 R6E port configuration: https://www.bestbuy.com/site/questions/msi-crosshair-15-rainbow-six-extraction-edition-b12u-15-6-gaming-laptop-intel-core-i7-16-gb-memory-nvidia-geforce-rtx-3070-multicolor-gradient/6501181
• [S24] Best Buy Q&A — Crosshair 16 Thunderbolt 4 confirmation (contrast case): https://www.bestbuy.com/site/questions/msi-crosshair-16-144hz-gaming-laptop-intel-13th-gen-core-i7-with-16gb-memory-nvidia-geforce-rtx-4070-1tb-ssd-black/6537000/question/e63c4191-9235-335c-a0ca-30b795a4039d
• [S25] PC Gamer — "Thunderbolt vs OCuLink external GPU interface-off": https://www.pcgamer.com/hardware/graphics-cards/state-of-play-egpus/
• [S26] egpu.io forums — "Buying a Laptop for eGPU?": https://egpu.io/forums/which-gear-should-i-buy/buying-a-laptop-for-egpu/
• [S27] HomeGuide — Geek Squad Prices 2026 (general labor rates, no repaste line item): https://homeguide.com/costs/geek-squad-prices
• [S28] phonerepairmore.com — Geek Squad Prices 2026: https://phonerepairmore.com/best-buy-geek-squad-repair-guide/
• [S29] Overclock.net — "Best way I've found to massively lower in-game CPU temps without impacting FPS": https://www.overclock.net/threads/best-way-ive-found-to-massively-lower-in-game-cpu-temps-without-impacting-fps.1800234/
• [S30] LinusTechTips forum — "GPU hotspot temp in excess of 100 degrees - 3070 ti": https://linustechtips.com/topic/1432702-gpu-hotspot-temp-in-excess-of-100-degrees-3070-ti/
• [S31] Tom's Hardware Forum — Gigabyte RTX 3070 hotspot temperature (delta heuristics): https://forums.tomshardware.com/threads/gigabyte-rtx-3070-hotspot-temperature.3756380/
• [S32] mundobytes.com — MSI Center performance/power profile explainer (mode-naming, no wattage table): https://mundobytes.com/en/MSI-Center-and-Windows/
• [S33] MSI — Crosshair 15 12U product page (spec reference): https://www.msi.com/Laptop/Crosshair-15-B12UX