Summary of ""Forbidden" Disassembly: NVIDIA Laptop RTX "5090" with Water Cooling"

Product reviewed

XMG 5090 laptop (featuring an “RTX 5090” laptop GPU) with a gimmick/option for external water cooling. The system uses two dedicated water-cooling ports and an external ~120mm radiator setup (120mm radiator + 120mm fan).

Key features mentioned

External water-cooling accessory ports on the laptop (two ports).
External cooling unit described as:
- ~120mm radiator + 120mm fan
- Tubes and a reservoir/radiator assembly, described as a “120 CLC”-style loop
Thermal design (laptop-only cooling):
- Multiple blower fans (two main bottom blowers; plus mention of a third blower fan)
- Large heat pipe(s) including one flat heat pipe across the center
- VRM cooling: copper cold plate / finning arrangement, with thermal pads/phase-change material and a noted flat plate contacting power components
Hardware observations relevant to performance claims:
- GPU power draw measured around 175W (for this laptop GPU)
- GPU silicon/die size suggested to be closer to 5080-class than true 5090-class based on measurements
- Memory configuration: 24GB GPU mentioned; later discussion suggests 8 memory modules total on the GPU side

Main findings / opinions (pros and cons)

Pros (what worked / looked good)

Thermal layout has lots of airflow paths: large ventilation area with relatively open fin-stack access (less obstructed than some laptops).
VRM cooling contact seems deliberate:
- VRM components get contact via thermal pads/phase-change material
- A “flat plate” cold-plate approach is described as likely adequate for concentrated VRM regions
Heatpipe appears to target the GPU directly, including a flat heat pipe across the center
In principle, external water cooling could help because it adds more radiator surface area than typical laptop cooling—if plumbed effectively

Cons (problems / negative points)

External water cooling did not work during testing:
- When connected by another team, it leaked
- The loop/pump behavior: it runs but turns off immediately and was not fully troubleshot; described as defective by XMG
- Because it failed, the reviewer could not test the thermal benefit
Leak/connection design issue:
- Tubes appear secured with magnets on the radiator side, which the reviewer dislikes
- Connection is described as not properly seated in the way it “needs to sit,” causing the leak
- Reviewer suggests radiator-side retention should use clamps/clip/QDC-style mechanisms rather than magnets
RTX “5090” naming skepticism:
- Reviewer is not impressed with 5090-vs-4090 laptop comparisons so far (expects more based on desktop context)
- CUDA core count estimate: near 5080 desktop class, not true 5090
- Die size measurement mismatch:
  - Reviewer estimates ~378 mm² actual (method-based), with external measurement estimates around ~403–406 mm²
  - Claimed “true 5090” die ~750 mm²
  - “5080” desktop die closer to ~403–406 mm²
- Implied conclusion: the laptop “5090” may be renamed/up-binned 5080-class silicon, not a full new 5090 die
Flat heat pipe uncertainty:
- Flat heat pipes are described as situational; performance depends heavily on fit/coverage
Disassembly nuisances:
- Lots of pre-installed “hair”/debris (repeatedly mentioned)
- Some parts appear heavily adhered (thermal putty suction/stuck components), increasing effort

User experience / practical testing experience

Reviewer had only ~6.5 hours (overnight shift) and ran thermals/audio plus one game test, so results are partial
Water cooling testing was limited by leak and pump shutoff, reducing usable real-world validation

Comparisons made

Laptop RTX 5090 vs Laptop RTX 4090
- Reviewer: “not impressed thus far” (details promised from another tester)
Laptop RTX 5090 vs desktop RTX 5090 expectations
- GPU power draw observed ~175W vs desktop being stated as far higher (claimed as ~400W lower than desktop actual 5090)
- Biggest comparison: die size and CUDA/core estimates implying 5080-class, not true 5090-class
On the cooling design concept
- External water cooling compared indirectly to typical desktop cooling:
  - A 120mm CLC would generally be insufficient for desktop usage, but may matter more on a laptop due to scaling of surface-area needs
  - However, this was not verifiable because the external loop did not run reliably

Numerical details / measurements explicitly mentioned

Test time: ~6.5 hours
GPU power (observed): ~175W
Memory: 24GB (also described as 8 modules on the GPU side)
Heat pipe size: ~10–12 mm, later clarified as ~13.6 mm
Die measurement (reviewer estimates):
- “Actual 58 is 378 mm²”
- External rough estimate: ~406 mm² (multiple approximations)
- Desktop references:
  - True 5090 die ~750 mm²
  - 5080 G desktop die closer to ~403 mm²
Battery: 6450 mAh typical, 15.48V, 99.9Wh
External radiator: 120mm class, described repeatedly as ~120 rad / 120mm fan

Overall verdict / recommendation (based on the subtitles)

Verdict: Interesting teardown and cooling layout, but real-world performance/value is questionable—especially because the “5090” appears much closer to 5080-class silicon based on die/CUDA/core reasoning, and because the external water-cooling feature failed (leaks + pump shutting off).
Recommendation: Don’t rely on the external water-cooling accessory until it’s proven reliable. Treat the laptop “RTX 5090” as potentially marketing-forward rather than true 5090-class at least per the reviewer’s silicon/power observations.

Unique points mentioned (consolidated)

“Forbidden/private disassembly” warning (general electrical safety note)
XMG laptop teardown connected through collaboration (not directly tied to Nvidia’s launch cycle)
External water cooling: two ports + external ~120mm radiator/fan
External water cooling failure: leaked, pump turned off immediately, marked defective
Laptop cooling airflow: blocked off areas, but relatively open fin-stack ventilation; chassis ventilation height used
Cooling fans: two blower fans plus a third blower fan discovered
Flat heat pipe across the center; performance uncertainty
VRM cooling: copper cold plate/VRM plate with thermal pads/phase-change pad; emphasis on thermal contact
Thermal interface note: thermal putty described as good gap filler but difficult to remove
GPU silicon skepticism: CUDA estimate ~5080-class; die estimates closer to 5080 than 5090
Memory configuration: 24GB, with multiple modules (noted as 8 on the GPU side at least)
Motherboard split/bridge design: directionality/junction likely for high-speed connections/manufacturing/flex benefits
Repeated teardown debris mention (“hair”)
Battery specs given (99.9Wh class)
External water-cooling critique: magnet-based connection seen as risky; should use clip/clamp/QDC retention
Power-cable routing above tubes noted as helping avoid power-cable leakage

Speaker views (who said what, separated)

Primary narrator / teardown voice (main contributor)
- Cooling layout observations
- Die-size and CUDA-core skepticism
- External water-cooling failure critique
- Overall impressions (including “not impressed” vs 4090 yet)
Referenced/other creator (“Josh from Just Josh”)
- Promised more detailed performance comparisons and testing methodology support (specific results not included in these subtitles)
XMG (company) / “XMG told Josh’s team”
- Claimed explanation: external water-cooling pump/system behavior described as defective (per their contact)