What arrived on the bench
Short answer: A postgraduate student from Hyderabad brought in an ASUS ROG Zephyrus gaming laptop with an NVIDIA RTX 4090 mobile GPU. The machine was being used for training a computer vision model and would shut down completely within 20–40 minutes of starting a training run. The student had already tried undervolting the GPU in software — it helped marginally but did not stop the shutdowns. The root cause turned out to be hardened thermal paste and a partially blocked exhaust duct, a combination that had pushed GPU temperatures to 97°C under sustained load in a room without air conditioning.
How we diagnosed and fixed the thermal trip
Step 1: Reproducing the fault on the bench
The first step was to run a controlled GPU stress test (a tool that drives the GPU to 100% load to check stability) while monitoring temperatures with HWiNFO64 (a hardware monitoring utility). Within 18 minutes, the GPU junction temperature hit 97°C and the laptop triggered a hard power-off. The CPU was running at a more reasonable 82°C, which told us the CPU heatsink was still functional — the problem was isolated to the GPU thermal stack.
We also noted that the fan was spinning at maximum speed from the first minute of the test. A laptop fan running at full speed from a cold start is almost always a sign the thermal paste has degraded — the system compensates for poor heat transfer by spinning the fans harder, but at some point there is not enough airflow to compensate for degraded compound.
Step 2: Opening the machine and the finding
The thermal paste on the GPU die — the small chip at the centre of the graphics processor — had turned from its original pliable consistency to a hard, cracked layer. Thermal paste works by filling microscopic gaps between the chip surface and the metal heatsink. When it dries out and cracks, air pockets form in those gaps, and air conducts heat fifty times worse than good thermal compound. We also found a layer of fine dust compressed into the fin stack of the GPU heatsink, reducing airflow by an estimated 30%.
Both findings together explained the symptoms perfectly. The student had been running training jobs for roughly 14 months on the same machine in an environment that was often 35–38°C ambient. Indian summer ambient temperatures reduce the thermal headroom on a high-end gaming laptop by 8–12°C compared to the same machine in a 24°C air-conditioned room — a difference that matters a great deal when the GPU is already running near its thermal limit.
Step 3: The fix
We cleaned the old compound, applied a fresh layer of liquid metal thermal compound (a high-performance compound that has significantly lower thermal resistance than standard paste, particularly suited to the heat output of top-tier mobile GPUs), replaced the thermal pads on the VRAM chips (the memory chips on the GPU board) and the VRM (voltage regulation module), and deep-cleaned the heatsink fin stack. After reassembly, the same stress test now peaked at 78°C — a 19-degree reduction — and the machine ran stably through a two-hour training run without incident. You can read more about laptop overheating causes on our overheating repair page.
Step 4: The India angle for AI/ML users
The AI and machine learning community in India has grown enormously, with students and professionals in Hyderabad, Bengaluru, Pune, and Chennai running training workloads on consumer gaming laptops. These machines were designed for gaming sessions of a few hours — not for 6–12 hour uninterrupted training runs at 100% GPU load.
India’s power infrastructure adds another variable: voltage fluctuations during power-cut recovery events can spike current to the GPU, accelerating component stress. A basic surge protector and a UPS for clean power delivery are worth considering for any workstation used for sustained compute. We see power-surge-related bench cases weekly.
When to bring it in — and what it costs
Signs your gaming or AI laptop needs thermal service
Watch for: fan noise starting immediately from a cold boot, GPU temperatures above 90°C within the first 10 minutes of a heavy workload, training runs that crash or produce CUDA out-of-memory errors (which can occur when GPU memory throttles under thermal stress), or any unexpected shutdown during compute. Do not ignore these — the next stage after thermal throttling is GPU damage.
Typical cost in India
Thermal paste replacement with fan cleaning on a gaming laptop: ₹800–₹1,500. With liquid metal compound and full thermal pad replacement: ₹1,500–₹3,000. Our Hyderabad doorstep repair service can handle most gaming laptops at your location; the ₹149 visit charge covers the diagnosis.
A note from the LRW Engineer Team
The single biggest preventable cause of gaming and AI laptop failure in India is degraded thermal paste combined with dust accumulation. A ₹1,500 annual service prevents a ₹30,000–₹60,000 GPU board replacement. If you are running sustained AI workloads on a consumer laptop, treat thermal maintenance as routine, not optional. Also see our Indian summer overheating case studies for more bench findings.