Graphics Cards & GPUs

Graphics cards (GPUs) are specialised processors designed to render 3D graphics, accelerate video encoding, and increasingly to run AI workloads. The two dominant brands are NVIDIA and AMD, with Intel as a growing third option. Modern consumer GPUs range from $200 entry-level cards capable of 1080p gaming to $2,000+ flagship cards designed for 4K gaming, professional 3D rendering, and large language model inference. Beyond gaming, GPUs are now critical for video editing (export speeds depend heavily on GPU encoding), 3D modelling, and any AI workload running locally.

A graphics card is essentially a computer-within-a-computer — its own processor (the GPU), its own memory (VRAM), its own power delivery, and its own cooling solution mounted on a PCIe expansion card that plugs into a desktop motherboard. Power consumption ranges from 75 watts (small cards that draw all power from the slot) to 450+ watts (high-end cards that need multiple dedicated power connectors). This power and thermal output is why graphics cards are physically the largest single component in most modern desktops.

Graphics cards are surprisingly reliable for the heat and power they handle. The most common failure is fan failure — after 3–5 years, fan bearings wear out and the card either runs noisier than new or stops cooling adequately, leading to thermal shutdowns. Replacing a fan or fan assembly is straightforward and costs $20–$60. Less common but more serious failures include VRAM module faults (showing as artifacts, crashes, or specific games failing while others work), VRM (voltage regulator) failures, and physical damage from improper handling.

GPU silicon itself almost never wears out under normal use. The biggest threat is heat over time — a card that consistently runs above 80°C ages much faster than one kept below 70°C. Dust accumulation on heatsinks is the silent killer, restricting airflow until temperatures climb past safe limits. Annual cleaning with compressed air keeps a quality graphics card running at peak performance for many years.

Graphics cards age in capability, not in functionality. A 5-year-old card still runs the games and software it ran on day one, but new games and software demand more performance every year. The replacement decision usually comes from wanting higher resolution, higher frame rate, ray tracing, or AI features rather than from the existing card actually failing.

When buying new, the key specifications are VRAM (12 GB minimum for modern AAA games at 1440p, 16+ GB for 4K and creative work), generation (newer architectures are dramatically more efficient), power draw (verify your power supply can handle it with margin), physical dimensions (some flagship cards don't fit in standard cases), and thermal design (better cooling means quieter operation and longer lifespan). Avoid the temptation to buy the absolute top tier — the price-to-performance curve is steep, and one tier down often delivers 90% of the performance for 60% of the cost.

Graphics cards run hotter and harder than any other component in a typical PC, and dust is their single biggest long-term enemy. Open the case once every three months and gently blow out the GPU's fan blades and heatsink fins with short bursts of compressed air, holding the fans still with a finger so they don't spin up and damage their bearings. A clean GPU runs ten to fifteen degrees cooler than a dust-clogged one, which translates directly into quieter operation, higher sustained clock speeds, and longer life from the silicon, the solder joints, the VRMs, and the fans themselves. The whole maintenance routine takes five minutes and is the single most impactful habit for any gaming or workstation PC.

Avoid extreme overclocking on a card you intend to keep for many years. Modest overclocks, well within the manufacturer's recommended voltage limits, are essentially free performance and don't shorten card life meaningfully. Aggressive overclocks with raised voltages accelerate electromigration in the silicon and dry out thermal pads and paste much faster. If a card already runs at its limits from the factory, undervolting can reduce temperatures and noise dramatically with minimal performance loss — a thirty-minute experiment with the manufacturer's tuning utility often pays dividends in quieter operation and longer life. Repaste the GPU every three to four years to restore the thermal performance that always degrades over time.

Modern GPUs hold their value remarkably well on the secondhand market, and a five-year-old card is still genuinely useful for many games and creative applications at lower settings or older titles. When upgrading, sell the old card on an enthusiast marketplace rather than letting it sit in a drawer — funded partial-upgrade cycles like this make staying current much more affordable. If a card has truly failed and isn't worth repairing, recycle it through the manufacturer's take-back programme or certified e-waste channels because GPUs contain valuable copper, aluminium, and rare-earth materials that should never end up in landfills.