Thermal Paste for GPUs: Repasting Your Graphics Card the Right Way

If you’ve ever watched your graphics card’s fans ramp up to jet-engine levels during a gaming session or a GPU-compute workload — even though the card isn’t particularly old — the culprit is often something surprisingly simple: the small layer of thermally conductive paste (the gooey compound sandwiched between the GPU chip and its metal cooler) has dried out, cracked, or pumped out from between those surfaces over time. That paste is the critical bridge transferring heat from the silicon die into the heatsink. When it degrades, temperatures spike, fans scream, and your expensive GPU either throttles its performance or runs stressed. The good news: repasting a GPU — replacing that factory compound with a fresh, high-quality product — is one of the highest-return maintenance tasks in PC building. This guide walks you through how to know when you need it, what products are worth considering, how to actually do it, and what to expect afterward.

Across aggregated long-run reviews, GPU repastes consistently deliver more dramatic temperature drops than CPU repastes — often 10–20°C on the hotspot reading — because reference and partner-card factories frequently apply paste under more mechanical stress than a careful hobbyist would, and because GPUs sustain higher, longer heat cycles that accelerate compound degradation. If you’re sitting on a card that’s running hotter than it did a year or two ago, keep reading.

How to Tell Whether Your GPU Actually Needs a Repaste

The diagnostic step most builders skip is distinguishing normal thermal degradation from thermal pump-out — these look similar in sensor data but have different causes and severities.

Normal degradation is gradual. Over one to three years of sustained use, organic carrier oils in silicone-based pastes slowly evaporate, leaving a drier, less conductive compound. GPU-load temperatures climb 5–10°C above their first-month baseline. HWiNFO64 (the free sensor-monitoring tool virtually every overclocker uses) is your best friend here: log your GPU core temperature, hotspot temperature, and junction temperature (on AMD cards) at full load and compare them against what you recorded — or what reviewers documented — at launch.

Thermal pump-out is more abrupt and more severe. It happens when repeated heat cycles — the GPU expanding when hot, contracting when cool — physically squeeze the paste out from between the die and heatsink contact plate. TechPowerUp’s thermal paste longevity study documents this failure mode in detail: the paste migrates toward the edges of the die, leaving a dry hotspot at the center. One reviewer community document describing a Ryzen 9 5900X after just 12 months of use with KPX compound found a clearly visible dried hotspot on the die surface when disassembled — the most specific pump-out failure description in recent hobbyist literature. On a GPU, you’d see this as a sudden, steep temperature increase rather than a slow creep, and the hotspot delta (the difference between GPU core temperature and hotspot temperature in HWiNFO64) widening noticeably.

Rule of thumb: if your GPU hotspot is reading above 95°C under sustained load and it wasn’t doing that 12–18 months ago, a repaste is a strong first move before you consider any hardware change.

Choosing Your Thermal Compound: The Real Tradeoffs

Not all thermal pastes are created equal, and the GPU application context introduces considerations that don’t exist on CPUs — larger die surfaces, non-square die geometries on some cards, and the presence of VRAM chips and VRM components that may also need attention.

Traditional Paste: Kryonaut and MX-Series

Thermal Grizzly Kryonaut remains a benchmark reference compound in long-run GPU reviews. Its viscosity is on the thicker side, which reduces pump-out risk compared to ultra-thin pastes like KPX — a meaningful advantage on GPU dies that see sustained high-temperature cycles. Reviewers at Tom’s Hardware and Igor’s Lab consistently note that Kryonaut’s included spatula and nib applicator tips make it genuinely easier to apply evenly across the larger surface area of a GPU die compared to pastes that ship with nothing but a syringe. If you’re doing a full repaste that includes VRM heatsink pads or any secondary contact points, budget accordingly — the 1g tube is often not quite enough for an exhaustive job (see the FAQ section below).

Thermal Grizzly KPX delivers better raw conductivity numbers on paper, but its lower viscosity makes it a pump-out risk over long GPU duty cycles. TechPowerUp’s analysis flags this tradeoff directly. It’s a reasonable choice if you plan to repaste on a regular maintenance schedule (annually), but if you want a longer service interval, Kryonaut or a graphene pad is the more defensible choice.

Arctic MX-6 has attracted a loyal following for GPU work partly because of its included cleaning wipes — a genuinely useful addition that makes old paste removal cleaner without sourcing isopropyl alcohol wipes separately. The MX series is non-electrically-conductive (safe around VRAM pads and PCB traces if you apply carefully), and long-run reviewers appreciate its stability over heat cycles. One consistent note in the review aggregate: the lemon fragrance added to MX-6 has prompted questions about whether it leaves any residue on contact surfaces. Based on Arctic’s published formulation notes and the absence of corrosion reports in multi-year use documentation, this appears to be cosmetic rather than functional — but it’s worth a clean IPA wipe-down of the die surface before applying if you’re cautious.

Graphene Phase-Change Pads: KryoSheet

The Thermal Grizzly KryoSheet graphene pad represents a different category altogether. Rather than a paste you spread, it’s a thin, semi-rigid sheet that you cut to size, peel, and press onto the die surface — it then transitions to a conforming state under heat and pressure. Igor’s Lab’s dedicated KryoSheet analysis shows it performing competitively with top-tier pastes across a range of die geometries.

The GPU-specific results are where it gets interesting. An RTX 5090 owner citing HWiNFO64 hotspot readings documented a drop from over 100°C to 81°C after switching to KryoSheet — a 19°C hotspot reduction on one of the hottest consumer GPUs currently available. Separately, an RX 7900 XTX owner using KryoSheet reported that it completely eliminated the reference card’s notorious junction temperature issue — a problem HardwareLuxx has documented extensively — and silenced fans that had previously been pegged at 98–100% speed. Those are specific, hardware-identified results, not generalized claims.

The tradeoff: KryoSheet costs more per application than paste, cutting it to fit non-square dies requires care and a sharp blade, and it’s less forgiving of repositioning errors than paste. We’ll address the non-square die question in the FAQ.

By the Numbers

Compound	Conductivity (W/m·K)	Pump-Out Risk	GPU Die Coverage
Kryonaut	~12.5	Low-Medium	Paste — spread manually
KPX	~14.2	Medium-High	Paste — thin, flows easily
MX-6	~12.6	Low	Paste — non-conductive
KryoSheet	~18 (graphene)	None	Pad — cut to shape

Conductivity values per manufacturer spec sheets; pump-out risk characterization per TechPowerUp longevity documentation.

The Repaste Process: What You’re Actually Doing

Detailed disassembly steps vary too much per card model to cover universally here — always locate a card-specific teardown guide before you start. The core process is consistent, though:

1. Discharge and ground yourself. Use an anti-static wrist strap. GPUs are expensive.
2. Remove the cooler. Heatsink screws on most reference and partner cards follow a spring-loaded pattern — loosen in a star sequence, don’t remove one fully before touching the others.
3. Document before you remove anything. Photo the thermal pad thickness and placement on VRAM chips and VRMs before you touch them. Replacing these pads with wrong-thickness replacements causes new problems.
4. Remove old paste. 90%+ isopropyl alcohol on a lint-free cloth or dedicated cleaning wipes. Multiple passes. The die surface should be visibly clean before you apply anything new.
5. Apply new compound. A pea or rice-grain-sized dot at the center of the die; the heatsink mounting pressure spreads it. For a graphene pad, cut to the die’s dimensions and press flat.
6. Reassemble, boot, monitor. Run HWiNFO64 at full GPU load (a 20-minute Unigine Superposition pass works well) and log your new baseline.

Frequently Asked Questions

How do I know if my GPU has thermal pump-out vs. normal degradation? The clearest indicator is rate of change. Normal degradation is a slow, months-long temperature creep. Pump-out tends to appear more suddenly — a 10–15°C spike over a short period — and the hotspot delta in HWiNFO64 widens disproportionately. If you disassemble and find paste pooled at the die edges with a dry center, that’s textbook pump-out as documented in TechPowerUp’s longevity research.

Does the KryoSheet graphene pad work on GPUs with non-square dies? Yes, but it requires careful trimming. KryoSheet ships in sheet form specifically so you can cut it to any die geometry. Use a metal ruler and a fresh X-Acto blade, cut slightly inside your measured die boundary (you don’t want overhang touching surrounding components), and take your time. Igor’s Lab’s application notes cover this workflow in detail. It’s manageable for any patient builder; it’s just less plug-and-play than paste.

Is Kryonaut 1g enough for a full GPU repaste including VRMs? Usually not, if “full repaste” means touching every thermal interface. The GPU die application itself uses well under 1g. But if you’re replacing thermal pads on VRAM or VRMs with paste rather than pads — which isn’t always the right move anyway — consumption adds up. For die-only application: 1g is sufficient. For a comprehensive contact-surface refresh: grab a 3g or 7g tube, or buy a separate thermal pad kit for the VRAM/VRM surfaces rather than using paste there.

Can I reuse the factory VRAM thermal pads or do I need to replace them too? Factory VRAM and VRM thermal pads can often be reused if they haven’t dried, cracked, or torn during removal. Gently peel them; if they come off cleanly and are still pliable, they’re reusable. If they crumble, harden, or leave chunks on the VRAM package, replace them — measure thickness with digital calipers before ordering replacements, because wrong-thickness pads create either a gap (no contact) or excess pressure. KitGuru’s RTX 5090 thermal analysis covers pad thickness measurement as a critical step in the repaste workflow.

How long does MX-4 or MX-6 last on a GPU running at high sustained temps? Arctic rates MX-4 and MX-6 for eight-year service life under typical conditions. Under sustained high-heat GPU workloads — think continuous rendering or AI-inference at near-TDP — real-world service intervals in the review aggregate suggest three to five years before meaningful performance degradation, with MX-6 showing modestly better thermal stability than the older MX-4 formulation. Annual monitoring with HWiNFO64 baselines is the practical hedge regardless of rated life.

Will repasting void my GPU warranty? Almost certainly yes, if your card is still under the manufacturer’s warranty period. Disassembly requires removing warranty stickers on most NVIDIA and AMD partner cards, which voids coverage. Weigh this against your card’s age: if you’re three or more years past purchase, warranty coverage is likely expired anyway and repasting is a straightforward maintenance call. If your card is under 12 months old and running hot, contact the manufacturer first — you may be entitled to an RMA rather than a DIY fix.

---

The decision frame here is relatively clean: if your GPU is out of warranty, running 10°C or more hotter than its launch-era benchmarks, and you’re comfortable with a disassembly process, a repaste is one of the most cost-effective interventions available. For cards with known junction temperature issues — the 7900 XTX being the textbook case — the KryoSheet graphene pad has produced some of the most dramatic documented results in the current review cycle. For straightforward paste replacement, Kryonaut’s application ergonomics and pump-out resistance make it the defensible default. And if you want lower friction in the cleaning step, MX-6’s included wipes earn their keep. Whichever product you choose, the HWiNFO64 before-and-after comparison is non-negotiable: log your baseline, repaste, log again, and let the data tell you what you actually gained.