← Back to Articles Guide

RTX 4090 Overclocking: Core, Memory and Power Limit

The RTX 4090 ships with thermal headroom that most users never touch. A systematic approach to core offset, GDDR6X memory overclocking, and power limit adjustment can extract 8 to 12 percent additional raster performance while keeping thermals inside acceptable margins.

Published June 2026 — VoltGround

The Ada Lovelace architecture that powers the RTX 4090 behaves differently under overclocking than Ampere did. The card boosts aggressively by default, and the gains from a simple core offset are smaller than they were on previous generations. That does not mean overclocking is pointless—it means you need to approach it systematically rather than just slamming the power limit to maximum and calling it done.

This guide assumes you are running a reference-class or AIB RTX 4090 with a three-fan or three-slot cooler. The process works on all board partners. The differences between AIB versions mostly affect the temperature ceiling you can sustain, not the overclocking headroom itself.

Tools you need before you start

You need MSI Afterburner (version 4.6.5 or later), HWiNFO64 for sensor logging, and a benchmark you trust. For raster workloads, 3DMark TimeSpy Extreme and Port Royal are both reliable. For games, you want something with a built-in benchmark that produces consistent results across runs—Cyberpunk 2077 in Overdrive mode with RT enabled and the built-in benchmark works well because the scene is GPU-limited and reproducible.

Before touching any settings, run your benchmark three times and record the average. Note your GPU core temperatures and GPU hotspot temperatures at peak load. Also note your GPU clock in the benchmark—check HWiNFO64 GPU Core Clock Maximum value. This baseline is your reference point. Without it, you cannot know what you actually gained.

Power limit first

Raise the power limit to maximum before doing anything else. On stock RTX 4090 cards, the power limit ceiling in Afterburner is typically 600W (up from the 450W default). Raising it tells the GPU it can draw more power to sustain boost clocks, which removes the primary bottleneck on stock hardware. Run your benchmark again after this change alone. On most cards this produces a 4 to 6 percent improvement in raster performance with no other modifications.

The cost is increased power draw and heat. If your cooling solution or PSU is marginal, address that before continuing. The RTX 4090 at full power limit and core OC should be treated as a sustained 600W+ component.

Core offset tuning

The Ada architecture responds to core offset tuning, but the gains are nonlinear. The GPU runs a performance-per-watt algorithm that makes conservative boost decisions unless you guide it. Start with a +100 MHz core offset in Afterburner. Run the benchmark. If it is stable and scores improve, move to +125 MHz. Continue in 25 MHz steps until you hit instability—typically black screens, driver crashes, or artifact patterns in the benchmark results.

Back off 25 MHz from your instability point and run a 20-minute stress test using OCCT GPU:3D Adaptive. If that passes, your core offset is stable. Most RTX 4090 cards land in the +100 to +150 MHz range. Outliers on good silicon can reach +200 MHz but they are genuinely rare.

Note: Core offset in MSI Afterburner applies as a fixed addition to the GPU's boost table. The actual in-game clock still varies based on power, temperature, and driver decisions. The offset shifts the entire curve upward, which is why results at the same offset vary between cards.

Memory overclocking on GDDR6X

GDDR6X memory responds well to overclocking but can produce instability that manifests as visual artifacts rather than outright crashes. Start at +500 MHz memory offset and run Port Royal, which is memory-bandwidth sensitive. Increment in 250 MHz steps. Watch for pixel shimmering, bright single-pixel dots, or corrupted geometry in the benchmark scene. These are memory errors, not GPU errors.

Most RTX 4090 cards can run +1000 to +1500 MHz memory offset stably. Higher settings exist but require temperature-controlled testing because GDDR6X error rates increase with memory temperature. At sustained load the memory junction temperature on the 4090 can reach 100 to 110 degrees Celsius, which is within specification but means your error margin narrows.

Memory overclocking helps most in bandwidth-limited workloads: high-resolution gaming, 4K RT, and compute tasks. In CPU-limited scenarios it is invisible in the benchmark results.

Thermal verification after overclocking

Run a full 30-minute gaming session or OCCT loop and log the following values in HWiNFO64: GPU Core Temperature, GPU Hotspot Temperature, and GPU Memory Junction Temperature. On a properly cooled three-fan AIB card with the settings above, you should expect GPU Core around 70 to 78 degrees Celsius, Hotspot around 85 to 92 degrees Celsius, and Memory Junction around 100 to 108 degrees Celsius.

If Hotspot is consistently above 95 degrees Celsius, your cooling is being stressed. Either add case airflow or consider whether your power limit is sustainable in your chassis. Hotspot temperature directly affects boost clock sustainability, so if it is high, your OC gains will partially disappear in extended runs.

Putting it together: what to expect

A well-tuned RTX 4090 overclock—maximum power limit, +125 to +150 MHz core offset, +1000 MHz memory—typically produces 8 to 12 percent improvement in 3DMark TimeSpy Extreme over stock settings. In games the gains are usually 5 to 9 percent at 4K, which is meaningful given what the card costs. At 1440p the gains are smaller because you are more likely to be CPU-limited at that resolution.

The investment is a few hours of testing and a solid understanding of what each variable does. The result is a card that performs close to what factory-overclocked versions cost several hundred dollars more. It is one of the better returns on time you can find in PC hardware tuning.