GPU Stress Test: How to Test Your Graphics Card Stability (2026) | GamerHardware

A GPU stress test pushes your graphics card to 100% load to check for stability issues, overheating, and visual artifacts. FurMark is the fastest free option for a quick thermal check, while 3DMark Time Spy Stress Test gives you the most reliable pass/fail result for overclocking validation. This guide covers the best stress test tools, safe temperature thresholds for every major GPU generation, how to read monitoring data in HWiNFO and MSI Afterburner, and what to do when your card fails. If you want the full context, see all our hardware tests and tools.

Why You Should Stress Test Your GPU

Gaming alone is not a reliable stability test. Most games load your GPU to 70-90% and cycle between different workloads constantly. A game crash could mean a buggy driver, a bad game patch, or an unstable GPU – you can’t tell which. A proper stress test isolates your graphics card by pinning it at maximum load for a sustained period, so any failure points straight at the hardware or its settings.

There are four situations where a GPU stress test is worth the 30 minutes it takes:

• After overclocking or undervolting – You changed voltage, clock speed, or memory frequency. The card seemed fine in a quick game, but that doesn’t mean it will survive 30 minutes of sustained peak load.
• New build or used GPU purchase – Verify the card works correctly before the return window closes. Used cards from mining rigs may pass a quick boot test but fail under extended stress.
• Random crashes or visual glitches – Screen flickering, colored pixels during gaming, or driver crashes (black screen followed by “Display driver stopped responding”) all point toward GPU instability.
• Thermal concerns – Your GPU hits 90C+ during gaming and you want to know if it will throttle or shut down under worst-case load.

Best GPU Stress Test Tools Compared

Not every stress test tool hits your GPU the same way. Some hammer the shader units, others stress the VRAM, and a few push both simultaneously. Here is how the four most popular tools compare.

FurMark – The Thermal Torture Test

FurMark renders a spinning fur-covered donut that loads your GPU harder than any game ever will. It is what GPU reviewers call a “power virus” – a workload designed to generate maximum heat. That makes it perfect for finding thermal limits but poor for simulating real gaming conditions.

How to run it: Download FurMark from Geeks3D, select your resolution (1080p or 1440p), check “GPU temperature alarm” and set it to 95C as a safety net, then click “GPU Stress Test.” Let it run for 15-30 minutes while watching temperatures in HWiNFO or the built-in overlay.

What to watch for: Temperature should stabilize within 5-10 minutes. If it keeps climbing after 10 minutes, your cooling solution is struggling. If the test crashes or your screen goes black, your GPU is hitting its thermal limit and shutting down to protect itself.

The catch: FurMark loads the GPU in a way no game does. Some GPUs that crash in FurMark run games perfectly for years. It is best used as a worst-case thermal check, not as a stability verdict for daily use. Also, some manufacturers (NVIDIA included) have built power-limiting firmware that detects FurMark-like loads and throttles the card to prevent damage.

3DMark Time Spy Stress Test – The Gold Standard

3DMark’s stress test runs the Time Spy benchmark 20 times in a row and checks whether your frame rate stays consistent across all runs. A passing score means less than 3% frame rate variation. This is the most commonly used tool for overclock validation because the workload resembles actual gaming – DirectX 12 rendering with complex shaders, tessellation, and compute tasks.

How to run it: Install 3DMark from Steam (the free Demo version includes Time Spy). Launch it, go to Benchmarks > Time Spy > Stress Test. Hit Run. The test takes roughly 20 minutes. At the end, you get a percentage score and a pass/fail result.

What to watch for: Below 97% means your overclock is unstable. Between 97-99% is borderline – the card might crash in demanding games. Above 99% is a pass. Pay attention to the frame time graph: if individual loops show progressively lower FPS, thermal throttling is kicking in.

The catch: The free version of 3DMark only includes Time Spy’s stress test. For Port Royal (ray tracing stress) or Speed Way, you need the paid version. Also, 3DMark’s stress test is less demanding than FurMark, so a card might pass 3DMark but still hit concerning temperatures under FurMark.

Unigine Heaven and Superposition – The Artifact Hunters

Unigine benchmarks loop a detailed 3D scene indefinitely, making them ideal for extended soak tests where you watch the screen for visual artifacts. Heaven is older and lighter. Superposition is newer and pushes modern GPUs harder. Both are free for the basic versions.

How to run it: Download from Unigine’s website. Set resolution to your monitor’s native res, select the highest preset your GPU supports, and click “Run.” Once the scene loads, press F9 to start benchmark mode. Let it loop and watch the screen carefully, or let it run for an hour and come back.

What to watch for: Colored pixels or blocks that flash on screen (artifacts), frame rate drops that get worse over time (thermal throttling), or full scene corruption (memory errors). Unigine’s looping nature means you can let it run as long as you want – 1 hour, 4 hours, overnight.

The catch: Superposition’s “8K Optimized” preset can bring even an RTX 4090 to its knees, but the lower presets are not stressful enough for high-end cards. Match the preset to your GPU tier. Also, Heaven is showing its age and may not push modern cards hard enough.

OCCT – The Error Detective

OCCT stands out from the others because it actively detects and logs GPU errors, not just crashes. Its 3D test stresses the GPU while monitoring for rendering errors that might not be visible to the naked eye. It also tests VRAM separately, which matters if you suspect bad memory chips.

How to run it: Download OCCT, go to the GPU tab, select “3D Standard” or “3D Adaptive” for the test type, set the duration to 1 hour, and start. The software logs every error it detects with timestamps. For VRAM testing, switch to the VRAM tab and run a separate test.

What to watch for: The error counter at the bottom. Even one error in an hour means instability. Check the graphs for temperature spikes that correlate with error timestamps. OCCT’s VRAM test is the best way to identify failing video memory – a growing problem with used GPUs.

The catch: OCCT’s free version limits test duration to 1 hour for personal use, which is usually enough. The GPU stress is moderate compared to FurMark – closer to real gaming loads. And the VRAM test takes a long time on cards with large memory pools (16 GB+ can take over an hour).

Setting Up GPU Monitoring (HWiNFO + MSI Afterburner)

Running a stress test without monitoring is like driving blind. You need to watch temperature, clock speed, power draw, and fan speed in real time to understand what is happening inside your GPU during the test.

HWiNFO64 – The Data Powerhouse

HWiNFO64 is the monitoring tool that hardware reviewers rely on. It reads every sensor on your GPU, including ones that manufacturer tools hide. Download it free from hwinfo.com.

Key sensors to watch during a stress test:

• GPU Temperature – The main die temperature. This is what most people mean by “GPU temp.”
• GPU Hot Spot (Junction) – The hottest point on the die. On NVIDIA 30-series and newer, this runs 10-15C hotter than the main sensor. This is the number that triggers thermal throttling.
• GPU Memory Junction Temperature – VRAM temperature. GDDR6X memory (on RTX 3080 and up) runs extremely hot – 90-100C is normal but above 105C causes throttling.
• GPU Clock – Real-time clock speed. If this drops during the test, the card is throttling.
• GPU Power – Wattage draw. Useful for verifying your PSU can handle the load.
• GPU Fan Speed (RPM / %) – Fan speed in percentage and RPM. If fans hit 100% and temps keep climbing, your cooling is maxed out.

Pro tip: In HWiNFO, right-click any sensor and select “Show Graph” to get a real-time chart. After the stress test ends, check the “Maximum” column for peak values. The “Average” column during the stabilized period (after the first 5 minutes) is your true operating temperature.

MSI Afterburner – The All-in-One

MSI Afterburner combines monitoring with overclocking controls. Its on-screen display (OSD) overlay puts GPU stats directly on your screen during the stress test, which is useful for tools like Unigine that run full-screen.

Setting up the OSD: Open Afterburner > Settings > Monitoring tab. Check the boxes for GPU temperature, GPU usage, core clock, memory clock, and fan speed. For each metric, check “Show in On-Screen Display.” Then open the stress test tool and the overlay appears in the corner.

Afterburner’s built-in monitoring is good enough for most users. HWiNFO shows more sensors (hot spot temp, VRAM temp, individual voltage rails) but requires more screen space. Use Afterburner for the OSD during tests, and HWiNFO for detailed post-test analysis.

Safe GPU Temperature Ranges

Every GPU generation has different thermal limits. A temperature that is perfectly safe for an RTX 4080 would be dangerously high for an older GTX 1060. The following chart shows safe operating ranges, warning zones, and critical thresholds for modern NVIDIA and AMD cards.

A few important notes on these ranges. AMD’s RX 7000 series cards are designed to run at higher temperatures than NVIDIA equivalents. AMD set the “junction temperature” limit at 110C, and the card boosts aggressively until it hits that wall. Seeing 95-100C on an RX 7900 XTX during a stress test is within spec, even though the same number on an RTX 4070 would be alarming.

For NVIDIA’s RTX 40 and 50 series, the GPU core typically throttles at 83-90C depending on the specific model and power limit. But the hot spot sensor can read 10-15C higher than the main temperature. If HWiNFO shows a 90C “GPU Temperature” but 105C “Hot Spot,” your card is already in trouble even though the main number looks acceptable.

GDDR6X memory has its own temperature limit of 110C. On cards like the RTX 3080, 3090, 4080, and 4090, the backside VRAM runs hot because it sits directly behind the GPU die. Replacing thermal pads on these cards with thicker, higher-quality pads can drop VRAM temps by 10-20C.

Step-by-Step: Running Your First GPU Stress Test

Here is the exact process to run a thorough GPU stability test from start to finish. This takes about 45 minutes and covers thermal limits, artifact detection, and overclock stability.

Step 1: Prepare Your System

1. Close all applications except your monitoring tool (HWiNFO or MSI Afterburner).
2. Set your Windows power plan to “High Performance” or “Ultimate Performance.”
3. If you overclocked, make sure your OC settings are applied. If testing stock, reset to defaults.
4. Wait until GPU idle temperature stabilizes (usually 35-50C depending on your case and ambient temp).

Step 2: Run FurMark for 15 Minutes (Thermal Baseline)

Start FurMark with GPU temperature alarm set to 95C. Watch the temperature curve in HWiNFO. Your GPU should stabilize within 5-8 minutes. Record the stabilized temperature, clock speed, and fan speed. If the card crashes or hits 95C, stop here and fix your cooling before proceeding.

Step 3: Run 3DMark Time Spy Stress Test (Stability Check)

After FurMark, let the GPU cool for 5 minutes back to idle temps. Launch 3DMark and run the Time Spy Stress Test (20 loops). This takes about 20 minutes. A score above 97% means the card is stable under realistic gaming loads. Below 97% means something needs adjustment.

Step 4: Run Unigine for 30+ Minutes (Artifact Scan)

Open Unigine Superposition (or Heaven if Superposition is too demanding). Set it to loop and watch the screen for visual glitches – flickering pixels, colored blocks, texture corruption, or geometry that warps incorrectly. These are artifacts. Even a single artifact during 30 minutes of testing indicates a problem.

Step 5: Review Results

Check HWiNFO’s “Maximum” column for peak temperatures during each test phase. Compare your 3DMark score to the online database for your card model. If everything passed – stable temps, 97%+ stress test score, zero artifacts – your GPU is healthy.

How to Identify GPU Artifacts

Artifacts are the most visible sign of GPU instability. They happen when the GPU or VRAM processes data incorrectly, producing visual errors on screen. Here is what each type looks like and what it means.

Colored pixels or blocks: Random bright-colored dots or small rectangles that flash on screen. Usually a VRAM issue. The memory chips are returning corrupted data. If this happens at stock settings, your VRAM may be failing.

Texture corruption: Game textures turn into a checkerboard pattern, go completely black, or display the wrong texture (a character’s face on a wall, for example). This points to VRAM errors or an aggressive memory overclock.

Screen flickering or flashing: The entire screen flickers between the correct image and a corrupted or black frame. Could be a driver issue, a failing GPU core, or a bad display cable. Test with a different cable first before blaming the GPU.

Geometry glitches: 3D objects stretch to infinity, creating long triangular spikes across the screen. This is a GPU core issue – the vertex shader is producing garbage data. Almost always caused by overclocking the core too high or a dying GPU.

Screen goes black (driver crash): The display cuts out and comes back after a few seconds with a “Display driver stopped responding and has recovered” message. This is the GPU’s way of resetting itself after an error. Frequent driver crashes during stress testing mean instability.

Important distinction: Artifacts during overclocking usually mean you pushed the clocks too high. Dial them back and retest. Artifacts at stock settings are a much bigger concern – they suggest hardware degradation. If your GPU shows artifacts at stock after being stable for months, back up your data and start shopping for a replacement.

Testing Overclock Stability

Overclocking your GPU is the most common reason to run stress tests. The process is straightforward: push the clocks up, stress test, and keep adjusting until you find the highest stable frequency.

The Overclock Testing Protocol

1. Establish your stock baseline. Run 3DMark Time Spy once at stock settings. Write down the Graphics Score. Run the Stress Test and confirm it passes above 99%.
2. Increase core clock by +50 MHz. Apply the offset in MSI Afterburner. Run 3DMark Stress Test again. Still above 97%? Good. Move to step 3. If it crashes or scores below 97%, dial back by 25 MHz.
3. Increase memory clock by +100 MHz. With your core clock set, bump memory. Run the stress test again. Memory overclocks are trickier – some cards show no artifacts but actually lose performance because error correction kicks in. Compare your Graphics Score to the baseline. If the score drops despite a higher memory clock, you’ve gone too far.
4. Run Unigine for 1 hour. 3DMark’s 20-minute stress test catches major instability. Unigine running for an hour catches the edge cases that only appear after thermal saturation.
5. Play your most demanding game for 2 hours. The final test. If your overclock survives Cyberpunk 2077 or a heavy ray-tracing title for 2 hours without crashes or artifacts, it is stable for daily use.

When “Stable” Is Not Actually Stable

A GPU overclock can pass every synthetic stress test and still crash in specific games. This happens because different games stress different parts of the GPU in different ratios. A game heavy on tessellation might expose an instability that a shader-heavy benchmark misses.

If you get random crashes in only one or two games after overclocking, try reducing the core clock by just 15-25 MHz. That small reduction often fixes game-specific instability without meaningfully impacting performance.

Signs Your GPU Is Failing

Stress testing can reveal early warning signs of hardware failure. Catching these early gives you time to back up data, order a replacement, and avoid a dead card in the middle of a ranked match.

Artifacts at stock settings that get worse over time. A healthy GPU never produces artifacts at stock clocks. If you see them and they become more frequent over weeks, the GPU is degrading. This is often caused by cracked solder joints (especially on cards that ran hot for years) or failing VRAM chips.

Progressively lower benchmark scores. Run 3DMark Time Spy today and save your score. Run it again in a month. A drop of more than 5% with the same settings and driver version suggests thermal throttling that is getting worse, or silicon degradation.

Fan behavior changes. Fans that suddenly spin to 100% at idle, fans that refuse to spin at all, or fans that make grinding or clicking noises. Fan failure is common and fixable (replacement fans cost little), but a card running with dead fans will overheat and die fast.

Driver crashes that increase in frequency. An occasional driver crash happens to everyone. Driver crashes multiple times per week, especially at idle or during light use, point to a hardware problem rather than a software bug.

Black screen on boot that requires reseating. If your PC boots to a black screen and only works after removing and reinserting the GPU, the PCIe contacts or the card’s VRM may be failing.

Diagnosing GPU Issues with Stress Tests

When something goes wrong during a stress test, the specific failure mode tells you what to fix. Use this flowchart to narrow down the problem.

Common Mistakes When GPU Stress Testing

These mistakes lead to misleading results or unnecessary panic. Avoid them.

Running FurMark and declaring your GPU “failing” because it hit 90C. FurMark generates more heat than any game. A GPU that hits 90C in FurMark might top out at 78C in the most demanding games. Use FurMark to find your thermal ceiling, not as a pass/fail test for daily use.

Not closing background applications. Chrome eating 4 GB of RAM, Discord screen sharing, and Windows Update downloading in the background all interfere with stress test results. Your GPU might throttle because the CPU is overwhelmed by background tasks, not because the GPU itself is unstable.

Testing for only 5 minutes. Some instabilities only appear after thermal saturation – when the heatsink, VRM, and surrounding PCB all reach their maximum temperature. This takes 15-30 minutes. A 5-minute test catches obvious problems but misses edge-case instability.

Daisy-chaining PCIe power cables. If your GPU needs two 8-pin power connectors, use two separate cables from your PSU. A single cable split with a daisy-chain adapter can’t deliver enough power under peak load, causing crashes that look like GPU instability but are actually a power delivery problem.

Ignoring VRAM temperatures on GDDR6X cards. Checking only the GPU core temperature and ignoring VRAM junction temp. On RTX 3080/3090/4080/4090 cards, the VRAM can hit thermal limits before the core does. HWiNFO shows this as “GPU Memory Junction Temperature.”

Panicking over a single crash. A one-time crash during a stress test could be a driver hiccup, a Windows background process, or a momentary power fluctuation. Run the test again. If it passes the second time, the crash was likely not GPU-related. Consistent, repeatable failures are what matter.

Which Stress Test Should You Run?

Here is the short version. Pick the scenario that matches your situation.

You just overclocked and want to know if it is stable: Run 3DMark Time Spy Stress Test. It is the industry standard for overclock validation. If you score above 97%, follow it with 1 hour of Unigine to catch edge cases.

You bought a used GPU and want to verify it works: Run FurMark for 30 minutes to check thermal limits, then OCCT for 1 hour to check for errors. If it passes both, the card is healthy.

Your PC crashes randomly during games: Start with OCCT’s GPU test (it logs errors with timestamps). If no GPU errors appear, the problem is likely your CPU, RAM, or PSU – not the GPU. Run OCCT’s VRAM test separately to rule out memory issues.

You want a quick 15-minute check: FurMark. It pushes your GPU to its absolute thermal limit faster than anything else. If it survives 15 minutes without crashing, your cooling is adequate for gaming.

You are troubleshooting visual artifacts: Run Unigine Heaven or Superposition on a loop and watch the screen. Artifacts during a Unigine loop at stock settings confirm a hardware problem.

Related: best GPU, identify GPU, temps.

2026 update: The RTX 5070 and RX 9070 XT redefined mid-range GPU value. Our GPU guide covers current recommendations.

Frequently Asked Questions

How long should I run a GPU stress test?

Run a minimum of 30 minutes for a basic stability check. For overclock validation, 1 hour is the standard. Most thermal issues show up within the first 15 minutes as the heatsink saturates, but VRAM-related instability can take longer to surface. If you want full confidence in a used GPU purchase, let OCCT run for 2-4 hours.

Can a GPU stress test damage my graphics card?

No, not on a healthy card with working cooling. Modern GPUs have thermal protection that throttles clock speed or shuts down the card before temperatures reach damaging levels. The only risk is running a stress test on a card with a dead fan or blocked heatsink – but that card would also overheat during gaming. Stress testing does not cause wear beyond normal heavy gaming use.

What temperature is too hot for a GPU during stress testing?

Above 95C on the main GPU sensor is too hot for most NVIDIA cards, which start heavy throttling at 83-90C. AMD RX 7000 series cards run hotter by design, with a junction limit of 110C. Check VRAM temps separately on GDDR6X cards – above 105C on the memory junction means your thermal pads need replacing.

Is FurMark safe to use on my GPU?

Yes, FurMark is safe on any modern GPU. NVIDIA cards from the GTX 900 series onward have firmware-level power limits that prevent FurMark from exceeding the card’s thermal design power for extended periods. The card will throttle itself long before reaching dangerous temperatures. Older myths about FurMark “killing” GPUs came from cards with no thermal protection, which have not been sold since roughly 2014.

My GPU passes stress tests but still crashes in games – what is wrong?

The most common cause is a driver bug specific to that game. Use DDU (Display Driver Uninstaller) to completely remove your current driver, then install the latest version fresh. If that does not help, check your RAM stability with MemTest86 – bad RAM causes game crashes that look identical to GPU problems. Also verify your PSU has enough wattage headroom – a PSU running at 95% capacity can fail under the combined load of GPU plus CPU during gaming.