How to Overclock Your GPU Safely (Step-by-Step 2026)

1. NEVER exceed 1.1V core voltage on NVIDIA RTX 40/50 series
2. NEVER exceed 1.2V core voltage on AMD RX 7000/9000 series
3. Memory voltage: NEVER touch it (auto only)
4. Keep core temperature below 85°C (ideally <75°C)
5. Keep hotspot temperature below 100°C (ideally <85°C)
6. VRAM temperature below 95°C (GDDR6X runs hot)
7. If you artifact, crash, or freeze: reduce clock by 30 MHz and retest
8. Always test for at least 30 minutes before calling it "stable"

Open MSI Afterburner → Click "Reset" button (the curved arrow)
This sets all sliders to default.

In MSI Afterburner:
┌────────────────────────────────┐
│ Power Limit:    ↑ [120%]  ... │ ← MAX it out (usually 110-120%)
│ Temperature Limit: ↑ [90°C] ...│ ← Set to 88-90°C for safety
│ Core Voltage:    ↑ [+0]    ...│ ← +0 until Phase 4
│ Core Clock:      ↑ [+0]    ...│ ← +0 (we'll increment)
│ Memory Clock:    ↑ [+0]    ...│ ← +0 (we'll increment)
│ Fan Speed:       ↑ [Auto]  ...│ ← Start auto, adjust later
└────────────────────────────────┘

1. 3DMark Time Spy → Default run → Record "Graphics Score"
2. 3DMark Speed Way → Default run → Record score (RT benchmark)
3. Run for 5 minutes → Record max temp, max fan speed, power draw

Write down:
  Baseline FPS: _______
  Baseline GPU temp: _______
  Baseline hotspot: _______
  Baseline VRAM temp: _______
  Baseline power draw: _______

For NVIDIA GPUs (40/50 series):
Step 1: +50 MHz core clock → Run Time Spy → If pass, go to Step 2
Step 2: +15 MHz → Run Time Spy → If pass, go to Step 2 again
Step 3: Repeat Step 2 until crash or artifact
Step 4: Back off 30 MHz from crash → This is your stable core OC

For AMD GPUs (RX 7000/9000 series):
Step 1: +100 MHz core clock → Run Time Spy → If pass, go to Step 2
Step 2: +20 MHz → Run Time Spy → If pass, go to Step 2 again
Step 3: Repeat Step 2 until crash or artifact
Step 4: Back off 40 MHz from crash → This is your stable core OC

After each increment:
- Run 3DMark Time Spy (1 loop): Must complete without crash
  → If crash: reduce by 15 MHz, test again

- After finding stable OC: Run Time Spy Stress Test (20 loops)
  → Pass threshold: 97%+ stability 
  → If lower: reduce by 15 MHz, retest

Final verification:
- 30 minutes of gaming (most demanding game you play)
- No crashes, no artifacts
- GPU temperature stays below 85°C
- Fan noise is acceptable

Step 1: Set your stable core OC from Phase 2
Step 2: +500 MHz memory clock → Run Time Spy→ If pass, go to Step 3
Step 3: +200 MHz → Run Time Spy → If pass, go to Step 3 again
Step 4: Repeat Step 3 until crash or artifacts
Step 5: Back off 300 MHz from crash → This is your stable memory OC

Critical: Memory overclocks degrade over time.
  - Test memory OC for 2+ hours (not 30 min)
  - One month later, reduce by 100 MHz (memory degrades)

Signs of memory instability:
1. Artifacts (visual glitches, flickering)
2. Driver crashes/code 43 in Windows
3. Benchmark scores DECREASING with more OC (ECC retries)
4. Color banding or weird colors in games
5. Sudden save file corruption

Check in GPU-Z → "Memory Errors" section:
  - ANY memory errors = unstable
  - Reduce by 200 MHz immediately

Undervolting gives you:
- Same performance at lower temps (-5 to -15°C)
- Same performance at lower power draw (-20 to -80W)
- Lower fan noise (fans spin slower)
- Same (or higher) boost clocks due to thermal headroom

vs Overvolting:
- Overvolting: +5% perf, +10°C, +50W = NOT worth it
- Undervolting: 0% perf gain, -10°C, -50W = Always worth it

1. Open MSI Afterburner → Ctrl+F (opens Voltage/Frequency curve)
2. Find 0.9V on the X axis (or whatever voltage you want)
3. Click on the point at 0.9V → Drag it UP to your target frequency
   → Example: For RTX 4090/5090, target 2700 MHz at 0.9V
4. All points to the right of this should be LOWERED
   → Click and drag each point to below your target line
   → Or: Hold Shift, select all points right of target, drag down
5. Click Apply (✓)
6. The curve should now be FLAT after your target point

Result: GPU will never go above 0.9V
  = Same performance at much lower temps and power

1. Open AMD Adrenalin → Performance → Tuning
2. Enable "GPU Tuning" → "Advanced Control"
3. Set "Voltage (mV)" to a lower value:
   - RX 9070 XT: Try 1050 mV (stock ~1150 mV)
   - RX 9070: Try 1000 mV (stock ~1100 mV)
   - RX 7900 XTX: Try 1050 mV (stock ~1150 mV)
4. Test with Time Spy → If stable, go lower by 10-15 mV
5. If crash, increase by 10 mV
6. Find the lowest stable voltage

Expected results:
  - RX 9070 XT: -8-15°C at same performance
  - RX 9070: -10-12°C
  - Power savings: 30-80W

✅ Stable = All of:
  - 3DMark Time Spy Stress Test: 97%+ pass
  - 30 minutes Unigine Superposition: no artifacts
  - 30 minutes OCCT VRAM: zero errors
  - 2 hours gaming: no crashes
  - Daily use for 1 week: no driver timeouts

❌ NOT Stable if:
  - Any crash in any benchmark
  - Artifacts at any point
  - Driver timeouts
  - Benchmark scores worse than stock
  - Memory errors (even 1!)

After you find your "stable" OC:
1. Run Time Spy each morning for 3 days (5 min each)
2. If it passes all 3 days → OC is stable as daily driver
3. If it fails on day 3 → Reduce by 15 MHz core or 100 MHz memory

Memory degradation:
  - Memory overclock often degrades after 2-4 weeks
  - If you crash after being stable for weeks → reduce memory by 100-200 MHz