AMD Radeon RX 9070 XT
17.1 GB reported video memory
- Video memory
- Image accuracy
- Graphics accuracy
- Stability under load
- Temperature monitoring
Detailed results
Hardware specifications
- Reported video memory
- 17.1 GB
Rendering feature checks
- Point-sampled texturesChecks that the correct texture pixels are selected, including where a texture repeats at its edges.
- Passed
- Filtered texturesChecks that colours between neighbouring texture pixels are blended correctly.
- Passed
- Depth orderingChecks that nearer surfaces hide farther surfaces and that depth values are written correctly.
- Passed
- Alpha blendingChecks that transparent colours combine correctly with the background.
- Passed
Small off-screen images are compared with independently calculated expected pixel values. Results show the initial check and the repeat near the end when recorded. These checks do not cover every graphics feature or the physical display output.
GPU load response
- Window completion
- Completion state not recorded by this checker version
- Observation
- Temperature control reduced the rendering workload
- Conclusion
- Temperature control reduced workload during the observation; unrestricted performance retention could not be assessed. The request ended at 86.0%, below the 95.0% request at the start of this observation.
- Observation elapsed time
- 30 seconds
- Time at original 95% request
- 3.28 seconds
- Observed time below original request
- Not recorded by this checker version
- Observed pause time
- 0 seconds
- Early / late comparison
- Not comparable or unavailable
- Comparison evidence
- Rendering settings changed or could not be confirmed; the early and late work is not comparable.
- Median GPU utilisation · rendering segment
- 79 %
- Peak GPU core · rendering segment
- 51 °C
- Peak GPU hotspot · rendering segment
- 87 °C
- Temperature-driven reductions · rendering segment
- 1
One fixed window after preparation; pauses and reductions remain inside it. Safety changes to rendering settings remain active and prevent a comparable performance result. Recorded active rendering time covers batch submission, readback and image checking; it excludes pacing, control and memory waits and is not GPU busy time. Boundary-crossing or missing batches are not credited. Below-original time includes pauses; unaccounted time includes telemetry gaps and request transitions. Time at the original request is accumulated, not uninterrupted recovery. GPU utilisation includes other applications, whose interference is not removed from the timing comparison. This short observation does not establish maximum cooling capacity or long-term stability. Existing accuracy and load-coverage results remain separate.
Graphics verification
- Graphics load errors recorded
- 0
Counts describe work verified by the checker. Draw operations and checked frames are different measures; neither is a gaming frame-rate benchmark.
Video-memory coverage
- Video-memory sample verified
- 12.05 GB (70.51 %)
- Memory architecture
- Dedicated graphics memory
- Added through bounded growth
- 50.33 MB
- Sample limited by
- Verification time remaining in the growth window
- Recorded sample evidence
- All required patterns checked across the recorded sample. Delayed reread: completed across the verified sample
Verification time remaining in the growth window. The verified allocation was 70.51 % of reported video memory. A passed memory result applies to the checked sample; these restrictions alone do not indicate faulty memory. Only the allocated sample was checked. Its percentage does not establish physical VRAM coverage; repeated sweeps do not increase coverage.
Graphics load coverage
- Peak GPU utilisation · graphics stage
- 86 %
Stage utilisation includes other applications, lighter work and pauses. Temperature control or scheduling can reduce load; averages and brief peaks do not establish sustained full-load cooling.
GPU-focused observation
- Coverage result
- Measured load-coverage target reached
- GPU-focused phase observed
- 90.19 seconds
- Qualifying higher-load time
- 65.03 seconds; 60 seconds required
- Median GPU utilisation · focused phase
- 71 %
Extra time compensates for delayed load coverage, up to one minute. The prototype target requires measured higher load, fresh sensor readings and a recent stable observation window. Temperature controls remain active. This is separate from calculation correctness and does not certify full-load cooling or long-term stability. GPU utilisation includes other applications.
Display-driver recovery
- Windows recovery events
- None recorded
What does display-driver recovery mean?
Windows reported that the display driver stopped responding and recovered. You may have noticed a brief freeze, flicker or black screen.
This is a stability warning, not proof that the graphics card is defective. The event alone does not distinguish a driver problem from a hardware or system issue. Review the Windows event at the test time and the saved diagnostics before drawing a conclusion.
A successful image or calculation check does not cancel a recovery event. The event therefore prevents an overall full pass.
Covers Windows Display event 4101. This does not catch every GPU reset or identify its cause.
Advanced details
Additional measurements
- Peak GPU core temperature
- 52 °C
- Peak GPU hotspot temperature
- 87 °C
Peak temperatures are brief observations across the run, not a sustained cooling measurement.
GPU memory temperature
- Peak GPU memory temperature
- 76 °C
Peak from the GPU memory temperature sensor, separate from GPU core and hotspot. A peak alone does not establish sustained cooling performance.
Graphics verification
- Rendered frames checked
- 11
- Pixels checked
- 132,227
- Load batches checked
- 1,361
- Draw operations submitted
- 140,421
Counts describe work verified by the checker. Draw operations and checked frames are different measures; neither is a gaming frame-rate benchmark.
Video-memory coverage
- Sample selection
- Based on reported capacity and available process budget
- Memory patterns completed
- 7
- Complete verification sweeps
- 4
This is an allocation-size comparison, not proof that the same percentage of physical VRAM cells was tested. Repeated sweeps recheck the allocation and do not increase unique memory coverage.
Graphics load coverage
- Paused stage time
- 0.2 seconds
Active stage time does not mean continuous full GPU utilisation. Early warm-up is included in stage time but excluded from dedicated load coverage. Utilisation includes other applications; the stage includes lighter work and pauses. A short peak does not establish sustained full-load cooling.
GPU load response · advanced details
- Rendering timing comparison
- Render configuration changed or could not be confirmed
- Requested workload · segment start
- 95 %
- Requested workload · segment end
- 86 %
- Earlier thermal restriction at entry
- Not recorded
- Fresh temperature coverage
- 30 seconds
- Median GPU clock · rendering segment
- 2,540 MHz
The short segment uses managed workload. Timing is comparable only when the render configuration and requested load match. Clock and power readings are observations, not a performance grade. Detailed batch, CPU, wait and pacing timings remain in diagnostic evidence.