TensorFlow GPU not detected should be diagnosed from the exact affected system, not from a code or screenshot alone. TensorFlow lists only CPU devices even though the machine has a supported NVIDIA GPU. The goal is to identify the responsible layer, make one reversible change, and verify the original failing action again.
Symptoms: recognize the failure pattern
Look for whether the failure is repeatable, limited to a single device or account, or started after a specific change. These common scenarios lead to different remedies:
- Installed TensorFlow build does not support that GPU/platform.
- CUDA, cuDNN or driver versions mismatch.
- Container/WSL GPU passthrough is not configured correctly.
Evidence to save: exact error text, affected version/model, timestamp, and the most recent relevant change. A screenshot without the surrounding context can hide the distinction between a device fault, a configuration issue and a service outage.
Before changing settings: protect your data and access
Do not mix arbitrary system CUDA DLLs or replace kernel drivers without a recovery path. Keep the current configuration, installer or backup available wherever a rollback could be needed. If a device or service is managed by an organization, obtain authorization before modifying its policy or production configuration.
First safe comparison: Record OS, GPU model, TensorFlow version and Python environment. Then note the result before moving to another action. This isolates cause and effect instead of stacking changes that cannot be evaluated separately.
Step-by-Step Troubleshooting
Start with low-risk evidence and controlled comparisons. Use Advanced for logs, targeted configuration checks and authorized repair; complete each step only after verifying its result.
Record OS, GPU model, TensorFlow version and Python environment
Record OS, GPU model, TensorFlow version and Python environment.
Do this before changing settings; it provides the reference point needed to tell a real repair from a coincidental retry.
Check the official platform support matrix; native Windows GPU
Check the official platform support matrix; native Windows GPU requirements vary by TensorFlow version.
This comparison narrows the cause while keeping unrelated accounts, files and device settings unchanged.
Test the GPU with vendor tools and confirm the Python environment
Test the GPU with vendor tools and confirm the Python environment running the script.
Write down the observed result and proceed only if the prior test gave a clear reason for this action.
Keep a working CPU environment before changing any driver or toolkit
Keep a working CPU environment before changing any driver or toolkit.
If the test does not improve the original symptom, restore the previous state and use the Advanced tab for evidence-based diagnosis.
Run TensorFlow physical device listing from the same interpreter used
Run TensorFlow physical device listing from the same interpreter used in production.
Record the exact timestamp and scope. The comparison is meaningful only if the same workload, user and target are involved; redact credentials or private records from collected logs.
Compare framework wheels and CUDA library versions against official
Compare framework wheels and CUDA library versions against official install documentation.
Compare the result against the baseline noted before changing settings. A log entry in isolation is not proof that it caused the user-visible failure.
import tensorflow as tfprint(tf.config.list_physical_devices("GPU"))For WSL/containers, inspect GPU device passthrough and supported
For WSL/containers, inspect GPU device passthrough and supported runtime configuration.
Apply one authorized, reversible change supported by the evidence. Keep the original configuration or version available so that a failed test can be undone.
Correct one dependency mismatch at a time and rerun a small GPU
Correct one dependency mismatch at a time and rerun a small GPU operation.
Repeat the original failure path and check for new errors. If evidence still points outside your access level, escalate rather than adding more speculative changes.
Diagnosis: distinguish the likely causes
The same visible failure can come from different subsystems. Use the table to decide what to inspect next; a possible cause is not a confirmed diagnosis.
| Possible cause | Evidence to collect | Decision |
|---|---|---|
| Installed TensorFlow build does not support that GPU/platform | Check the official platform support matrix; native Windows GPU requirements vary by TensorFlow version. | Proceed only if that observation supports this cause. |
| CUDA, cuDNN or driver versions mismatch | Run TensorFlow physical device listing from the same interpreter used in production. | Proceed only if that observation supports this cause. |
| Container/WSL GPU passthrough is not configured correctly | Compare framework wheels and CUDA library versions against official install documentation. | Proceed only if that observation supports this cause. |
How to verify that the repair worked
Success criterion: TensorFlow enumerates the GPU and completes a small supported compute operation without library errors. Do not declare success merely because one retry finishes; check the original operation under the same conditions that previously produced the failure.
- Repeat the exact failing action with the same authorized user, device or application settings.
- Check the relevant status, logs or user-facing result for the previously recorded signature.
- Repeat once after a normal restart or reconnect if that was part of the original failure pattern.
- Record what changed, the version or model involved, and the successful evidence.
Solution Checklist
If it still fails: rollback and escalation
If the test remains unchanged, stop applying unrelated fixes. Restore the last known working configuration and continue from the remaining evidence. For this issue, a useful next technical check is: Run TensorFlow physical device listing from the same interpreter used in production. Provide a support technician with the sanitized exact error, reproduction steps and findings, rather than passwords or personal records.
How to prevent a repeat failure
Keep compatible software, firmware, drivers or service configuration under documented change control. The most useful preventive habit for this case is to preserve a known-good baseline so you can compare installed tensorflow build does not support that gpu/platform with later behavior. Verify backup and rollback procedures before major upgrades.
Frequently asked questions
Does TensorFlow GPU not detected have one universal cause?
No. For this issue, compare installed tensorflow build does not support that gpu/platform against cuda, cudnn or driver versions mismatch. Their evidence differs, so choosing a fix from the error name alone is unreliable.
Should I reset everything to solve this problem?
No. Do not mix arbitrary system CUDA DLLs or replace kernel drivers without a recovery path. Start with the Beginner checks and move to advanced diagnostics only when the evidence supports them.
How do I know the problem is actually fixed?
TensorFlow enumerates the GPU and completes a small supported compute operation without library errors. Keep the result and the exact version information in your notes so future recurrences can be investigated more quickly.
Further reading: official documentation
- Official product or platform documentation — check the currently supported instructions for your exact version and device.
This guide prioritizes reversible diagnostic steps. Specific options and error meanings can differ by application version, device model and managed-environment policy; confirm current manufacturer guidance before potentially destructive work.