Real GPU smoke test
`cuda-doctor validate`
validate is the truth serum. If validation fails, the install is not healthy even if reporting tools look fine.
What validation should prove
- Device selection works.
- Memory allocation and transfer work.
- A simple kernel launches and completes.
- Results are numerically sane.
- The runtime, toolkit, and driver combination works in practice.
Why this command matters most
CUDA environments often fail only when a real workload reaches the GPU. That means validation must sit above package detection, version inspection, and even compiler presence. If the smoke test cannot run, the system is not actually usable.
Device: selected
Memory: allocation + transfer passed
Kernel: launch completed
Results: sane
Status: healthyHow failures should be framed
Validation failure is final
If validate fails, the install is not healthy, even if the driver appears present and frameworks import cleanly.
Failures should point back to the likely root cause: architecture support gaps, driver/runtime incompatibility, broken framework stacks, or toolchain drift that made the system look healthier than it really is.
Related docs
Diagnose
doctor
Run a full environment diagnosis for the GPU, driver, toolkit, runtime, build chain, and validation risk.
Repair
doctor auto
Apply compatible repairs to a broken or misleading CUDA environment and refuse success until validation passes.
Execution
build
Build CUDA code in the current project with the correct toolkit, compiler, and architecture settings for the local machine.