CUDA error codes, with the name and the string

Every value the CUDA runtime's error enum can return, with the name cudaGetErrorName gives back and the message cudaGetErrorString gives back, printed by a program on a real GPU rather than copied from a header.

Captured on	Tesla T4 (sm_75), driver 595.84, CUDA 12.6 (V12.6.85)
Date	2026-09-01
Program	code/errors/_strings/dump_error_strings.cu
Transcript	code/errors/_strings/evidence/run-2026-09-01.txt

These strings are version-specific, and that is the point of naming the toolkit. The table below is what CUDA 12.6 printed. The 13.3 documentation lists more enumerators than 12.6 defines, so a reader on 13.x will find codes here that their runtime does not have, and codes in their runtime that are not here. Read the header above before quoting a row.

The code and the string are two different things, and neither is the enum name. A program that prints only the number gives you nothing searchable; a program that prints only the string cannot be grepped for reliably, because the text changes between releases. Print all three.

Two quirks worth knowing

cudaGetErrorString(cudaSuccess) returns no error, not an empty string. A logger that prints the message unconditionally will print "no error" on every successful call, which is how a clean run comes to look like it is reporting failures.

Exactly one message in the whole table ends in a full stop: code 222, the provided PTX was compiled with an unsupported toolchain. Every other string has no terminating punctuation. If you are matching text, copy it exactly.

Which codes poison the context

The distinction that matters at three in the morning is not severity, it is whether the CUDA context survives. Some errors kill it, and every later call in that process fails until the process restarts, so the second, third and fourth messages you see are noise. Others leave the context healthy and are safe to handle and continue from.

Sticky, measured on this card. After an illegal memory access (700) the next call returns 700 as well: cudaFree after the failure -> 700 in code/errors/unspecified-launch-failure/evidence/run-2026-09-01.txt. Same shape for a device-side assert (710), where next call (cudaFree) -> 710 in code/errors/device-side-assert-triggered/evidence/run-2026-09-01.txt. An unspecified launch failure (719) behaves the same way. Restart the process; there is nothing to recover.

Not sticky, also measured. An invalid device ordinal (101) leaves everything working, and cudaSetDevice(0) succeeds immediately after cudaSetDevice(1) fails. A failed allocation (2) takes nothing and leaks nothing, with the free-memory figure byte-identical either side of it. A rejected launch configuration (1) is refused before anything runs, and the following synchronize returns success. A missing kernel image (209) reports at the launch and leaves the context usable. And an initialization error (3) in a forked child does not harm the parent. Each of those is a transcript under code/errors/, named on the matching error page.

The rule to carry: a code raised at the launch, before any device work happened, is recoverable. A code raised by device work that already went wrong is not.

The table

All 131 rows exactly as the program printed them, cudaSuccess included. Rows with a page link to it; rows without do not pretend to have one.

Code	cudaGetErrorName	cudaGetErrorString	Page
0	cudaSuccess	no error
1	cudaErrorInvalidValue	invalid argument	/errors/invalid-argument
2	cudaErrorMemoryAllocation	out of memory	/errors/out-of-memory
3	cudaErrorInitializationError	initialization error	/errors/initialization-error
4	cudaErrorCudartUnloading	driver shutting down
5	cudaErrorProfilerDisabled	profiler disabled while using external profiling tool
6	cudaErrorProfilerNotInitialized	profiler not initialized: call cudaProfilerInitialize()
7	cudaErrorProfilerAlreadyStarted	profiler already started
8	cudaErrorProfilerAlreadyStopped	profiler already stopped
9	cudaErrorInvalidConfiguration	invalid configuration argument	/errors/invalid-configuration-argument
12	cudaErrorInvalidPitchValue	invalid pitch argument
13	cudaErrorInvalidSymbol	invalid device symbol	/errors/invalid-device-symbol
16	cudaErrorInvalidHostPointer	invalid host pointer
17	cudaErrorInvalidDevicePointer	invalid device pointer
18	cudaErrorInvalidTexture	invalid texture reference
19	cudaErrorInvalidTextureBinding	texture is not bound to a pointer
20	cudaErrorInvalidChannelDescriptor	invalid channel descriptor
21	cudaErrorInvalidMemcpyDirection	invalid copy direction for memcpy
22	cudaErrorAddressOfConstant	invalid address of constant
23	cudaErrorTextureFetchFailed	fetch from texture failed
24	cudaErrorTextureNotBound	cannot fetch from a texture that is not bound
25	cudaErrorSynchronizationError	incorrect use of __syncthreads()
26	cudaErrorInvalidFilterSetting	linear filtering not supported for non-float type
27	cudaErrorInvalidNormSetting	read as normalized float not supported for 32-bit non float type
28	cudaErrorMixedDeviceExecution	device emulation mode and device execution mode cannot be mixed
31	cudaErrorNotYetImplemented	feature not yet implemented
32	cudaErrorMemoryValueTooLarge	memory size or pointer value too large to fit in 32 bit
34	cudaErrorStubLibrary	CUDA driver is a stub library
35	cudaErrorInsufficientDriver	CUDA driver version is insufficient for CUDA runtime version
36	cudaErrorCallRequiresNewerDriver	API call is not supported in the installed CUDA driver
37	cudaErrorInvalidSurface	invalid surface reference
43	cudaErrorDuplicateVariableName	duplicate global variable looked up by string name
44	cudaErrorDuplicateTextureName	duplicate texture looked up by string name
45	cudaErrorDuplicateSurfaceName	duplicate surface looked up by string name
46	cudaErrorDevicesUnavailable	CUDA-capable device(s) is/are busy or unavailable
49	cudaErrorIncompatibleDriverContext	incompatible driver context
52	cudaErrorMissingConfiguration	global function call is not configured
53	cudaErrorPriorLaunchFailure	unspecified launch failure in prior launch
65	cudaErrorLaunchMaxDepthExceeded	launch would exceed maximum depth of nested launches
66	cudaErrorLaunchFileScopedTex	launch failed because kernel uses unsupported, file-scoped textures (texture objects are supported)
67	cudaErrorLaunchFileScopedSurf	launch failed because kernel uses unsupported, file-scoped surfaces (surface objects are supported)
68	cudaErrorSyncDepthExceeded	cudaDeviceSynchronize failed because caller's grid depth exceeds cudaLimitDevRuntimeSyncDepth
69	cudaErrorLaunchPendingCountExceeded	launch failed because launch would exceed cudaLimitDevRuntimePendingLaunchCount
98	cudaErrorInvalidDeviceFunction	invalid device function	/errors/invalid-device-function
100	cudaErrorNoDevice	no CUDA-capable device is detected	/errors/no-cuda-capable-device
101	cudaErrorInvalidDevice	invalid device ordinal	/errors/invalid-device-ordinal
102	cudaErrorDeviceNotLicensed	device doesn't have valid Grid license
103	cudaErrorSoftwareValidityNotEstablished	integrity checks failed
127	cudaErrorStartupFailure	startup failure in cuda runtime
200	cudaErrorInvalidKernelImage	device kernel image is invalid
201	cudaErrorDeviceUninitialized	invalid device context
205	cudaErrorMapBufferObjectFailed	mapping of buffer object failed
206	cudaErrorUnmapBufferObjectFailed	unmapping of buffer object failed
207	cudaErrorArrayIsMapped	array is mapped
208	cudaErrorAlreadyMapped	resource already mapped
209	cudaErrorNoKernelImageForDevice	no kernel image is available for execution on the device	/errors/no-kernel-image-is-available
210	cudaErrorAlreadyAcquired	resource already acquired
211	cudaErrorNotMapped	resource not mapped
212	cudaErrorNotMappedAsArray	resource not mapped as array
213	cudaErrorNotMappedAsPointer	resource not mapped as pointer
214	cudaErrorECCUncorrectable	uncorrectable ECC error encountered
215	cudaErrorUnsupportedLimit	limit is not supported on this architecture
216	cudaErrorDeviceAlreadyInUse	exclusive-thread device already in use by a different thread
217	cudaErrorPeerAccessUnsupported	peer access is not supported between these two devices
218	cudaErrorInvalidPtx	a PTX JIT compilation failed
219	cudaErrorInvalidGraphicsContext	invalid OpenGL or DirectX context
220	cudaErrorNvlinkUncorrectable	uncorrectable NVLink error detected during the execution
221	cudaErrorJitCompilerNotFound	PTX JIT compiler library not found
222	cudaErrorUnsupportedPtxVersion	the provided PTX was compiled with an unsupported toolchain.
223	cudaErrorJitCompilationDisabled	PTX JIT compilation was disabled
224	cudaErrorUnsupportedExecAffinity	the provided execution affinity is not supported
225	cudaErrorUnsupportedDevSideSync	the provided PTX contains unsupported call to cudaDeviceSynchronize
300	cudaErrorInvalidSource	device kernel image is invalid
301	cudaErrorFileNotFound	file not found
302	cudaErrorSharedObjectSymbolNotFound	shared object symbol not found
303	cudaErrorSharedObjectInitFailed	shared object initialization failed
304	cudaErrorOperatingSystem	OS call failed or operation not supported on this OS
400	cudaErrorInvalidResourceHandle	invalid resource handle
401	cudaErrorIllegalState	the operation cannot be performed in the present state
402	cudaErrorLossyQuery	attempted introspection would be semantically lossy
500	cudaErrorSymbolNotFound	named symbol not found
600	cudaErrorNotReady	device not ready
700	cudaErrorIllegalAddress	an illegal memory access was encountered	/errors/illegal-memory-access
701	cudaErrorLaunchOutOfResources	too many resources requested for launch	/errors/too-many-resources-requested
702	cudaErrorLaunchTimeout	the launch timed out and was terminated
703	cudaErrorLaunchIncompatibleTexturing	launch uses incompatible texturing mode
704	cudaErrorPeerAccessAlreadyEnabled	peer access is already enabled
705	cudaErrorPeerAccessNotEnabled	peer access has not been enabled
708	cudaErrorSetOnActiveProcess	cannot set while device is active in this process
709	cudaErrorContextIsDestroyed	context is destroyed
710	cudaErrorAssert	device-side assert triggered	/errors/device-side-assert-triggered
711	cudaErrorTooManyPeers	peer mapping resources exhausted
712	cudaErrorHostMemoryAlreadyRegistered	part or all of the requested memory range is already mapped
713	cudaErrorHostMemoryNotRegistered	pointer does not correspond to a registered memory region
714	cudaErrorHardwareStackError	hardware stack error
715	cudaErrorIllegalInstruction	an illegal instruction was encountered
716	cudaErrorMisalignedAddress	misaligned address	/errors/misaligned-address
717	cudaErrorInvalidAddressSpace	operation not supported on global/shared address space
718	cudaErrorInvalidPc	invalid program counter
719	cudaErrorLaunchFailure	unspecified launch failure	/errors/unspecified-launch-failure
720	cudaErrorCooperativeLaunchTooLarge	too many blocks in cooperative launch
800	cudaErrorNotPermitted	operation not permitted
801	cudaErrorNotSupported	operation not supported
802	cudaErrorSystemNotReady	system not yet initialized
803	cudaErrorSystemDriverMismatch	system has unsupported display driver / cuda driver combination
804	cudaErrorCompatNotSupportedOnDevice	forward compatibility was attempted on non supported HW
805	cudaErrorMpsConnectionFailed	MPS client failed to connect to the MPS control daemon or the MPS server
806	cudaErrorMpsRpcFailure	the remote procedural call between the MPS server and the MPS client failed
807	cudaErrorMpsServerNotReady	MPS server is not ready to accept new MPS client requests
808	cudaErrorMpsMaxClientsReached	the hardware resources required to create MPS client have been exhausted
809	cudaErrorMpsMaxConnectionsReached	the hardware resources required to support device connections have been exhausted
810	cudaErrorMpsClientTerminated	the MPS client has been terminated by the server
811	cudaErrorCdpNotSupported	is using CUDA Dynamic Parallelism, but the current configuration, like MPS, does not support it
812	cudaErrorCdpVersionMismatch	unsupported interaction between different versions of CUDA Dynamic Parallelism
900	cudaErrorStreamCaptureUnsupported	operation not permitted when stream is capturing
901	cudaErrorStreamCaptureInvalidated	operation failed due to a previous error during capture
902	cudaErrorStreamCaptureMerge	operation would result in a merge of separate capture sequences
903	cudaErrorStreamCaptureUnmatched	capture was not ended in the same stream as it began
904	cudaErrorStreamCaptureUnjoined	capturing stream has unjoined work
905	cudaErrorStreamCaptureIsolation	dependency created on uncaptured work in another stream
906	cudaErrorStreamCaptureImplicit	operation would make the legacy stream depend on a capturing blocking stream
907	cudaErrorCapturedEvent	operation not permitted on an event last recorded in a capturing stream
908	cudaErrorStreamCaptureWrongThread	attempt to terminate a thread-local capture sequence from another thread
909	cudaErrorTimeout	wait operation timed out
910	cudaErrorGraphExecUpdateFailure	the graph update was not performed because it included changes which violated constraints specific to instantiated graph update
911	cudaErrorExternalDevice	an async error has occured in external entity outside of CUDA
912	cudaErrorInvalidClusterSize	a kernel launch error has occurred due to cluster misconfiguration
913	cudaErrorFunctionNotLoaded	the function handle is not loaded when calling an API that requires a loaded function
914	cudaErrorInvalidResourceType	one or more resources passed in are not valid resource types for the operation
915	cudaErrorInvalidResourceConfiguration	one or more resources are insufficient or non-applicable for the operation
999	cudaErrorUnknown	unknown error

---

Written 2026-09-01. Every row above is the output of code/errors/_strings/dump_error_strings.cu, captured to code/errors/_strings/evidence/run-2026-09-01.txt on 2026-09-01 on ornn-internal-t4-spot-5v08: Tesla T4 (sm_75), driver 595.84, CUDA 12.6 (V12.6.85). No value in this table was typed by hand or taken from documentation. Author and reviewer: not yet assigned; this page does not publish until both are named.