2.3.1. General CUDA

• Arm platforms are now unified in CUDA Toolkit, enabling single-install and consistent builds across all Arm architectures. Note that this applies to new architectures only and does not apply to Jetson Orin, which remains as-is.

• Updated vector types with 32-bit alignment for increased load/store performance on Blackwell (see additional details below).

• Added support for the following new operating system distributions:

  • Red Hat Enterprise Linux 10.0 and 9.6

  • Debian 12.10

  • Fedora 42

  • Rocky Linux 10.0 and 9.6

• SM101 has been renumbered as SM110 from this release.

• CUDA 13.0 supports all NVIDIA architectures Turing through Blackwell, including GB200, GB300 NVL72, RTX PRO Blackwell, and the GeForce RTX 5000 series.

• Added support for cuMemCreate and cudaMallocAsync on the host, with CU_MEM_LOCATION_TYPE_HOST.

• Improved error reporting from CUDA batched memcpy APIs, using CUDA rich error reporting introduced in CTK 12.9. As a result, the batched memcpy APIs have been superseded with a new function prototype eliding the now-unnecessary failIdx out parameter.

• Added CUDA Runtime API support for rich error reporting.

• cuda-checkpoint utility updated to allow GPU migration. Users can now specify how GPUs from the old and new machines should be matched by specifying UUID pairs.

• Coherent memory platforms (including Grace Hopper) can now be initialized in non-NUMA mode, where video memory and kernel memory are managed separately as on more traditional platforms (this does not affect coherency or bandwidth, only memory organization).

• Fatbin file compression now uses Zstd instead of LZ4, improving compression ratios and resulting in smaller generated binaries.

• Added cuGreenCtxGetId function to allow unique identification of green context IDs without needing to convert from a primary context.

• Registers can now additionally spill to shared memory, which is approximately 10x lower latency than spilling to L2 cache. This feature is opt-in and controllable via the pragma enable_smem_spilling.

• CUDA Runtime now uses contextless loading.

• Hostnames are now supported as part of nvidia-imex nodes_config domain definitions. This allows dynamic reassignment of the underlying IP addresses of compute nodes by updating the DNS or /etc/hosts files.

• All Windows userspace .exe and .dll files are now signed.

• Added support for managed memory discard in UVM, through the following new APIs:

  • cuMemDiscardBatchAsync

  • cuMemDiscardAndPrefetchBatchAsync

2.3.2. CUDA Compiler

• For changes to PTX, refer to https://docs.nvidia.com/cuda/parallel-thread-execution/#ptx-isa-version-9-0.

• CUDA Runtime API:

  • The prototype for the CUDA Runtime API function cudaGraphKernelNodeCopyAttributes has been updated. It now takes hDst followed by hSrc to align with the calling convention used by most copy APIs.

• nvJitLinkCreate:

  • The options parameter of nvJitLinkCreate is now of type const char * const * instead of const char **.

• General Compiler (NVCC, nvFatbin, fatbinary):

  The default compression mode is now equivalent to using --compress-mode=balance. As a result, newly generated binaries will benefit from the higher compression ratio provided by the Zstandard algorithm.

• NVCC:

  A new flag, -jobserver, has been added to NVCC to control thread usage by -split-compile and --threads when invoked under GNU Make with -jN (where N > 1). This option is supported on Linux only.

  • Requires GNU Make 4.4+ to work out of the box.

  • Can also work with GNU Make 4.3+ with appropriate Makefile modifications.

  For more details, see the NVCC option listing.

• Fatbinary:

  The -image option, which was previously deprecated, has been removed in CUDA 13.0. Users should use -image3 or -image2 instead, depending on their requirements.

• New host compiler support in CUDA 13.0:

  • Clang 20

  • GCC 15

• libNVVM

  • libNVVM has been updated to align with the LLVM 20 release.

  • Split from the NVCC component

• libNVPTXCompiler

  • Split from the NVCC component

• CUDA CRT

  • Split from the NVCC component

2.3.3. CUDA Developer Tools

• A new developer tool, Compile Time Advisor (ctadvisor), has been introduced in CUDA 13.0. This tool helps users analyze the compilation time of their CUDA C++ code and provides actionable suggestions to reduce build time.

For details on new features, improvements, and bug fixes, see the changelogs for:

• Nsight Systems.

• Nsight Visual Studio Edition.

• CUPTI.

• Nsight Compute.

• Compute Sanitizer.

• CUDA-C++-Programming-Guide.

2.3.4. CUDA Runtime

• CUDA cuLibOS

  Split from cudart

2.4.1. General CUDA

• The behavior of cuInit on Windows has been unified with Linux for systems using NVSwitch. If the Fabric Manager service is not running and the system is expected to use NVSwitch or NVLink for P2P connectivity, cuInit will return CUDA_ERROR_NOT_READY until the configuration issue is resolved.

2.4.2. CUDA Compiler

• Under certain scenarios involving an over-aligned type and the MSVC #pragma pack directive, NVCC incorrectly computed the size, alignment, and other layout characteristics of the type.

  This issue can occur when using an over-aligned type as the embedded object of std::optional<T> from the MSVC STL. It only affects Windows when NVCC is used with an MSVC host compiler.

  For example, NVCC may incorrectly determine the layout of variable z in the following code:

  struct alignas(16) complex {
      double x;
      double y;
  };
  
  std::optional<complex> z;
  

  This may lead to a crash or out-of-memory error during later compilation stages. The issue was introduced in CUDA 12.9. It is fixed in CUDA 13.0.

• When the [[no_unique_address]] attribute is applied to an empty member variable within an otherwise empty object, NVCC may crash during the device compilation stage.

  For example, the following code could trigger a crash:

  struct Foo { };
  struct Bar {
    [[no_unique_address]] Foo x{};
  };
  

  This issue is resolved in CUDA 13.0.

2.5.1. General CUDA

• Certain Linux kernels with KASLR enabled have a known issue in HMM initialization, causing CUDA initialization to fail. This issue is indicated by the following debug message:

[64689.125237] nvidia-uvm: uvm_pmm_gpu.c:3176 devmem_alloc_pagemap[pid:92821] request_free_mem_region() err -34

Fixes to this issue are being handled in upstream kernels. In the meantime, you can use one of the following workarounds:

• Option 1: Disable KASLR (Preferred option)

  If using GRUB, edit /etc/default/grub and add nokaslr to GRUB_CMDLINE_LINUX_DEFAULT:

  GRUB_CMDLINE_LINUX_DEFAULT="quiet splash nokaslr"
  

  Then, update GRUB and reboot:

  sudo update-grub
  sudo reboot
  

• Option 2: Disable HMM for UVM

  1. Create or edit /etc/modprobe.d/uvm.conf.

  2. Add or update the following line:

     options nvidia_uvm uvm_disable_hmm=1
     

  3. Unload and reload the nvidia_uvm kernel module or reboot the system:

  sudo modprobe -r nvidia_uvm
  sudo modprobe nvidia_uvm
  

• On Mediatek Dimensity Auto Cockpit platforms, setting the NvSciBufGeneralAttrKey_EnableGpuCompression attribute causes NvSciBuf import into CUDA to fail with the following error:

  cudaErrorInvalidValue / CUDA_ERROR_INVALID_VALUE

  Compression is not supported in CUDA for NvSciBuf interoperability on this platform.

• Starting with CUDA 13.0, on Windows Server 2022 systems with NVSwitch deployments in TCC mode, cuInit() may return CUDA_ERROR_NOT_READY if the Fabric Manager (FM) is not running or the GPU fabric is misconfigured. [5328746]

  This change aligns Windows behavior with Linux, where the driver now fails initialization instead of silently falling back to reduced performance.

  To avoid this issue:

  • Ensure the Fabric Manager service is correctly installed and running.

  • Alternatively, configure the RmGpuFabricProbe registry key if the system is intended to run without NVSwitch and Fabric Manager.

  This change may introduce regressions in legacy workflows that relied on the previous fallback behavior. It is recommended to validate affected systems using NVSwitch on Windows.

2.5.2. cuDLA

• Calling cuDLA APIs on the Tegra THOR platform is not supported, as the THOR SoC does not include a DLA engine. Attempting to do so may result in unexpected behavior.

2.6.1. General CUDA

• CUDA 13.0 deprecates the following legacy vector types:

  • double4

  • long4

  • ulong4

  • longlong4

  • ulonglong4

  These types are being replaced by new aligned variants:

  • *_16a and *_32a (e.g., double4_16a, double4_32a)

  Deprecation warnings can be managed as follows:

  • Globally silenced by defining __NV_NO_VECTOR_DEPRECATION_DIAG

  • Locally suppressed using the macro pair __NV_SILENCE_HOST_DEPRECATION_BEGIN / __NV_SILENCE_HOST_DEPRECATION_END

  These legacy types are planned for removal in CUDA 14.0.

• The CUDA installer for Windows no longer bundles the display driver. Users must install the display driver separately, either before or after installing the CUDA Toolkit.

• Multi-device launch APIs and related references for Cooperative Groups have been removed. These APIs were previously marked as deprecated in CUDA 12.x:

  • cudaLaunchCooperativeKernelMultiDevice has been removed from cuda_runtime_api.h.

  • The accompanying parameter struct cudaLaunchParam has been removed from driver_types.h.

  • this_multi_grid and multi_grid_group have been removed from cooperative_groups.h.

• Changes to cudaDeviceProperties structure:

  In CUDA 13.0, several deprecated fields have been removed from the cudaDeviceProperties structure. To ensure forward compatibility, use the recommended replacement APIs listed below:

  Removed Fields and Their Replacements

  Removed Field	Replacement API
  clockRate	cudaDeviceGetAttribute(cudaDevAttrClockRate)
  deviceOverlap	Use the asyncEngineCount field
  kernelExecTimeoutEnabled	cudaDeviceGetAttribute(cudaDevAttrKernelExecTimeout)
  computeMode	cudaDeviceGetAttribute(cudaDevAttrComputeMode)
  maxTexture1DLinear	cudaDeviceGetTexture1DLinearMaxWidth()
  memoryClockRate	cudaDeviceGetAttribute(cudaDevAttrMemoryClockRate)
  singleToDoublePrecisionPerfRatio	cudaDeviceGetAttribute(cudaDevAttrSingleToDoublePrecisionPerfRatio)
  cooperativeMultiDeviceLaunch	No replacement available

  Removed cudaDeviceAttr Types (No Replacement Available)

  • cudaDevAttrCooperativeMultiDeviceLaunch

  • cudaDevAttrMaxTimelineSemaphoreInteropSupported

• The following legacy header files related to deprecated texture and surface references have been removed from the CUDA 13.0 runtime:

  • cuda_surface_types.h

  • cuda_texture_types.h

  • surface_functions.h

  • texture_fetch_functions.h

2.6.2. Deprecated Architectures

• Architecture support for Maxwell, Pascal, and Volta is considered feature-complete. Offline compilation and library support for these architectures have been removed in CUDA Toolkit 13.0 major version release. The use of CUDA Toolkits through the 12.x series to build applications for these architectures will continue to be supported, but newer toolkits will be unable to target these architectures.

2.6.3. Deprecated or Dropped Operating Systems

• Support for Ubuntu 20.04 has been dropped starting with this release. Users are advised to migrate to Ubuntu 22.04 LTS or later.

2.6.4. Deprecated or Dropped CUDA Toolchains

CUDA Tools

• The profiling tools NVIDIA Visual Profiler and nvprof have been removed in CUDA 13.0. Users are encouraged to migrate to the next-generation tools:

  • NVIDIA Nsight Systems for GPU and CPU sampling and tracing

  • NVIDIA Nsight Compute for detailed GPU kernel profiling

• The pre-built CUDA Demo Suite applications have been removed in CUDA 13.0. Users are encouraged to build CUDA Samples from source, available on GitHub.

• The nv-p2p APIs are deprecated starting with CUDA 13.0 and will be removed in CUDA 14.0. These APIs remain functional in CUDA 13.0 but will no longer receive enhancements or bug fixes.

• The following CUPTI APIs and features have been removed in CUDA 13.0:

  • CUPTI Event API (cupti_events.h)

  • CUPTI Metric API (cupti_metrics.h)

  • PC Sampling Activity API (cupti_activity.h)

  • Source/SASS-level metrics (cupti_activity.h)

• The following APIs have been deprecated in CUDA 13.0:

  • CUPTI Profiling API (cupti_profiler_target.h)

  • Perfworks Metric API (nvperf_host.h)

  Users are encouraged to transition to the following replacements:

  • CUPTI Range Profiling API (cupti_range_profiler.h)

  • CUPTI Profiler Host API (cupti_profiler_host.h)

CUDA Compiler

• The following host compilers are no longer supported as of CUDA 13.0:

  • Intel ICC 2021.7

  • Microsoft Visual Studio 2017 (MSVC 2017)

3.1.1. cuBLAS: Release 13.0

• New Features

  • The cublasGemmEx, cublasGemmBatchedEx, and cublasGemmStridedBatchedEx functions now accept CUBLAS_GEMM_AUTOTUNE as a valid value for the algo parameter. When this option is used, the library benchmarks a selection of available algorithms internally and chooses the optimal one based on the given problem configuration. The selected algorithm is cached within the current cublasHandle_t, so subsequent calls with the same problem descriptor will reuse the cached configuration for improved performance.

    This is an experimental feature. Users are encouraged to transition to the cuBLASLt API, which provides fine-grained control over algorithm selection through the heuristics API and includes support for additional data types such as FP8 and block-scaled formats, as well as kernel fusion. (see autotuning example in cuBLASLt).

  • Improved performance of BLAS Level 3 non-GEMM kernels (SYRK, HERK, TRMM, SYMM, HEMM) for FP32 and CF32 precisions on NVIDIA Blackwell GPUs.

  • This release adds support of SM110 GPUs for arm64-sbsa on Linux.

• Known Issues

  • cublasLtMatmul previously ignored user-specified auxiliary (Aux) data types for ReLU epilogues and defaulted to using a bitmask. The correct behavior is now enforced: an error is returned if an invalid Aux data type is specified for ReLU epilogues. [CUB-7984]

• Deprecations

  • The experimental feature for atomic synchronization along the rows (CUBLASLT_MATMUL_DESC_ATOMIC_SYNC_NUM_CHUNKS_D_ROWS) and columns (CUBLASLT_MATMUL_DESC_ATOMIC_SYNC_NUM_CHUNKS_D_COLS) of the output matrix which was deprecated in 12.8 has now been removed.

  • Starting with this release, cuBLAS will return CUBLAS_STATUS_NOT_SUPPORTED if any of the following descriptor attributes are set but the corresponding scale is not supported:

    • CUBLASLT_MATMUL_DESC_A_SCALE_POINTER

    • CUBLASLT_MATMUL_DESC_B_SCALE_POINTER

    • CUBLASLT_MATMUL_DESC_D_SCALE_POINTER

    • CUBLASLT_MATMUL_DESC_D_OUT_SCALE_POINTER

    • CUBLASLT_MATMUL_DESC_EPILOGUE_AUX_SCALE_POINTER

  • Previously, this restriction applied only to non-narrow precision matmuls. It now also applies to narrow precision matmuls when a scale is set for a non-narrow precision tensor.

3.2.1. cuFFT: Release 13.0

• New Features

  • Added new error codes:

    • CUFFT_MISSING_DEPENDENCY

    • CUFFT_NVRTC_FAILURE

    • CUFFT_NVJITLINK_FAILURE

    • CUFFT_NVSHMEM_FAILURE

  • Introduced CUFFT_PLAN_NULL, a value that can be assigned to a cufftHandle to indicate a null handle. It is safe to call cufftDestroy on a null handle.

  • Improved performance for single-precision C2C multi-dimensional FFTs and large power-of-2 FFTs.

• Known Issues

  • An issue identified in CUDA 13.0 affects the correctness of a specific subset of cuFFT kernels, specifically half-precision and bfloat16 size-1 strided R2C and C2R transforms. A fix will be included in a future CUDA release.

• Deprecations

  • Removed support for Maxwell, Pascal, and Volta GPUs, corresponding to compute capabilities earlier than Turing.

  • Removed legacy cuFFT error codes:

    • CUFFT_INCOMPLETE_PARAMETER_LIST

    • CUFFT_PARSE_ERROR

    • CUFFT_LICENSE_ERROR

  • Removed the libcufft_static_nocallback.a static library. Users should link against libcufft_static.a instead, as both are functionally equivalent.

3.3.1. cuSOLVER: Release 13.0

• New Features

  • cuSOLVER offers a new math mode to leverage improved performance of emulated FP32 arithmetic on Nvidia Blackwell GPUs.

    To enable and control this feature, the following new APIs have been added:

    • cusolverDnSetMathMode()

    • cusolverDnGetMathMode()

    • cusolverDnSetEmulationStrategy()

    • cusolverDnGetEmulationStrategy()

  • Performance improvements for cusolverDnXsyevBatched() have been made by introducing an internal algorithm switch on Blackwell GPUs for matrices of size n <= 32.

    To revert to the previous algorithm for all problem sizes, use cusolverDnSetAdvOptions().

    For more details, refer to the cusolverDnXsyevBatched() documentation.

• Deprecations

  • cuSOLVERMg is deprecated and may be removed in an upcoming major release. Users are encouraged to use cuSOLVERMp for multi-GPU functionality across both single and multi-node environments. To disable the deprecation warning, add the compiler flag -DDISABLE_CUSOLVERMG_DEPRECATED.

  • cuSOLVERSp and cuSOLVERRf are fully deprecated and may be removed in an upcoming major release. Users are encouraged to use the cuDSS library for better performance and ongoing support.

    For help with the transition, refer to the cuDSS samples or CUDA samples for migrating from cuSOLVERSp to cuDSS.

    To disable the deprecation warning, add the compiler flag: -DDISABLE_CUSOLVER_DEPRECATED.

• Resolved Issues

  • The supported input matrix size for cusolverDnXsyevd, cusolverDnXsyevdx, cusolverDnXsyevBatched, cusolverDn<t>syevd, and cusolverDn<t>syevdx is no longer limited to n <= 32768.

    This update also applies to routines that share the same internal implementation: cusolverDnXgesvdr, cusolverDnXgesvdp, cusolverDn<t>sygvd, cusolverDn<t>sygvdx, and cusolverDn<t>gesvdaStridedBatched.

3.4.1. cuSPARSE: Release 13.0

• New Features

  • Added support for 64-bit index matrices in SpGEMM computation. [CUSPARSE-2365]

• Known Issues

  • cuSPARSE logging APIs can crash on Windows.

  • CUSPARSE_SPMM_CSR_ALG3 does not return deterministic results as stated in the documentation.

• Deprecation

  • Dropped support for pre-Turing architectures (Maxwell, Volta, and Pascal).

• Resolved Issues

  • Fixed a bug in cusparseSparseToDense_bufferSize that caused it to request up to 16× more memory than required. [CUSPARSE-2352]

  • Fixed unwanted 16-byte alignment requirements on the external buffer. Most routines will now work with any alignment. In the generic API, only cusparseSpGEMM routines are still affected. [CUSPARSE-2352]

  • Fixed incorrect results from cusparseCsr2cscEx2 when any of the input matrix dimensions are zero, such as when m = 0 or n = 0. [CUSPARSE-2319]

  • Fixed incorrect results from CSR SpMV when any of the input matrix dimensions are zero, such as when m = 0 or n = 0. [CUSPARSE-1800]

3.5.1. CUDA Math: Release 13.0

• New Features

  • Single and double precision math functions received targeted performance and accuracy improvements through algorithmic simplifications, reduced branching, and tighter approximations.

    • atan2f, atan2: Up to 10% faster with minor improvements in accuracy.

    • sinhf, coshf, acoshf, asinhf, asinh: Up to 50% speedups with minor improvements in accuracy.

    • cbrtf, rcbrtf: 15% faster with minor improvements in accuracy.

    • erfinvf, erfcinvf, normcdfinvf: Minor accuracy improvements, performance neutral.

    • ldexpf, ldexp: Up to 3x faster in single precision and 30% faster in double precision, with no accuracy loss.

    • modff, modf: Up to 50% faster in single precision and 10% faster in double precision, with no accuracy loss.

3.6.1. nvJPEG: Release 13.0

• Deprecations

  • Removed the nvjpegEncoderParamsCopyHuffmanTables API.

Resolved Issues

• nvJPEG is now more robust and no longer crashes or exhibits undefined behavior when decoding malformed or truncated bitstreams. [5168024, 5133845, 5143450]

• nvjpegEncodeYUV now avoids reading outside of allocated device memory in certain cases. [5133826]

• Optimized memory usage when encoding RGB inputs using the hardware encoder.

• Fixed issues related to rounding in various transform, sampling, and conversion steps, improving image quality for both encoder and decoder. [5064901, 3976092]

• Various bug fixes for improved security.

3.7.1. NPP: Release 13.0

• Deprecations

  • Removal of Legacy Non-Context APIs

    All legacy NPP APIs without the _Ctx suffix have been deprecated and are now removed starting with this release. Developers should transition to the context-aware (_Ctx) versions to ensure continued support and compatibility with the latest CUDA releases.

  • Deprecation of ``nppGetStreamContext()``

    The nppGetStreamContext() API has been deprecated and removed. Developers are strongly encouraged to adopt application-managed stream contexts by explicitly managing the NppStreamContext structure. For guidance, refer to the NPP Documentation – General Conventions and the usage demonstrated in the StreamContexts example.

• Resolved Issues

  • Fixed an issue in nppiFloodFillRange_8u_C1IR_Ctx where the flood fill operation did not correctly fill the full target area. [5141474]

  • Resolved a bug in the nppiDebayer() API that affected proper reconstruction of color data during Bayer pattern conversion. [5138782]