NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
crates.io
crates.io, #3304 most downloaded on crates.io
Safe and minimal CUDA bindings
Latest stable
0.19.10
Last release 1 months ago
24 Sep 2026
Ships fairly regularly
a new release about every 5 weeks
Nearly every release is documented
notes for 58 of the last 60 stable releases
2 versions withdrawn
withdrawn after publishing
4 years old
89 releases · first in 2022
Drag across the ruler to compare, or tap a release to open it.
feat(driver): allow selecting pinned host allocation flags by @jordan-wu-97 in #580
Full Changelog: v0.19.9...v0.19.10
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.9...v0.19.10
Added some missing cupti headers for new cuda versions by @amotzop in #592
Full Changelog: v0.19.8...v0.19.9
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.8...v0.19.9
Add CUDA 13.3 support by @corrm in #582
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.7...v0.19.8
Nothing published for this version
Fixing bug with nccl-version-from-build-system by @chelsea0x3b in #574
Full Changelog: v0.19.5...v0.19.6
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.5...v0.19.6
Add CUDA stream priority API by @w4nderlust in https://github.com/chelsea0x3b/cudarc/pull/554
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.4...v0.19.5
Add total_mem() and mem_get_info() safe methods to CudaContext by @OneThing98 in #534
has_async_alloc field in CudaContext by @LateinCecer in #553Full Changelog: v0.19.3...v0.19.4
has_async_alloc field in CudaContext by @LateinCecer in https://github.com/chelsea0x3b/cudarc/pull/553Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.3...v0.19.4
Implement DeviceRepr for arrays by @kaathewisegit in #523
DeviceRepr for arrays by @kaathewisegit in #523Full Changelog: v0.19.2...v0.19.3
DeviceRepr for arrays by @kaathewisegit in https://github.com/chelsea0x3b/cudarc/pull/523Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.2...v0.19.3
fix: add support for cufft 12.x by @mayocream in #530
Full Changelog: v0.19.1...v0.19.2
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.1...v0.19.2
Bump float8 to 0.7.0 by @EricLBuehler in #527
Full Changelog: v0.19.0...v0.19.1
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.19.0...v0.19.1
Fix memory safety issue in CudaSlice::leak and optimize Drop by @wizenink in #516
Full Changelog: v0.18.2...v0.19.0
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.18.2...v0.19.0
Add CUBIN binary format support to module loading by @SpenserCai in https://github.com/chelsea0x3b/cudarc/pull/505
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.18.1...v0.18.2
More cublas filters for cuda-13000 by @chelsea0x3b in https://github.com/chelsea0x3b/cudarc/pull/496
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.18.0...v0.18.1
Deprecate memcpy_stod,memcpy_dtov. Adds clone_htod,clone_dtoh by @chelsea0x3b in https://github.com/chelsea0x3b/cudarc/pull/493
Full Changelog: https://github.com/chelsea0x3b/cudarc/compare/v0.17.8...v0.18.0
Unified memory: slice methods. by @npatsakula in https://github.com/coreylowman/cudarc/pull/485
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.7...v0.17.8
Expose a getter for compute capability by @wizenink in https://github.com/coreylowman/cudarc/pull/477
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.6...v0.17.7
Only add typical CUDA paths to the linker search paths if no explicit path is supplied by @johanpel in https://github.com/coreylowman/cudarc/pull/475
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.5...v0.17.6
Make the driver feature not depend on nvrtc by @johanpel in https://github.com/coreylowman/cudarc/pull/468
driver feature not depend on nvrtc by @johanpel in https://github.com/coreylowman/cudarc/pull/468Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.4...v0.17.5
nccl: update bindings to 2.28.3 by @zackangelo in https://github.com/coreylowman/cudarc/pull/463
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.3...v0.17.4
Support cuda 13 for cudnn by @coreylowman in https://github.com/coreylowman/cudarc/pull/456
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.2...v0.17.3
fp8 and fp4 support by @zackangelo in https://github.com/coreylowman/cudarc/pull/449
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.1...v0.17.2
Adding safe apis for cusolver & cusolvermg by @coreylowman in https://github.com/coreylowman/cudarc/pull/438
is_culib_present function by @kaathewisegit in https://github.com/coreylowman/cudarc/pull/430Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.17.0...v0.17.1
Fix nvrtc lib for CUDA 12.6 by @jw1912 in https://github.com/coreylowman/cudarc/pull/431
nvrtc lib for CUDA 12.6 by @jw1912 in https://github.com/coreylowman/cudarc/pull/431Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.6...v0.17.0
Removing deprecated functions from cusparse by @coreylowman in https://github.com/coreylowman/cudarc/pull/421
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.5...v0.16.6
Feat: add softmax support by @kckeiks in https://github.com/coreylowman/cudarc/pull/413
CUDARC_CUDA_VERSION override to build.rs by @coreylowman in https://github.com/coreylowman/cudarc/pull/417Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.4...v0.16.5
filter cudaDeviceGetNvSciSyncAttributes by @coreylowman in https://github.com/coreylowman/cudarc/pull/407
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.3...v0.16.4
See cuda release notes here: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
See cuda release notes here: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.2...v0.16.3
Adds UnifiedSlice - safe unified memory api by @coreylowman in https://github.com/coreylowman/cudarc/pull/400
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.1...v0.16.2
Add Send & Sync for PinnedHostSlice by @torymur in https://github.com/coreylowman/cudarc/pull/394
Send & Sync for PinnedHostSlice by @torymur in https://github.com/coreylowman/cudarc/pull/394Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.16.0...v0.16.1
Add CudaContext::disable_event_tracking - a way to disable CudaEvent tracking for CudaSlice by @coreylowman in https://github.com/coreylowman/cudarc/p
CudaContext::disable_event_tracking - a way to disable CudaEvent tracking for CudaSlice by @coreylowman in https://github.com/coreylowman/cudarc/pull/391Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.15.2...v0.16.0
driver/safe test: fix 20% relative tolerance instead of absolute by @jedbrown in https://github.com/coreylowman/cudarc/pull/384
ctx::get_current() before calling ctx::set_current() by @coreylowman in https://github.com/coreylowman/cudarc/pull/388Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.15.1...v0.15.2
Add cuSparse and cuSolver unsafe FFI bindings (after v0.15 merge) by @ajz34 in https://github.com/coreylowman/cudarc/pull/381
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.15.0...v0.15.1
The breaking changes are that we are going back to the format of calling sys:: () rather than sys::lib(). ().
This is only breaking release if you depend on sys level APIs. No result or safe apis have been broken in this release
The breaking changes are that we are going back to the format of calling sys::<fn name>() rather than sys::lib().<fn name>().
Wonderful work by @Narsil to support all of this 🎉
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.14.0...v0.15.0
Export PinnedHostSlice as pub use by @aseyboldt in https://github.com/coreylowman/cudarc/pull/339
See PRs for more info
&Arc<CudaStream> by @coreylowman in https://github.com/coreylowman/cudarc/pull/350spin for no-std builds by @coreylowman in https://github.com/coreylowman/cudarc/pull/353CudaGraph api by @coreylowman in https://github.com/coreylowman/cudarc/pull/357SyncOnDrop & DevicePtr/Mut now return (CUdeviceptr, SyncOnDrop) by @coreylowman in https://github.com/coreylowman/cudarc/pull/360Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.9...v0.14.0
make build.rs check if we are building for docs.rs by @jonaspleyer in https://github.com/coreylowman/cudarc/pull/338
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.8...v0.13.9
Adds PinnedHostSlice to support pinned memory copies by @coreylowman in https://github.com/coreylowman/cudarc/pull/336
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.7...v0.13.8
Skipping layout tests for all cuda bindings by @coreylowman in https://github.com/coreylowman/cudarc/pull/334
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.6...v0.13.7
feat(adding cuFuncSetCacheConfig) by @wcork in https://github.com/coreylowman/cudarc/pull/330
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.5...v0.13.6
feat(add cuda-12080) by @wcork in https://github.com/coreylowman/cudarc/pull/329
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.4...v0.13.5
allow up to 32 parameters in launch interface of CudaFunctions by @dkales in https://github.com/coreylowman/cudarc/pull/322
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.3...v0.13.4
allow up to 20 parameters in launch interface of CudaFunctions by @dkales in https://github.com/coreylowman/cudarc/pull/321
query and synchronize event function wrappers by @OlivierDehaene in https://github.com/coreylowman/cudarc/pull/320Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.2...v0.13.3
Adds u64 support for curand::CudaRng::fill_with_uniform by @coreylowman in https://github.com/coreylowman/cudarc/pull/319
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.1...v0.13.2
Fix WSL by @CatsAreFluffy in https://github.com/coreylowman/cudarc/pull/313
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.13.0...v0.13.1
fix runtime/result.rs import_external_memory_opaque_win32() bug by @guyoung in https://github.com/coreylowman/cudarc/pull/305
&Option<T> -> Option<&T> in nccl broadcast. Add Option argument to nccl reduce. by @coreylowman in https://github.com/coreylowman/cudarc/pull/292Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.12.2...v0.13.0
Add safe methods set_pointer_mode and get_pointer_mode by @MathisWellmann in https://github.com/coreylowman/cudarc/pull/291
set_pointer_mode and get_pointer_mode by @MathisWellmann in https://github.com/coreylowman/cudarc/pull/291Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.12.1...v0.12.2
Support CUDA 12.6. by @paulstansifer in https://github.com/coreylowman/cudarc/pull/286
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.12.0...v0.12.1
[Breaking] Adds cooperative launch to LaunchAsync by @JRazek in https://github.com/coreylowman/cudarc/pull/221
cudnn and nccl non-default features in Cargo.toml by @coreylowman in https://github.com/coreylowman/cudarc/pull/275Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.9...v0.12.0
Adding options for dynamic loading that include versions after .so by @coreylowman in https://github.com/coreylowman/cudarc/pull/276
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.8...v0.11.9
Nothing published for this version
Add documentation on how to synchronize non-default streams by @sebhtml in https://github.com/coreylowman/cudarc/pull/261
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.6...v0.11.7
add another windows lib_name workaround by @brandonros in https://github.com/coreylowman/cudarc/pull/250
CudaDevice::name() and result::device::get_name() by @coreylowman in https://github.com/coreylowman/cudarc/pull/252Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.5...v0.11.6
Add CudaSlice::split_at_mut and CudaViewMut::split_at_mut by @dkales in https://github.com/coreylowman/cudarc/pull/235
CudaSlice::split_at_mut and CudaViewMut::split_at_mut by @dkales in https://github.com/coreylowman/cudarc/pull/235Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.4...v0.11.5
Using nvcc to get cuda toolkit version instead of cuda.h by @coreylowman in https://github.com/coreylowman/cudarc/pull/241
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.3...v0.11.4
Adding fallback lib name options for dynamic loading by @coreylowman in https://github.com/coreylowman/cudarc/pull/240
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.2...v0.11.3
support CUDA Toolking 12.5 by @vartec in https://github.com/coreylowman/cudarc/pull/238
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.1...v0.11.2
CUDA 12.4 by @bitemyapp in https://github.com/coreylowman/cudarc/pull/227
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.11.0...v0.11.1
Added result and safe functionality to support the cuFuncSetAttribute function. by @GaryMcD in https://github.com/coreylowman/cudarc/pull/199
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.10.0...v0.11.0
feat: add support for batch matrix-matrix product in cuBLASLt by @OlivierDehaene in https://github.com/coreylowman/cudarc/pull/186
cublaslt::result::get_matmul_algo_heuristic is now unsafe by @coreylowman in https://github.com/coreylowman/cudarc/pull/189Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.9.15...v0.10.0
Fixing 07-build-workflow. by @Narsil in https://github.com/coreylowman/cudarc/pull/175
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.9.14...v0.9.15
Enabling runtime choice for mallocAsync or sync version. by @Narsil in https://github.com/coreylowman/cudarc/pull/174
Full Changelog: https://github.com/coreylowman/cudarc/compare/v0.9.13...v0.9.14
One column per quarter.