The following issues have been fixed in this release:Ĭorrected the documentation for cudnnBatchNormalization* API functions, clarifying which are optional arguments and when the user needs to pass them to the API.įixed a lack-of-synchronization issue when cudnnRNNBackwardData() and cudnnRNNBackwardDataEx() calls a kernel that is not synchronized back to the application's stream. When buffers of only one size are available, the behavior of cuBLAS calls is deterministic in multi-stream setups. In earlier cuBLAS libraries, such as cuBLAS 10.0, it used the :16:8 non-adjustable configuration. The default buffer configuration in cuBLAS 10.2 and 11.0 is :16:8:4096:2, i.e., we have two buffer sizes. The first configuration instructs cuBLAS to allocate eight buffers of 16 KB each in GPU memory while the second setting creates two buffers of 4 MB each. The user can eliminate the non-deterministic behavior of cuDNN RNN and multi-head attention APIs, by setting a single buffer size in the CUBLAS_WORKSPACE_CONFIG environmental variable, for example, :16:8 or :4096:2. The kernel selection may affect numerical results. When a larger buffer size is not available at runtime, instead of waiting for a buffer of that size to be released, a smaller buffer may be used with a different GPU kernel. This is caused by two buffer sizes (16 KB and 4 MB) used in the default configuration. As described in the Results Reproducibility section in the cuBLAS Library User Guide, numerical results may not be deterministic when cuBLAS APIs are launched in more than one CUDA stream via the same cuBLAS handle. This is the result of a new buffer management and heuristics in the cuBLAS library. RNN and multi-head attention API calls may exhibit non-deterministic behavior when the cuDNN 7.6.5 library is built with CUDA Toolkit 10.2 or higher. Published Best Practices For Using cuDNN 3D Convolutions.įor the latest compatibility software versions of the OS, CUDA, the CUDA driver, and the NVIDIA hardware, see the cuDNN Support Matrix for v7.6.5.
Separated the cuDNN datatype references and APIs from the cuDNN Developer Guide into a new cuDNN API. Made performance improvements to several APIs including cudnnAddTensor, cudnnOpTensor, cudnnActivationForward and cudnnActivationBackward. The following features and enhancements have been added to this release: These release notes are applicable to both cuDNN and JetPack users unless appended specifically with (not applicable for Jetson platforms).įor previous cuDNN release notes, see the cuDNN Archived Documentation. This release includes fixes from the previous cuDNN v7.x.x releases as well as the following additional changes.
0 kommentar(er)
