Software
MAGMA Downloads

MAGMA provides implementations for CUDA, Intel Xeon Phi, and OpenCL. The latest releases are MAGMA 2.5.2, MAGMA MIC 1.4.0, and clMAGMA 1.3, respectively. The libraries available for download are listed below in the order of their release dates.

Please use any of the following publications to reference MAGMA.

MAGMA Bitbucket repository: https://bitbucket.org/icl/magma.


 

Showing 1-5 of 51 Entries
MAGMA 2.5.2
2019-11-24

MAGMA 2.5.2 is now released. Updates include:

  • New routine: magmablas_hgemm_batched for fixed size batched matrix multiplication in FP16 using the Tensor Cores.
    The routine does not currently support pre-Volta GPUs.
    The routine outperforms cuBLAS for sizes less than 100, as well as for general sizes that are not multiple of 8.
    The kernel is tuned for the notrans-notrans case only.
    Comprehensive tuning is planned in future releases;
  • Fix magmablas_?gemm_vbatched routines to correctly handle batch sizes over 65535. The same fix is applied to vbatched syrk, herk, syr2k, her2k, symm, hemm, and trmm;
  • Fix a bug in the FP32 <-> FP16 conversion routines (magmablas_hlag2s and magmablas_slag2h). The bug used to cause a launch failure for very large matrices;
  • Fix a bug in batched LU factorization to avoind NaNs when singularity is ancountered;
  • Fix a bug in batched LU factorization to ensure that the first pivot is always returned even when multilpe pivots with the same absolute value are found;
  • Add Frobenius norm for general matrices
    (supported as option to magmablas_Xlange for X = 's', 'd', 'c', or 'z').
magma-2.5.2.tar.gz   Download View License

MagmaDNN 1.1
2019-11-23

MagmaDNN 1.1 is now available. MagnaDNN provides HP data analytics and machine learning tools using MAGMA as its computational backend. Updates in this release include:

  • Bug fixes and performance improvements;
  • Distributed training;
  • Hyperparameter optimization framework improvements;
  • Benchmarks using MagmaDNN;
  • Performance comparisons, accuracy validations, etc. (w\ TensorFlow, Theano, and PyTorch).

More information on MagmaDNN 1.1 is given in this paper and presentation.

MagmaDNN's repository is on Bitbucket: https://bitbucket.org/icl/magmadnn.

release-magmadnn-v1.1.tar.gz   Download View License

MAGMA 2.5.1
2019-08-02

MAGMA 2.5.1 is now released. Updates include:

  • Updates and improvements in CMakeLists.txt for improved/friendlier CMake and spack installations;
  • Fixes related to MAGMA installation on GPUs and CUDA versions that do not support FP16 arithmetic;
  • Added support for Turing GPUs;
  • Removed some C++ features from MAGMA Sparse for friedlier compilation (using nvcc and various CPU compilers);
  • New routine: magmablas_Xherk_small_reduce (X = 's', 'd', 'c', or 'z') is a special HERK routine that assumes that the output matrix is very small (up to 32), that that the input matrix is very tall-and-skinny.
     
magma-2.5.1.tar.gz   Download View License

MAGMA 2.5.1-alpha1
2019-05-10

MAGMA 2.5.1 Alpha is now released. Updates include:

  • Updates and improvements in CMakeLists.txt for improved/friendlier CMake and spack installations;
  • Fixes related to MAGMA installation on GPUs and CUDA versions that do not support FP16 arithmetic;
  • Added support for Turing GPUs;
  • Removed some C++ features from MAGMA Sparse for friedlier compilation (using nvcc and various CPU compilers).
magma-2.5.1-alpha1.tar.gz   Download View License

MAGMA 2.5.0
2019-01-02

MAGMA 2.5.0 is now released. Updates include:

  • New routines: Magma is releasing the Nvidia Tensor Cores version of its linear mixed-precision solver that is able to provide an FP64 solution with up to 4X speedup using the fast FP16 Tensor Cores arithmetic. The release includes:
    magma_dhgesv_iteref_gpu (FP64-FP16 solver with FP64 input and solution);
    magma_dsgesv_iteref_gpu (FP64-FP32 solver with FP64 input and solution);
    magma_hgetrf_gpu        (mixed precision FP32-FP16 LU factorization);
    magma_htgetrf_gpu       (mixed precision FP32-FP16 LU factorization using Tensor Cores).
    Further details for the function names and the testing routines are given in file:
    README_FP16_Iterative_Refinement.txt
  • New routine: magmablas_Xgemm_batched_strided (X = {s, d, c, z}) is the stride-based variant of magmablas_Xgemm_batched;
  • New routine: magma_Xgetrf_native (X = {s, d, c, z}) performs the LU factorization with partial pivoting using the GPU only. It has the same interface as the hybrid (CPU+GPU) implementation provided by magma_Xgetrf_gpu. Testing the performance of this routine is possible through running testing_Xgetrf_gpu with the option (--version 3);
  • New routine: magma_Xpotrf_native (X = {s, d, c, z}) performs the Cholesky factorization using the GPU only. It has the same interface as the hybrid (CPU+GPU) implementation provided by magma_Xpotrf_gpu.
    Testing the performance of this routine is possible through running testing_Xpotrf_gpu with the option (--version 2)
  • Added benchmark for GEMM in FP16 arithmetic (HGEMM) as well as auxiliary functions to cast matrices from FP32 to FP16 storage (magmablas_slag2h) and from FP16 to FP32 (magmablas_hlag2s).
magma-2.5.0.tar.gz   Download View License

Showing 1-5 of 51 Entries
License

Copyright © 2019 The University of Tennessee. All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
· Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
· Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer listed in this license in the documentation and/or other materials provided with the distribution.
· Neither the name of the copyright holders nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

This software is provided by the copyright holders and contributors "as is" and any express or implied warranties, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose are disclaimed. in no event shall the copyright owner or contributors be liable for any direct, indirect, incidental, special, exemplary, or consequential damages (including, but not limited to, procurement of substitute goods or services; loss of use, data, or profits; or business interruption) however caused and on any theory of liability, whether in contract, strict liability, or tort (including negligence or otherwise) arising in any way out of the use of this software, even if advised of the possibility of such damage.

Dec 13 2019 Admin Login