Abstract
A high-performance SYMV kernel is implemented on Fermi-core GPUs using an atomic-operation based algorithm. The algorithm is effective for the memory bandwidth and reduced memory usage. On a Tesla C2050, sustained double-precision and single-precision performances of approximately 43 GFLOPS and 78 GFLOPS, respectively, were achieved. The proposed SYMV kernel also performs on a GeForce GTX580 with 72 GFLOPS and 128 GFLOPS in the double-precision and single-precision modes, respectively. The proposed SYMV kernel outperforms major CUDA BLAS kernels, CUBLAS, MAGMABLAS, and CULA-BLAS. This performance improvement has a significant impact when the SYMV kernel is plugged into user codes.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
Imamura, T., Yamada, S., Machida, M.: Development of a High Performance Eigensolver on the Peta-Scale Next Generation Supercomputer System, the Atomic Energy Society of Japan. Progress in Nuclear Science and Technology 2, 643–650 (2011)
Nath, R., Tomov, S., et al.: Optimizing symmetric dense matrix-vector multiplication on GPUs. In: Proc. of the Intl. Conf. High Performance Computing, Networking, Storage and Analysis, SC 2011 (2011)
Imamura, T.: Performance-stabilization and automatic performance tuning for DGEMV routines on a CUDA environment. IPSJ Journal, Transaction of Advanced Computing Systems, ACS 4(4), 158–168 (2011) (in Japanese)
Schäfer, A., Fey, D.: High Performance Stencil Code Algorithm for GPGPUs. In: Proc. of ICCS 2011, Procedia Computer Science, vol. 4, pp. 2077–2036 (2011)
Hwu, W.W. (ed.): GPU Computing Gems Jade Edition (Applications of GPU Computing Series). Morgan Kaufmann (2011)
NVIDIA: whitepaper NVIDIA’s Next Generation CUDA Compute Architecture: Fermi, http://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIAFermiComputeArchitectureWhitepaper.pdf
NVIDIA: CUDA CUBLAS Library, http://developer.download.nvidia.com
Agullo, E., Demmel, J., et al.: Numerical linear algebra on emerging architectures: The PLASMA and MAGMA projects. J. of Physics: Conference Series 180 (2009)
Humphrey, J.R., Price, D.K., et al.: CULA: Hybrid GPU Accelerated Linear Algebra Routines. In: SPIE Defense and Security Symposium (DSS) (2010)
Sørensen, H.H.B.: Auto-tuning Dense Vector and Matrix-Vector Operations for Fermi GPUs. In: Wyrzykowski, R., Dongarra, J., Karczewski, K., Waśniewski, J. (eds.) PPAM 2011, Part I. LNCS, vol. 7203, pp. 619–629. Springer, Heidelberg (2012)
GPUlab: GLAS library version 0.0.2, http://gpulab.imm.dtu.dk/docs/glas_v0.0.2_C2050_cuda_4.0_linux.tar.gz
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2013 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Imamura, T., Yamada, S., Machida, M. (2013). A High Performance SYMV Kernel on a Fermi-core GPU. In: Daydé, M., Marques, O., Nakajima, K. (eds) High Performance Computing for Computational Science - VECPAR 2012. VECPAR 2012. Lecture Notes in Computer Science, vol 7851. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-38718-0_9
Download citation
DOI: https://doi.org/10.1007/978-3-642-38718-0_9
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-642-38717-3
Online ISBN: 978-3-642-38718-0
eBook Packages: Computer ScienceComputer Science (R0)