AMD APU systems as a platform for scientific computing

Filip Krużel¹ , Krzysztof Banaś²

¹Cracow University of Technology,Warszawska 24, 31-155 Kraków, Poland.
²AGH University of Science and Technology, al. Mickiewicza 30, 30-059, Kraków.

DOI:

https://doi.org/10.7494/cmms.2015.2.0541

Abstract:

In our current work we investigate the possibility of using modern AMD APU architecture in scientific and technical computing. The architecture combines both a CPU and a GPU in a single Accelerated Processing Unit, which theoretically allows for shortening the time of exchanging the data between the two hardware units. This capability solves the problem of performance bottleneck related to the exchange of data between the CPU and GPU memory. Due to the structure of this architecture, it can be considered as a natural evolution of the concept presented in the IBM PowerXCell processors that have been tested during our past research (Krużel & Banaś, 2013). As reference systems we use both a system based on similar AMD architecture and a specialized Nvidia Tesla Accelerator card. Moreover, due to comparable characteristics of the CPU and GPU parts of APU we have run our computations on both hardware units separately to see the difference in performance. For testing we used our previously developed finite element numerical integration algorithm implemented in OpenCL programming framework. This algorithm has been tested with various organizations of memory and computing techniques to fully check the hardware capabilities of the APU architecture, both in terms of data exchange and calculations acceleration. Our research brings an answer to the question whether this architecture is the right future for scientific computing and whether in the next few years will be able to play a significant role in many areas of computational science.

Cite as:

Krużel, F., & Banaś, K. (2015). AMD APU systems as a platform for scientific computing. Computer Methods in Materials Science, 15(2), 363-369. https://doi.org/10.7494/cmms.2015.2.0541

Article (PDF):

Keywords:

AMD, Accelerated processing unit, APU, OpenCL, Heterogeneous system architecture, Finite element

Publication dates:

Received: 15.02.2015, accepted: 24.04.2015, published:

Publication type:

Original scientific paper

References:

AMD, AMD Developer Summit 2013, San Jose Convention Center.

Banaś, K., Krużel, F., 2014, OpenCL Performance Portability for Xeon Phi Coprocessor and NVIDIA GPUs: A Case Study of Finite Element Numerical Integration, Lecture Notes in Computer Science, 8806, 158-169.

Banaś, K., Płaszewski, P., Macioł, P., 2014, Numerical integration on GPUs for higher order finite elements, Computers & Mathematics with Applications, 67 (6), 1319-1344.

Barker, K. J., Davis K., Hoisie, A., Kerbyson, D. K., Lang, M.,

Pakin, S., Sancho J. C., 2008, Entering the petaflop era: The architecture and performance of Roadrunner, High Performance Computing, Networking, Storage and Analysis, 1-11. –

Gaster, B., Kaeli, D., Howes, L., Mistry, P., Schaa, D., 2011, Heterogeneous Computing With OpenCL, Elsevier Science & Technology.

Graczyk, R., Intel Iris Pro 5200 – test; Crysis na integrze?, available online at: http://pclab.pl/art54267.html, PClab.pl digital community, 2013, [Accessed February 10, 2015].

Halfacree, G., , 2013, AMD announces Heterogeneous Queuing tech, available online at: bit-tech http://www.bittech.net/ news/hardware/2013/10/22/amdhq/1,bit-tech [Accessed February 10, 2015].

Heirich, A., Bavoil L., 2006, Deferred Pixel Shading on the PLAYSTATION 3, available online at: http:// research.scea.com/ps3_deferred_shading.pdf, Sony Computer Entertainment US Research & Development, [Accessed February 10, 2015].

Howes, L., Munshi, A., 2014, The OpenCL Specification, Khronos OpenCLWorking Group, version 2.0, revision 26. HSA Foundation, [online] http://www.hsafoundation.com, 2013, [Accessed February 10, 2015]. Intel, OpenCL 2.0 Shared Virtual Memory Overview, Intel, 2014.

Krużel, F., Banaś, K., 2010, Finite element numerical integration on PowerXCell processors, Lecture Notes in Computer Science, 6067, 517-524.

Krużel, F., Banaś, K., 2014, Finite Element Numerical Integration on Xeon Phi coprocessor, Annals of Computer Science and Information Systems, 2, 603-612.

Krużel, F., Banaś, K., 2013, Vectorized OpenCL implementation of numerical integration for higher order finite elements, Computers & Mathematics with Applications, 66(10), 2030-2044.

Kyriazis, G., 2012, Heterogeneous System Architecture: A Technical Review, AMD, revision 1.0.

Landaverde, R., Zhang, T., Coskun, A.K., Herbordt, M., 2014, An Investigation of Unified Memory Access Performance in CUDA, IEEE High Performance Extreme Computing.

Michalik, K., Bana´s, K., Płaszewski, P., Cybułka, P., ModFem 2013, A computational framework for parallel adaptive finite element simulations, Computer Methods in Materials Science, 13 (1), 3-8. NVIDIA, CUDA C Programming Guide Design Guide, version 6.5, August 2014. NVIDIA, NVIDIA NVLink High-Speed Interconnect: Application Performance, Whitepaper, 2013.

Rul, S., Vandierendonck, H., D’ Haene J., De Bosschere, K., 2010, An experimental study on performance portability of OpenCL kernels, Application Accelerators in High Performance Computing, 2010 Symposium, Knoxville, TN, USA, 3. Van Winkle, W., 2012, AMD Fusion: How It Started, Where It’s Going, And What It Means, available online at: http://www.tomshardware.com/reviews/fusionhsa- opencl-history,3262-12.html, tom’s Hardware, [Accessed February 10, 2015].