Download as PDF, PPTX
Uploaded byOfer Rosenberg
The GPGPU Continuum
The document explores the evolution and challenges of the General-Purpose computation on Graphics Processing Units (GPGPU) continuum, detailing developments from 2004 to 2013. It highlights the diverse mobile GPGPU architectures and the associated high development costs due to immature tools, distribution models, and security concerns. The vision for a unified ecosystem based on common software architecture is emphasized, along with the need for evolving programming languages and enhancing the software stack.
More Related Content
The GPGPU Continuum
- 1. THE GPGPU CONTINUUM OferRosenberg The GPU continuum workshop, April 25 2013
- 2. CONTENT • Intel’s ComputeContinuum • GPGPU Evolution • The GPGPU Continuum • Mobile GPGPU challenges • GPGPU Continuum challenges • Towards the Continuum
- 3. INTEL’S “COMPUTE CONTINUUM”FROM IDC 2010
- 4. INTEL’S “COMPUTE CONTINUUM”FROM IDC 2010
- 5. GPGPU EVOLUTION G80 –346 GFLOPS 2004 – Stanford University: Brook for GPUs 2006 – AMD releases CTM NVIDIA releases CUDA 2008 – OpenCL 1.0 released R580 – 375 GFLOPS
- 6. GPGPU EVOLUTION Nov 2009- First Hybrid SC in the Top10: Chinese Tianhe-1 1,024 Intel Xeon E5450 CPUs 5,120 Radeon 4870 X2 GPUs Nov 2010 – First Hybrid SC reaches #1 on Top500 list: Tianhe-1A 14,336 Xeon X5670 CPUs 7,168 Nvidia Tesla M2050 GPUs Source: http://www.top500.org/lists/ Tianhe-1 : 563 TFLOPS Tianhe-1A : 2577 TFLOPS
- 7. GPGPU EVOLUTION 2013 -OpenCL on : Nexus 4 (Qualcomm Adreno 320) Nexus 10 (ARM Mali T604) Android 4.2 adds GPU support for Renderscript 2014 – NVIDIA Tegra 5 will support CUDA 2013 – GPGPU Continuum becomes a reality
- 8. THE GPGPU CONTINUUM AppleA6 GPU 25 GFLOPS < 2W AMD G-T16R 46 GFLOPS* 4.5W Intel i7-3770 511 GFLOPS* 77W NVIDIA GTX Titan 4500 GFLOPS 250W ORNL TITAN SC 27 PFLOPS 8200 KW * GFLOPS of CPU+GPU Take Intel’s vision on Compute Continuum, and aspire for that on the GPGPU continuum: A common ecosystem built on a common (SW) architecture
- 9. INTRO TO LEADINGMOBILE GPU VENDORS Imagination PowerVR 543 • Apple, Samsung, Motorola, Intel • Unified Shaders • Supports OpenCL 1.1 (E) • 38 Gflops (Apple’s MP4 ver) Vivante CG4000 • Unified Shaders • 4 Cores, SIMD4 each • Supports OpenCL 1.2 • 48 Gflops Qualcomm Adreno 320 • Part of Snapdragon S4 • Unified Shader • SIMD4 ? • Supports OpenCL 1.1 (E) • 50 GFlops ARM Mali T604 • 4 Cores • Multiple “pipes” per core • Supports OpenCL 1.1 • 68 GFlops NVIDIA Tegra 4 • 6 X 4-wide Vertex shaders • 4 X 4-wide Pixel Shaders • No GPGPU support • 74 GFLOPS http://kyokojap.myweb.hinet.net/gpu_gflops/
- 10. MOBILE GPGPU CHALLENGES • ManyDifferent GPU Architectures • Optimizing for each sets high bar on development costs • Development Tools • Immature (stability, performance) • No common SDK / Debugger / Profiler (different per vendor) • Ecosystem • • Lack of libraries, wizards, middleware Slow & expensive development Distribution Model • Driver updates are part of OS distribution (no more per-month updates…) • End users are less likely to update version higher standards on stability & performance of driver release • Security – the unspoken issue (hole) …
- 11. GPGPU CONTINUUM CHALLENGES • ManyDifferent GPU Architectures • Optimizing for each sets high bar on development costs • Development Tools • Immature (stability, performance) • No common SDK / Debugger / Profiler (different per vendor) • Ecosystem • • Lack of libraries, wizards, middleware Slow & expensive development Distribution Model • End users are less likely to update version higher standards on stability & performance of driver release • Security – the unspoken issue (hole) … These challenges are a barrier to GPGPU adoption across the continuum
- 12. TOWARDS THE CONTINUUM(1) - LANGUAGES • Welcome to the GPGPU (SW) jungle … GPU
- 13. TOWARDS THE CONTINUUM(1) - LANGUAGES • Welcome to the GPGPU (SW) jungle … OpenCL Render Script GPU Direct Compute CUDA
- 14. TOWARDS THE CONTINUUM(1) - LANGUAGES • Welcome to the GPGPU (SW) jungle … PyOpenCL WebCL Aparapi (Java) OpenCL OpenACC Render Script GPU Direct Compute C++ AMP CUDA Fortran NumbaPro (Python)
- 15. TOWARDS THE CONTINUUM(1) - LANGUAGES • Welcome to the GPGPU (SW) jungle … PyOpenCL WebCL Aparapi (Java) OpenCL OpenACC Render Script GPU Direct Compute C++ AMP CUDA Fortran NumbaPro (Python) A Jungle of languages… but are these the right ones ?
- 16. TOWARDS THE CONTINUUM(1) - LANGUAGES • Current GPGPU languages are C/C++ based • There are “binding” to Python, Java, Javascript – but kernels are still C/C++ • Current developers trends: • Managed languages (Java , C#) • Scripting languages (Python, PHP) https://sites.google.com/site/pydatalog/pypl/PyPL-PopularitY-ofProgramming-Language • Higher abstraction & manageability: • More room for tools to excel on optimization • Mitigate difference between GPU architectures GPGPU languages need to evolve Data from CodeEval.com, based on 100K+ code samples
- 17. TOWARDS THE CONTINUUM(2) - SOFTWARE STACK CUDA LLVM IR Vendor X IL Vendor X GPU
- 18. TOWARDS THE CONTINUUM(2) - SOFTWARE STACK OpenCL LLVM IR Vendor X IL Vendor X GPU CUDA
- 19. TOWARDS THE CONTINUUM(2) - SOFTWARE STACK • Most GPGPU languages already use LLVM compilation framework • Slight “flavors” of LLVM IR • Most languages also posses similar “API capabilities” set OpenACC Render Script OpenCL LLVM IR Vendor X IL Vendor X GPU CUDA
- 20. TOWARDS THE CONTINUUM(2) - SOFTWARE STACK • Most GPGPU languages already use LLVM compilation framework • Slight “flavors” of LLVM IR • • Most languages also posses similar “API capabilities” set Defining a common stack based on LLVM & common API will: • Improve the compiler OpenACC Render Script OpenCL LLVM IR Vendor X IL • Increase driver quality & stability • Enable unified debugger / profiler Vendor X GPU • … Define GPGPU Virtual Machine based on LLVM CUDA
- 21. TAKEAWAYS • GPGPU Continuumis here - from Mobile devices to HPC • Vision: A common ecosystem built on a common (SW) architecture • Challenges: many architectures, immature tools, ecosystem
- 22. QUESTIONS • Q: Whatabout “Heterogeneous Computing” ? • A: Go back, replace each “GPGPU” with “Heterogeneous Computing” – and it all fits… • More ?
- 23.