GTC 2012 Jen-Hsun Huang Keynote
•
4 likes•1,354 views
This document summarizes Jen-Hsun Huang's presentation on NVIDIA's graphics technologies like CUDA and Kepler. It shows growth in CUDA usage over time in academic papers and downloads. It highlights success stories from universities using CUDA and demonstrates new capabilities of the Kepler architecture like Hyper-Q and Dynamic Parallelism. It also introduces NVIDIA's virtualized GPU technology and cloud graphics computing platform to enable graphics-intensive applications in the data center.
Report
Share
GTC 2012 Jen-Hsun Huang Keynote
- 1. Jen-Hsun Huang Co-founder, President and CEO, NVIDIA GTC | San Jose Convention Center | May 15, 2012
- 2. CUDA by the Numbers 4,000 Academic Papers 2008 2012 22,500 Academic Papers 150K CUDA Downloads 1.5M CUDA Downloads 60 Universities 560 Universities 35 Supercomputers 1 Supercomputer
- 3. Technical Papers Published 25000 20000 15000 10000 5000 0 CUDA Hadoop OpenMP 2007 2008 2009 2010 2011
- 4. “When I first programmed in CUDA I was so excited to see the progress of the algorithm literally within seconds... I was unable to sleep. ” Dr. Gary Junkin Universidad Autónoma de Barcelona “I want to say thanks to the engineers at NVIDIA! CUDA has saved my thesis. ” Stanley Reid Seibert University of Texas, Austin “GPU computing has utterly transformed the science we can do. ” Iain Couzin Princeton University
- 5. 2007 Supercomputing Exhibit Floor
- 6. 2011 Supercomputing Exhibit Floor
- 7. The Science of GTC ADAPTIVE RADIOTHERAPY DENDRITE SIMULATION COLLECTIVE BEHAVIOR COMBUSTION EFFICIENCY AIR TRAFFIC MODELING SEISMIC TOMOGRAPHY
- 11. DEMO: WATER
- 12. KEPLER THE WORLD’S FASTEST, MOST EFFICIENT HPC ACCELERATOR SMX Hyper-Q Dynamic Parallelism
- 13. 3x Energy-Efficient Performance CONTROL LOGIC 192 cores Fermi SM CONTROL LOGIC 32 cores Kepler SMX
- 14. Hyper-Q FERMI 1 Work Queue KEPLER 32 Concurrent Work Queues
- 15. Without Hyper-Q Time 100 50 0 GPU Utilization %
- 16. With Hyper-Q Time 100 50 0 GPU Utilization %
- 17. Dynamic Parallelism CPU Fermi GPU CPU Kepler GPU
- 18. Dynamic Parallelism Too coarse Too fine Adaptive
- 19. DEMO: NBODY
- 20. DEMO: GALAXY
- 21. Tesla K10 Tesla K20 3x Single Precision 1.8x Memory Bandwidth Imaging, Signal, Seismic 3x Double Precision Hyper-Q, Dynamic Parallelism CFD, FEA, Finance, Physics Available Now Available Q4 2012
- 22. Something New…
- 24. KEPLER THE WORLD’S FIRST GPU for CLOUD COMPUTING Virtualized GPU Low-Latency Remote Display Super Energy-Efficient
- 25. World’s First Virtualized GPU
- 26. “BYOD” Windows PCs Tablets and Smartphones Macs
- 28. “PC as an APP” Virtual Machine Guest OS Application Citrix XenDesktop Software Graphics Software Encoder Hypervisor
- 29. NVIDIA VGX GPU-Accelerated VDI Hypervisor Virtual Machine Guest OS Application Citrix XenDesktop NVIDIA Driver & Software Remote Display KEPLER GPU GPU MMU Per-VM Channel Virtual GPU Manager
- 30. SUMIT DHAWAN Group VP and GM
- 31. DEMO: CITRIX
- 32. GRADY COFER Visual Effects Supervisor
- 33. DEMO: ILM
- 35. DAVID YEN GM and SVP, Data Center Group
- 38. Video Distribution 250 200 150 100 50 0 Blu Ray/DVD Sales & Rentals Video Services (Pay TV + Streaming) Sources: PWC, ABI Research Billion Dollars
- 39. GeForce® GRID
- 40. Game Input Lag in Milliseconds 100 100 50 30 10 75 30 15 5 66 66 66 0 50 100 150 200 250 300 350 400 Gaikai powered by Geforce Grid Cloud Gen I Console + TV Game pipeline Display Pro Athlete Pro Gamer Average human response time MLB pitch
- 41. Game Input Lag in Milliseconds 100 100 50 30 10 75 30 15 5 66 66 66 0 50 100 150 200 250 300 350 400 Gaikai powered by GeForce GRID Cloud Gen I Console + TV Game pipeline Capture/Encode Network Decode Display
- 42. DAVID PERRY CEO and Co-Founder
- 43. DEMO
- 44. NVIDIA Cloud GPU Ecosystem Enterprise VDI Cloud Gaming Service