Nvidia Corporation · NASDAQ:NVDA
Nvidia — Technical Presentation (December 2019)
13 slides from Nvidia's technical presentation, December 2019
Slide 1: CUDA on Arm Update
CUDA on Arm Update Duncan Poole | December 2019
Slide 2: Open Acceleration Platform
OPEN ACCELERATION PLATFORM NVIDIA GPU Systems, I/O Technology, Optimized Software Stacks X86 Arm POWER
Slide 3: CUDA on ARM
CUDA on ARM Technical Preview Release – Available https://developer.nvidia.com/cuda-toolkit/arm LAMMPS CONTAINERS LAMMPS GROMACS GROMACS Relion HPC APPS MILC HPC APPS MILC CONTAINERS QE DEEP QE LEARNING NGC TensorFlow NVIDIA NAMD NAMD HOOMD-blue Container CUDA Base HOOMD-blue Runtime Containers GRAP
Slide 4: CUDA on ARM: What's Missing & Target Availability
CUDA on ARM What’s Missing from Technical Preview Release & Target Availability • CUDA-X Libraries • System Management • CUDA-X Libraries • cuTENSOR • DCGM • TensorRT • Graphics • Operating Systems • Communication Libraries • EGL • SUSE SLES 15.1 • GPUDirect Storage • Compilers • OEM Systems and GPU
Slide 5: HPC Containers for Arm
HPC CONTAINERS FOR ARM • HPC | Visualization | DL GROMACS HOOMD-blue LAMMPS MILC NAMD • Runs in Singularity • EA Now – ngc.nvidia.com • GA Q1’20 NVIDIA IndeX ParaView TensorFlow VMD 5
Slide 6: NVIDIA Now Accelerates HPC Applications on Arm
NVIDIA NOW ACCELERATES HPC APPLICATIONS ON ARM Up to 20x application speedup with minimal effort ThunderX2 + V100 Application Performance 20x 15x GPU Speedup 13x 10x 7x 7x GROMACS LAMMPS MILC NAMD QE Relion Benchmark Application [Dataset]: GROMACS [ADH Dodec- Dev Sandbox], LAMMPS [LJ 2.5], MILC [Ape
Slide 7: Quantum Espresso Speedup
QUANTUM ESPRESSO (QE) QE AUSURF112 Makefile Multicore CPU vs GPU Performance 4 3.3x 3.4x 3.4x Source Code 3.0x 3 Speedup Skylake qe.exe 2 Rome qe.exe 1 OpenPOWER qe.exe 0 Dual-socket Dual-socket Dual-socket Dual-socket Dual-Socket Skylake Skylake+2 V100 Rome+2 V100 P9+2 V100 Arm+2 V100 Runs complete
Slide 8: Announcing NVIDIA Server Reference Design for Arm
ANNOUNCING NVIDIA SERVER REFERENCE DESIGN FOR ARM NVIDIA Collaborating with Arm, Ampere, Cray, Fujitsu, Hewlett Packard Enterprise, and Marvell Defining a Server Reference Platform Enabling a Range of GPU-Accelerated Servers for: • Hyperscale-Cloud to Edge • Simulation to AI • High-Performance Stora
Slide 9: Enabling the Arm HPC Ecosystem
ENABLING THE ARM HPC ECOSYSTEM CONTAINERS & ORCHESTRATION EARLY ENGAGEMENTS ML/DL/ANALYTICS COMPUTATIONAL SCIENCE VISUALIZATION APPLICATIONS & IndeX OptiX FRAMEWORKS CoMet DCA++ GAMERA Omniverse PACKAGING ML / DL MATHEMATICS ALGORITHMS COMMUNICATIONS CUDA-X cuDNN cuBLAS cuFFT ArrayFire NCCL HPC-X LI
Slide 10: Customer and Partner Testimonials
“ Performance of NVIDIA’s GPU drivers on Arm is really good, comparable with the x86 platforms we tried, which is an impressive achievement in such a short time period.” — Simon McIntosh-Smith, professor of high performance computing at the University of Bristol “ We’ve started evaluating Riken’s co
Slide 11: NVIDIA+ARM HPC Software Stack Evaluation Using Wombat at NCCS
NVIDIA+ARM HPC software stack evaluation using Wombat at NCCS Applications: CoMet LAMMPS NAMD VMD DCA++ Gromacs Gamera LSMS Comparative Molecular Molecular MD Material Molecular Earthquake Material Genomics Dynamics Dynamics Vizualiz. Science Dynamics Simulator Science Benchmarks & Mini-apps: Parall
Slide 12: Summary & Takeaways
SUMMARY & TAKEAWAYS • NVIDIA is enabling customer choice and innovations in system design • Extending the CUDA platform to Arm has mostly “just worked” • Feedback from ORNL: “Most HPC applications compile and run “out of the box” with CUDA, GCC, Open MPI, RHEL 8. Preliminary performance results are
Slide 13: NVIDIA
cS NVIDIA.