How GPUs Are Taking Over The World: Implementing AI With
How GPUs are Taking Over the World:Implementing AI with GPUsChristopher Geiger, PE2019 FES Annual ConferenceSt Petersburg, FL1
Agenda GPUs and their Origin Practical Uses in Engineering What’s Next?2
GPUs and their Origin In the 1990’s the term “GPU” was invented to describe parallelprocessing chips used to speed image processing Led to high-end video game graphics Parallel Processing– Modern GPUs have 100’s or 1,000’s of cores– Modern CPUs have less than a s-the-difference-between-a-cpu-and-a-gpu/3
Parallel ComputingSerial Computing (Single Core) Parallel Computing is the use of multiple cores toexecute a single process’s task at the same timeand then the results are combined Note: Multiprocessing is the parallel execution ofmultiple processes on different cores.Parallel Computing (Multi-Core) Note: This is different from multithreading.Multithreading is the use of a single core formultiple processes (still in serial rallel comp//4
Repurposed Intelligently Machine learning applications soon realized they could useGPUs to do parallel computing and accelerate learning With the advent of big data anddeep learning – GPUs areessential for machine learningand other AI training in areasonable timeframe at areasonable priceNvidia5
Current Contenders NVIDIA Frontrunner AMD Challenger Intel Plans to enter the market (2020)GPU Market is only 10% as large as the CPU Market.6
Languages CUDA– Enables Nvidia cards to be used as General Purpose GPUs (GPGPUs) TensorFlow– Application library created by Google for machine learning PyTorch– Application library created by Facebook for machine learning7
GPU Practical Uses in Engineering1. Original Purpose–Rendering CAD models and advanced graphics2. General Purpose Parallel Processing–Training and production of Machine Learning/AI (e.g. NLP, objectrecognition, language translation)3. Generative Design–Using machine learning in automated design to augment thedevelopment process (e.g. mechanical structures, BIM, CCA layout)NLP Natural Language Processing; BIM Building Information Modeling; CCA – Circuit Card Assembly8
Generative Design Starting from design constraints, software develops solutions Found in CAD software from: Autodesk PTC (Creo) Siemens (NX) Dassault n/autodesk-what-is-generative-design.mp49
Engineering Software nTopologyParaMattersAltair (OptiStruct)Human & AI Collaborative DesignANSYS (Discovery Live)Enabled byDyndriteGPUsFusion360Sciart (Generative Design in Solidworks or Creo)10
Cloud GPUs GPUs are available from the major Cloud vendors AWS,Google, Microsoft and industrial players (IBM, Oracle, etc.)– Primarily used for training Deep Learning algorithms and simulations(engineering, scientific, etc.)– Benefit of elasticity– Can cut processing times significantly (hours instead of weeks)11
Custom AI Chips Google– TPUs: Built to run deployed machine learning algorithms Graphcore– IPUs: Custom built to train machine learning and run deployedalgorithmsGPUs were not invented for machine learning. Now machine learning optimized chips are entering the market.TPU Tensor Processing Unit; IPU Intelligent Professing Unit12
Edge AI Edge devices (IoT, autonomous vehicles, smartphones, etc.)need to have AI onboard to save latency and make nextgeneration applications possible– Reduced power requirements– 5G doesn’t eliminate this needNvidia Jetson Nano (70 X 45 mm) Developer Kit13
Ubiquitous Hardware For the world of IoT and 5G to make a huge positive impact onhuman life (e.g. ubiquitous computing) GPUs are involvedevery step of the wayCloud AI(GPU, TPU)Machine Learning EdgeAISmartBuildingsMachine Learning Algorithm Deployment14
What about Virtual GPUs? vGPU technology allows multiple Virtual Machines (VMs) toshare a hardware GPU (or GPUs)– A hardware GPU is still required Nvidia has GRID software to manage vGPUs1Data from Lakeside Software’s SysTrack Community, 2017. /15
Takeaways GPU-accelerated machine learning should be embedded in yourengineering practice:– In the engineering development process– In the products we produce (to best use available data) GPU (and rival) technologies are rapidly evolving- this is an areato watch or you will be disrupted16
? Questions ? Questions ? Questions ? Questions ? Questions ?17
1 How GPUs are Taking Over the World: Implementing AI with GPUs Christophe
L’ARÉ est également le point d’entrée en as de demande simultanée onsommation et prodution. Les coordonnées des ARÉ sont présentées dans le tableau ci-dessous : DR Clients Téléphone Adresse mail Île de France Est particuliers 09 69 32 18 33 are-essonne@enedis.fr professionnels 09 69 32 18 34 Île de France Ouest
Single Thread, Multiple GPUs A single thread will change devices as-needed to send data and kernels to different GPUs Multiple Threads, Multiple GPUs Using OpenMP, Pthreads, or similar, each thread can manage its own GPU Multiple Ranks, Single GPU Each rank acts as-if there’s just 1 GPU, but multiple ranks per node use all GPUs
1Conformance logs submitted for the ATI Radeon HD 5800 Series GPUs, ATI HD 5700 Series GPUs, ATI HD 5600 Series GPUs, ATI HD 5400 Series GPUs, ATI HD 5500 Serie
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 Review: Blocks and Threads With M threads/block, unique index per thread is : int index threadIdx .
2 / 44 Contents Motivation Recent Media Articles Nvidia launched their RTX GPUs CPU vs GPUs GPU Architecture Basics GPU Programming Model: CUDA Game AI on GPUs? Investigating Common AI Techniques Neural Networks and Deep Learning Nvidia’s RTX Architecture Real-Time Rendering now relies on AI!? Selected Topics of A
Kubernetes is an open-source platform for automating deployment, scaling and managing containerized applications. Kubernetes on NVIDIA GPUs includes support for GPUs and enhancements to Kubernetes so users can easily configure and use GPU resources for accelerating w
M5. The Cisco UCS C240 M5 Rack Server can host up to four NVIDIA T4 Tensor Core GPUs for AI inferencing, or up to two NVIDIA Tesla V100 Tensor Core GPUs for training workloads. The compact, 1RU Cisco UCS C220 M5 Rack Server can host up to two NVIDIA T4 Tensor Core GPUs. NetApp ONTAP. The ONTAP software built into
Studies have shown veterinary surgeons do not feel they receive adequate training in small animal nutrition during veterinary school. In a 1996 survey among veterinarians in the United States, 70% said their nutrition education was inadequate. 3. In a 2013 survey in the UK, 50% of 134 veterinarians felt their nutrition education in veterinary school was insufficient and a further 34% said it .
