The GTC Europe 2018 opening keynote delivered by NVIDIA Founder and CEO, Jensen Huang, speaking on the future of computing.
This panel will showcase how management teams can implement new AI/DL solutions quickly and effectively, by developing talented teams successfully. It will also discuss how POCs can be moved seamlessly into productive use.
GTC Europe will feature groundbreaking work from startups using artificial intelligence to transform the world in the fields of autonomous machines, cyber security, healthcare and more. Join us to watch the hottest startups in Europe take to the stage and pitch their work for a chance to win $100,000 and a DGX Station.
Neural Networks have capitalized on recent advances on HPC, GPUs, GPGPUs, and the rising amounts of publicly available labeled data. In doing so, NN have and will revolutionize virtually every current application domain, as well as enable novel ones such as those on recognition, autonomous, predictive, resilient, self-managed, adaptive, and evolving applications.
Nevertheless, it is to point out that NN training is rather resource intensive in data, time and energy; turning the resulting trained models into valuable assets represents an IP imperatively worth of being protected.
Furthermore, in the wake of Edge computing, NNs are progressively deployed across decentralized landscapes; as a consequence, IP owners are very protective of their NN based software products.
In this session, we propose to leverage Fully Homomorphic Encryption (FHE) to protect simultaneously the IP of trained NN based software and the input and the output data.
Within the context of a smart city scenario, we outline our NN model-agnostic approach, approximating and decomposing the NN operations into linearized transformations while employing a SIMD for vectorization.
A Norwegian oil company needed to fix 1 million images of seabed, taken with artificial light, to be able to create a geological orthophoto. Due to light absorption in the water, all images were bright in center and dark at the sides. The poor quality of these images stopped us from machine analysis, so we used NVIDIA CUDA to create a routine that automatically analyzed all images one by one and fixed the inconsistent lighting. Afterwards, the images could be analysed with machine learning. The routine analysed every image separately and repaired them automatically.
Learn how to train and employ state-of-the-art object localization in a real-time safety application. In Petawatt laser systems, firing at 10Hz, suddenly appearing scatterers can damage components. Damage(-spreading) can be avoided by suspending operation immediately upon occurrence of such an event.
We present our approach for the automatic detection of critical failure states from intensity profiles of the laser beam. In order to minimize the rate of false alarms, which would reduce productivity or even render our system useless, we refrain from general anomaly detection and instead detect known error patterns. In this talk we present how we fitted the You Look Only Once(YOLO) approach, which is suited to low-latency object detection, to our problem and how we adapted the required multi-step training protocol to the available experimental data.
In this session we would like to demonstrate the benefits of multi-agent reinforcement learning for real world applications. We highlight our multi-agent system which is capable to learn an efficient communication protocol. The agents transmit relevant information through a low bandwidth channel to collectively solve a complex rescheduling problem. This approach proves beneficial whenever a feasible solution needs to be found within a short timeframe - in contrast to the computationally expensive optimal solution. Thus, our results can be applied to many different problems in the domain of operations research and transportation.
Key words: Deep Learning and AI, DGX, Reinforcement Learning, Traffic Management
We will present "Deep Variational Reinforcement Learning for POMDPs" (DVRL), a state-of-the-art algorithm to handle uncertainty in partially observable environments.
To make the step from the ivory tower into the real world, it is important for autonomous agents to learn how to deal with missing information. This can, for example, occur when sensors are noisy, objects are occluded in videos or the underlying disease of a patient is still unknown.
In contrast to previous approaches, our agent learns a generative model of its surroundings and uses it to explicitly reason about the true (unseen) world state. This requires additional computations that can be executed in parallel on a GPU: GPUs in reinforcement learning are often under-utilized when performing the action is the bottleneck, especially when the environment cannot be easily simulated or the agent is trained on-policy.
We show the success of our approach on two environments, including the standard benchmark tasks in the Atari Learning Environment.
Familiarity with Reinforcement Learning will be helpful but is not required.
This work was generously supported by NVIDIA, providing access to DGX-1s for running experiments.
In this technical deep dive, get an in-depth look at the GPU-accelerated containers for deep learning and high performance computing available on NVIDIA GPU Cloud (NGC) and learn how they can simplify your projects. NVIDIA pre-integrates and optimizes the top deep learning software such as TensorFlow, PyTorch, MXNet, and TensorRT, and makes it available on NVIDIA GPU Cloud, removing time consuming do-it-yourself software integration. We'll look at the NVIDIA framework optimizations, such as reducing GPU memory overhead, improving multi-GPU scaling, and reducing latency. We'll also talk about the integration of runtimes and drivers in the containers to ensure the best versions are all working together for peak performance. You'll leave with an understanding of what makes an NVIDIA GPU Cloud container tick.
At RoboVision we built an industrial pipeline that starts from scalable data curation to on premise deployment. Our goal is to put the internal teams at the customer in the driving seat, making them able to generate powerful deep learning models, deploying them, and getting more value from their DGX1 investments, without lengthy consultancy. With android integration and predictive labeling we enable a big crowd to annotate data, directly accessible for multi-gpu deep learning sessions. The data is stored on high speed storage systems like Pure. The tool uses nifty techniques like predictive labeling and multi-user curation to guarantee high quality input for the configurable deep learning stack. This stack is made scalable and robust, with the help of Kubernetes and MySQL clusters, ending in a restful API system for rapid integration in the client's ecosystem. After pioneering diverse applications in agriculture, RoboVision expanded its focus and is now active in industrial automation, safety and security, smart city and surveillance markets.
In this session we'll explore some of the common challenges with scaling-out deep learning training and inference deployment on data centers and public cloud using Kubernetes on NVIDIA GPUs. Through examples we'll review a typical workflow for AI deployments on Kubernetes. We'll discuss advanced deployment options such as deploying to heterogenous GPU clusters, specifying GPU memory requirements, and analyzing and monitoring GPU utilizations using NVIDIA DCGM, Prometheus and Grafana.
In current wireless networks, most algorithms are iterative and might not be able to meet the requirements of some 5G technologies such as ultra-reliable low-latency communication within a very low latency budget. For instance, requiring and end-to-end latency below 1ms, many signal processing tasks must be completed within microseconds. Therefore, only a strictly limited number of iterations can be performed, which may lead to uncontrollable excessive errors.
We argue in favor of formulating the underlying optimization problems as convex feasibility problems in order to enable massively parallel processing on GPUs for online learning for fast and robust tracking. Moreover, convex feasibility solvers allow for an efficient incorporation of context information and expert knowledge, and can provide robust results based on relatively small data sets. Our approach has numerous applications, including channel estimation, peak-to-average power ratio (PAPR) reduction in Orthogonal Frequency Division Multiplexing (OFDM) systems, radio map reconstruction, beam forming, localization, and interference reduction. We show that they can greatly benefit from the parallel architecture of GPUs.
We'll present Twitter's ML Platform and explain how it allows teams inside the company to run their models in production at Twitter's scale. Machine Learning has allowed Twitter to drive engagement, promote healthier conversations, and deliver catered advertisements. Over the past year, we have been working on a new chapter of ML at Twitter by migrating our machine learning platform to Tensorflow. This talk will be mainly focusing on this new Machine Learning platform and how we can productionalize our models with it. We will also discuss some of the ways that machine learning is used at Twitter and how we can leverage GPUs to train our models.
Interpretability of deep neural networks has focused on the analysis of individual samples. This can distract our attention from patterns originating at the distribution of the dataset itself. We broaden the scope of interpretability analysis, from individual images to entire datasets, and found that some high-performing classifiers use less than half the information contained in any given sample. While the learned features are more intuitive to visualize for image-centric neural networks, in time-series it is much more complicated as there is no direct interpretation of the filters and inputs as compared to image modality. In this talk we are presenting two approaches to analyze the behavior of networks with respect to used input signal, which pave the way to go beyond simple layer stacking and towards a more principled design of neural networks.
AI is currently penetrating many technological areas. Even though the term is used inflationary today, it often refers to machine learning, usually with deep neural networks (Deep Learning). Internet giants such as Google, Facebook, Microsoft with their almost unlimited computing capacities achieved spectacular results in image classification, text translation or in the Go game. At the same time, Earth observation has irreversibly arrived in the Big Data era with the Sentinel satellites (and in the future with Tandem-L). This requires not only new technological approaches to manage large amounts of data, but also new analysis methods. We are one of the pioneers in using Deep Learning in Earth observation and are enthusiastic about its possibilities. Going beyond quick-wins by fine-tuning existing architectures for the usual classification and detection tasks, we take particular care of the fact that Earth observation data and problems are in many aspects different from standard imagery found in the internet. In this talk, a wide spectrum of possibilities where Earth observation could tremendously benefit from methods from AI and Data Science, like deep learning, will be presented.
Analytics and AI present a serious challenge to businesses in developing new expertise and transforming data architectures from enterprise-class to AI-ready. AI workloads demand a different approach to managing the data lifecycle. The new AI datacenter must be optimized for ingesting, storing, transforming and optimizing data and feeding that data through hyper-intensive analytics workflows and ultimately, extracting value. Ensuring the maximum value of your investment into GPU platforms like NVIDIA's DGX-1 requires careful planning. Learn how to architect and deploy data platforms with robust and balanced performance for all I/O patterns.
AI has proceeded over the last 50 years in fits and starts and today the momentum has once again picked up dramatically. New discoveries and techniques in Deep Learning, the best performing AI, are emerging at a rate of one every two months. If anything, this pace is likely to accelerate in the future as this becomes relevant across many industries. While we are all worried about how to faster accelerate our own workloads, we need to also start thinking about what will come in 10 or 50 years' time. Will we be building an AI beneficial for all or for just some? Are we sure that we are putting the right goals into our AI systems? Shouldn't we make sure that the goals of a General AI are aligned with ours? What are the possible scenarios looking ahead? Let's take an objective look and discuss.
For your business projects you want to rely on solid partners to master their development and deployment. How to avoid the nightmare of cost increase or exceeding deadlines? How to benefit from industrialized solutions, avoiding demos that have been freshly issued from labs?
In this session, you will learn how Atos, with a proven set of products and services, helps you accelerate your projects in HPC, enterprise and Internet of Things domains, from cloud to on-premises, from central to edge while leveraging the most powerful NVIDIA technologies.
Because AI applications and models rely on secure, reliable and up-to-date data, this session will also introduce how Atos is managing, updating and securing data and will end up with a presentation of operational applications in the domains of image recognition, video intelligence, prescriptive maintenance and cyber security.
Mixed precision training of deep neural networks provides tremendous benefits: it requires half the storage and data movement of single-precision values, and starting with the Volta GPU's Tensor Cores, provides up to 120 TFLOPS of math throughput, an 8x speedup over FP32. In this talk, we first present the considerations and techniques when training with reduced-precision, including master weights and automatic loss scaling. After, we discuss real-world training in mixed precision with a particular focus on the PyTorch and TensorFlow frameworks.
IBM Developer Day (Includes 6 sessions)
11:00 - General Welcome (Host: Dilek SezgÃ¼n)
11.15 - #AI4Good on PowerAI: How can your coding skills help others? (Hackathon) Speaker: Carmen Recio
12:00 - Watson-Studio: Putting AI to Work for Business: Umit Mert Cakmak
13:00 - Virtualization is real: Cloud-enabled CAD (Cloud session) Speaker:Alex Hudak, Charlie Dawson (IMSCAD)
14:00 - Integrated IBM AI session (Systems & Cloud) Speaker: Florin Manaila, Alex Hudak
15:00 - Fuel Pipeline for AI (Storage Session),Daniel Reiberg
16:00 - Accelerate and Scale High Performance Computing with IBM Cloud and Rescale, Speaker: Jerry Gutierrez, IBM & Joris Poort, Rescale
At Pure Storage, we have helped deploy a number of AI systems including one of the world's fastest supercomputers dedicated to AI and many of the world's most recognized brands in autonomous cars. During our session, Lee Razo, a data technology specialist at Pure Storage will be sharing what we've learned in working with these customers.
This session examines the full pipeline and infrastructure needed for production Deep Learning, including ingest, data processing and preparation, and storage. We first present a reference data pipeline for AI and explain the role of each stage: ingest, inference, ETL, experimentation, and training. The needs for each stage dictate infrastructure decisions: shared storage, mixed compute servers, and a single unified network.
Second, we present a set of benchmarks that run multi-node training on Imagenet that include realistic IO from persistent storage to the CPU and GPU. For this training, we utilize four DGX-1 servers, a single FlashBlade shared storage tier, and an ethernet-only network. Our key result is that we achieve linear scalability of training performance up to 32 GPUs by utilizing FlashBlade, RDMA over Ethernet, and Horovod for multi-GPU scheduling.
Edge computing and IoT use cases have inspired machine inference based software stacks on the end and deep learning stacks in the cloud. However, challenges remain. The edge software stack and underlying systems configuration tends to be simplistic against a growing need to enable a richer set of edge applications in a multi-tenant format. The edge and cloud ends are typically developed in a disjoint manner in terms of services offered to the end user. And the workload provisioning frameworks at the edge and in the cloud make it such that ML hardware capabilities are leveraged inefficiently for the purposes of the application. This talk will describe a, edge-to-cloud application runtime, and a "ML hypervisor" to efficiently create applications that span the edge and the cloud and to efficiently map their needs to the underlying hardware's full capabilities.
In this session, we will explore the latest work, showcase benchmarks, and provide demos of the GPU Open Analytics Initiative (GoAi), a collection of open-source libraries, frameworks, and APIs established to standardize GPU analytics to allow for easier development and enhanced performance for GPU-accelerated analytics technologies. Numerous Fortune 500 customers experience latency and performance issues in their data pipeline. Big data frameworks and solutions tried to address this problem, but the cost to scale to the volume and velocity of current needs has proven to be prohibitively expensive. GoAi is addressing these challenges with a vision is to create an end-to-end GPU-accelerated data pipeline that will smooth onboarding ramp for enterprises to explore and integrate AI into their core data driven decision making processes. The session will also provide examples of how customers are benefiting from early primitives and outperforming CPU equivalents.
Since 2009, our deep learning artificial neural networks have won numerous contests in pattern recognition and machine learning. Today, they are used billions of times per day by the world's most valuable public companies. True AI, however goes far beyond slavishly imitating teachers through deep learning. That's why we have also focused, since 1990, on unsupervised AIs that invent their own goals and experiments to figure out how the world works and what can be done in it. Many of them model the world through a recurrent neural network that learns to predict the consequences of their action sequences. Without a teacher, they derive rewards from continually creating and solving their own, new, previously unsolvable problems, a bit like playing kids do, to become more and more like general problem solvers in the process. Relevant buzzwords include "artificial curiosity" (since 1990) and PowerPlay (since 2011). I will also briefly outline how AIs that set their own goals will eventually colonise the entire universe and make it intelligent.
Roborace Chief Strategy Officer Bryn Balcombe and the Technical University of Munich discuss their collaboration on a pilot project which has given TUM the opportunity to use the Roborace platform to test and develop their autonomous software. TUM have had access to Roborace's simulated and real environments, in order to help progress their self driving algorithms with the goal of being able to run one of Roborace's DevBot vehicles at the Formula E event in Berlin. The talk will discuss how the collaboration works, the successes and learnings, and how the platform will be available for more organisations to use in the future.
According to a major car manufacturer, modern vehicles are collecting and sharing more than 25 gigabytes of data per hour, from dozens of sensors focused inside and outside the car. Compound that rate of collection across the growing fleets of connected vehicles, and the automotive industry is facing a stiff new challenge: making hundreds of billions of location-intelligent data points comprehensible, actionable, and predictive. GPUs running OmniSci's extreme analytics platform are uniquely capable of solving this problem, with orders-of-magnitude faster SQL queries, and full-fidelity rendering on the GPU. In this talk, Aaron Williams will use a real-world example to share best practices for analyzing a large dataset of driving behavior, to lower risk and cultivate better drivers.
Continental is partnering with NVIDIA to provide a complementary solution for the Automotive AV market. We are jointly working on some of the mayor OEM opportunities. The session will provide an overview on ContiÂ´s AV products with infused NVIDIA performance and will give an outlook on future SD challenges to tackle â such as for example simulation and validation.
Autonomous Vehicles are transforming automotive industry and how humans interact with these intelligent machines. While we transition from traditional driving to autonomous, in-cabin monitoring becomes key to ensure passengers' safety, attention and comfort.
The opportunities in this area are huge, ranging from HMI/UX using NLP/Lip reading to co-pilot, eye tracking and monitoring of emotional reactions of the passengers using vision, voice, bio-feedback, sentiment analysis, etc.
This panel will explore the current challenges and opportunities of in-cabin monitoring and the role of AI in this area.
The rise of AI has the effect as big as the rise of computers. Autonomous driving is a super-computing problem and latest advances in AI and GPU based computing has enabled the development and acceleration in the self-driving vehicle space where NVIDIA is fundamentally changing and shaping the future. Furthermore latest advances of the machine learning and neural network based algorithms as well as sensor technology will shape the design of the future car and the evolution of transportation. Artificial intelligence algorithms, intelligent maps and simulation are the main components of the solution to the complex environments and dynamic driving conditions of the autonomous driving problem. There is no way for engineers to hard-code and test every possible variable or situation a car may face in a daily drive and test it on the road. This talk will provide technical insights to the NVIDIA Drive Platform for autonomous vehicle development and its applications in self-driving car space.
This panel will explore the current challenges and developments of AI tied specifically to automotive â how is AI being used to develop autonomous vehicles to revolutionize the transportation and mobility.
This session will explore challenges and innovative approaches to simulation techniques for the development and validation of autonomous vehicles. Panelists will present and discuss opportunities to realistically simulate sensor properties and data, to generate content and how to integrate with NVIDIA's DRIVE ecosystem in a closed-loop environment to combine real driving scenarios with simulation. \n
This session will discuss how Deep Learning is applied to improve real-time video data analysis for autonomous vehicles, in particular, semantic segmentation. The results of two pilot projects that tested both autonomous and connected drive will be presented as well as the intelligent connected infrastructure required for full Autonomous Driving.
Autonomous driving space is getting crowded these days. With over 50 companies doing public road tests in California and other parts of the world, getting one more self-driving car out there is not overly exciting any longer. Deployments - and by deployments we mean launch of a service powered by self-driving cars - is entirely different story.
During our session we will study the case of launching a public transportation service in a small Russian city powered by Yandex Taxi self-driving car. We're planning to discuss technical aspects of launching the service and specific engineering challenges we faced. We will also update the audience on the state of Yandex autonomous driving technology enabling the service and on our plans for the ongoing development.
Porsche's view on ADAS, with a focus on the specific profile that Porsche sees for sports cars. Porsche's experience with advanced systems like Innodrive as well as topics like the demands on V&V and data driven development.
While we are still at the nascent phase of the Intelligent Industrial Revolution, one industry already started its mutation â transportation. From heavy trucks to shuttles, everything that moves will one day become autonomous. Companies around the world are exploring AI-led technologies to enable new applications and business models.
This panel will explore the current challenges and opportunities of AI tied specifically to Commercial Vehicles."
This talk will be an opportunity to explain what challenges exist with introducing autonomous vehicles in Europe, how we are addressing them, what the progress in each field looks like, how we think services will be launched, and what remains to be done (which is a lot). We will include examples of how we are using NVIDIA technology to help us work on problems, show how we've configured our hardware, and discuss how we expect that to evolve from one platform to the next. Tech companies are set to deliver safe AVs ahead of OEMs and tier vendors, but no one has a safe solution for our complex cities or European cities that are more complex than either the US or China. Five AI is Europe's fastest-growing tech company, with just under 100 people from 15 a year ago, created to deliver self-driving technology to Europe's city dwellers.
In this talk, we'll discuss Project MagLev, NVIDIA's internal end-to-end AI platform that enables the development of NVIDIA's self-driving car software, DRIVE. We'll explore the platform that supports continuous data ingest from multiple cars (each producing TBs of data per hour) and enables autonomous AI designers to iterate training new neural network designs across thousands of GPU systems and validate their behavior over multi PB-scale data sets. We will talk about our overall architecture, from data center deployment to AI pipeline automation, large-scale AI dataset management, AI training & testing.
Digital Homologation with Simulation for R&D and validation using Nvidia GPU in cloud and HIL. Securing and homologating automated driving functions presents a huge challenge for market introduction due to an enormous number of scenarios and environment parameter combinations. Confronting conventional real world tests with the new challenges of automated driving is not feasible anymore, and yields to a virtualisation of the testing methods by means of X-in-the-Loop approaches. Especially when using Deep Learning Algorithms for automated driving functions, a scalable, powerful and consistent toolchain is required. Together with TÃV SÃD and the University of Applied Sciences in Kempten, AVL is working on such a consistent toolchain. Different configurations for the development and also approaches for the homologation shall be introduced. Especially the challenge of close-loop-testing, including the vehicle and powertrain dynamics, should be addressed. The challenge is to provide a modular framework integrating existing tools and platforms like NVIDIA's DRIVE platform or NVIDIA's DRIVE SIM to increase the efficiency during development and homologation.
Advanced driver assistance systems (ADAS/autonomous driving) are becoming part of all vehicles. All major OEM and Tier-1 auto manufacturers are implementing and testing AD facilities. We examine how real-time sensors, big data computing, data storage and data archiving are integrated in today's ADAS/AD systems, providing a fascinating case study, best practices for workflow design, testing and development, data storage and archiving, applicable to all industries.
Clearly, Autonomous Driving has the unique potential to change the way we think about transportation. The rapid evolution of sensors, artificial intelligence and IT-infrastructure paves the way to a driverless future much faster than many think. Let's have a look at what is out on the streets today and how we approach the fascinating future of Autonomous Driving at Mercedes-Benz. Title: The Reinvention of the Car - How to Consider Safety Aspects for Launching Autonomous Driving Abstract: Clearly, Autonomous Driving has the unique potential to change the way we think about transportation. The rapid evolution of sensors, artificial intelligence and IT-infrastructure paves the way to a driverless future much faster than many think. Let's have a look at what is out on the streets today and how we approach the fascinating future of Autonomous Driving at Mercedes-Benz Starting with short Video CASE @ Daimler (Connected, Autonomous, Shared, Electric) 1. Innovations in series cars: ADAS Update -History of ADAS Systems -Current Level of Automation (Driver Assistance Systems) -Field Validation 2. Why full vehicle automation makes sense: Motives -Reasons for Vehicle Automation 3. Where are we heading to: Technology -Sensor Setup -Use Case -How to understand Sensor Data -How to collect the needed information -Short Demo Drive (Video) -Experience counts (field testing around the world) -Safe System Architecture -ISO 26262 Development for AD -Strong Partners -Outlook
As developers of deep learning based applications for extreme environments ranging from Robotic Exploration on Mars through to remote inspection of critical assets we face a host of technology challenges and demands. NVIDIA's portfolio of GPU and embedded platforms allow us to both develop new analytic functions and offer orders of magnitude better performance year on year. Critically we are able to migrate our cloud based analysis to the edge using Jetson. This talk looks at some of our current work for both Mars exploration and commercial inspection services using AI at the edge.
Designing an autonomous machine are about much more than just the AI. Electrical, Mechanical, Connectivity, and Security are just a few of the disciplines where you will require expertise. Not all companies will have complete expertise in all these areas. In this session, we will provide examples followed by design considerations, strategies and solutions to begin to address these challenges.
Find out how financial companies analyse data using NVIDIA GPU-acceleration, impacting real-time risk management, regulatory reporting, fraud detection and cybersecurity, anti-money laundering, and trader surveillance.
We will discuss real-world examples, including how a specific multinational bank uses a real-time risk management engine running on GPU cloud instances. The bank's analysts can now make time-sensitive, computation-intensive risk calculations involving hundreds of variables, using a real-time, interactive dashboard. This produces meaningful, timely, and consistent financial analysis, optimised to maximise profitability and power business in motion.
This approach lends itself to a data-powered business. It allows banks to move applications â such as counterparty risk analysis â from batch overnight processing to streaming and real-time, creating flexible real-time monitoring of extreme data that makes it easy for traders, auditors, and management to take action.
This session will explore how auditors can be misguided or "fooled" by adversarial accounting records or adversarial financial transactions.
Recent discoveries in deep learning research revealed that learned models are vulnerable to "adversarial examples," or a sample of slightly modified input data that intends to cause a human and/or machine to misclassify it. Such examples exhibit the potential to be dangerous, since they could be specifically designed to misguide auditors or an accountant. Securing accounting information systems against such "attacks" can be difficult. In this talk, we'll explain why such "adversarial examples" are of vital relevance in the context of fraud detection and financial statement audits. We will demonstrate how autoencoder neural networks can be trained in an adversarial setup to generate "fake" accounting records or financial transactions. Such financial transactions might be misused to "attack" an organization's internal control system or obfuscate fraudulent activities. The training of such examples was conducted by training several adversarial autoencoders using NVIDIA's DGX-1 system.
Financial transaction data are one of the most privacy sensitive datasets out there, and yet their digital trail of data points result in highly unique fingerprints, making information-retaining anonymization thereof a hard problem. But one, that turns out to be solvable thanks to advances in AI.
For one of the largest retail banks operating in Central East Europe we deployed deep generative models, that train on over half a billion of transactions, in order to then generate highly realistic & representative synthetic customers, matching the patterns of the actual customers. By adapting auto-regressive neural networks to a highly heterogeneous data structure, we learn and thus retain detail, structure as well as variation of your privacy-sensitive data at an unprecedented level, while rendering the re-identification of any individual impossible. As this data can then be utilized without putting the privacy of customers at risk, this enables open collaboration for the development and testing of their digital services.
This type of modeling at that scale would not have been feasible without the advances in generative modeling as well as GPU technology
This session will be about deep learning frameworks that we have developed for interactive character control. The first approach is called a Phase-Functioned Neural Network (PFNN). The entire network is trained in an end-to-end fashion on a large dataset composed of locomotion, such as walking, running, jumping, and climbing movements, fitted into virtual environments. Our system can therefore automatically produce motions where the character adapts to different geometric environments such as walking and running over rough terrain, climbing over large rocks, jumping over obstacles, and crouching under low ceilings. Once trained, our system is also extremely fast and compact, requiring only milliseconds of execution time and a few megabytes of memory, even when trained on gigabytes of motion data.
The second approach is called Mode-Adaptive Neural Networks. This is an extension of the PFNN and has the capability to control quadruped characters, where the locomotion is multimodal. Once trained, the quadruped produce different gait types such as walk, pace, trot, and canter by simply changing the velocity of the character. The frameworks are suitable for computer games and VR.
Raytracing is the hallmark feature of NVIDIA's just-released Turing GPU in the GeForce RTX and Quadro RTX series. RTX for the first time brings raytracing to a mass consumer audience â a feat that just a few months ago was considered years in the future. This talk by one of the lead engineers on the project will walk you through some core principles of raytracing, give an overview of the RTX technology, and explain how it applies to real-time applications such as games
Remedy Entertainment is one of the first game studios that have started working with NVIDIA RTX technology. The session describes how RTX is used today in the company's in-house game engine, Northlight. Implementations of shadows, reflections, ambient occlusion and indirect diffuse illumination effects with RTX are explained. Additionally, the session talks about general best practices about how to most efficiently use RTX through Microsoft DXR API in a game engine.
Oxford Nanopore has built the first and only real-time, portable DNA sequencer - the MinION. It is being used to bring DNA information to researchers in many sectors, including biomedical/cancer research, environmental monitoring, agriculture, food/ water testing, and education. Oxford Nanopore is using GPUs to make sure that genomic data can be processed in real time, delivering potential benefits of rapid insights to users in any environment. Leila Luheshi and Rosemary Dokos will talk about current and potential healthcare applications of Nanopore technology, and how GPUs will turn sequence data into rapid insights for disease or environmental management.
When talking about GPU benchmarking and making the right choice of GPU for your organisation or workload, you'll probably think of using fancy benchmark tools. But there is a much more fun and low-cost way to do so. This session is about how to benchmark and make a GPU choice with the game Skyrim. It's a very low level way of testing using the NVIDIA-smi on the hypervisor and Powershell in the Windows VDI. Make your own charts with the information you collect yourself. This session is based on Windows 10 VDI, and VMware vSphere 6.7.
This session will involve a live case study of how the Willmott Dixon and ebb3 partnership are leveraging NVIDIA Quadro Virtual Data Center Workstation (Quadro vDWS) enabled services to support its internal teams and customers as the AEC sector in the UK moves rapidly towards digital transformation. We will explain one of the current problems that exist with BIM and the journey towards Level 3 in the UK, the barriers to users working with federated BIM models, and the ever increasing use of data within an analogue industry.The session will cover a live project visualisation and examples of using it within a collaborative environment to reduce time and costs.
With the latest release of NVIDIA vGPU software the world's most powerful virtual workstation gets even more powerful. Learn more about how our latest enhancements enable your data center to be more agile and scale your data center to meet the needs of thousands to ten-thousands and even hundreds of thousands of users. The newest release of NVIDIA virtual GPU software adds support for more powerful VMs, which can be managed from the cloud or from the on premises data center, or private cloud. With support for live migration of GPU-enabled VMs, IT can truly deliver high availability and a quality user experience. IT can further ensure they get the most out of their investments with the ability to re-purpose the same infrastructure that runs VDI during the day to run HPC and other compute workloads at night. In this session, we will unveil the new features of NVIDIA vGPU solutions and demonstrate how GPU virtualization enables you to easily support the most demanding users and scale virtualized, digital workspaces on an agile and flexible infrastructure, from the cloud and as well as the on premises data center.
During this session, attendees will learn about sizing and optimising a Citrix Xenapp / Microsoft RDSH environment with NVIDIA virtual GPU (vGPU) technology. Learn why it is especially important in XenApp environments to keep an eye on the framebuffer and make sure the load is not too high. Get the best performance out of your environment by using the best architecture and sizing on the basics of a real world scenario with Grimme, the world market leader for potato farming technology. In 2016, Grimme launched a project to replace their existing Citrix XenApp Infrastructure and decided to use NVIDIA Tesla cards and NVIDIA vGPU software in the new environment to improve performance and user experience.
The Hartree Centre, a department of the UK National Labs, focusses on industry-led challenges in HPC, High Performance Data Analytics, and AI. Its mission is to make UK industry more competitive through the uptake of novel technologies. Historically the focus has been on HPC (simulation and modelling), and more recently on data centric computing. This sessions focuses on on how AI can best be applied to add value for industry partners.
In pushing the limits of throughput of floating-point operations, GPUs have become a unique technology. During this session, we'll explore the current state of affairs from an application perspective. For this, we'll consider different computational science areas including fundamental research on matter, materials science, and brain research. Focusing on key application performance characteristics, we review current architectural and technology trends to derive an outlook towards future GPU-accelerated architectures.
We present in this talk a portable matrix assembly strategy used in solving PDEs, suited for co-execution on both the CPUs and accelerators. In addition, a dynamic load balancing strategy is considered to balance the workload among the different CPUs and GPUs available on the cluster. Numerical methods for solving partial differential equations (PDEs) involve two main steps: the assembly of an algebraic system of the form Ax=b and the solution of it with direct or iterative solvers. The assembly step consists of a loop over elements, faces and nodes in the case of the finite element, finite volume, and finite difference methods, respectively. It is computationally intensive and does not involve communication. It is therefore well-suited for accelerators.
This talk provides an overview of the key strategies used to design and implement OpenStaPLE, an application for Lattice QCD (LQCD) Monte Carlo simulations. LQCD are an example of HPC grand challenge applications, where the accuracy of results strongly depends on available computing resources. OpenStaPLE has been developed on top of MPI and OpenACC frameworks. It manages the parallelism across multiple computing nodes and devices, while OpenACC exploits the high level parallelism available on modern processors and accelerators, enabling a good level of portability across different architectures. After an initial overview, we also present performance and portability results on different architectures, highlighting key improvements of hardware and software key that may lead this class of applications to exhibit better performances.
We present our experiences implementing GPU acceleration in the massively parallel, real space FHI-aims electronic structure code for computational materials science. For fourteen years, FHI-aims has focused on high numerical accuracy for current methods, such as Kohn-Sham density-functional theory and beyond, and on outstanding scaling on distributed-parallel high-performance computers. We show how to exploit vectorized implementations in FHI-aims to achieve an overall 3x-4x GPU acceleration with minimal code rewrite for complete simulations. Furthermore, FHI-aims' domain decomposition scheme on non-uniform grids enables compute and memory-parallel computing across thousands of GPU-containing nodes for real-space operations.
This talk will present the roadmap, the strategy and the currently ongoing efforts to port the fundamental building blocks of the QuantumESPRESSO suite of codes to accelerated architectures. QuantumESPRESSO is an integrated suite of codes providing computational methods to estimate a vast number of physical properties at the nanoscale. It features high modularity and a user-oriented design, and it can efficiently exploit standalone workstations as well as state-of-art HPC systems. The differences characterizing this new work and the original GPU porting done in CUDA C back in 2012 will be used to discuss aspects of code evolution and maintainability. Special attention will also be devoted to the performance-critical kernels shared by most of the components of the suite.
VASP is a software package for atomic-scale materials modeling. It's one of the most widely used codes for electronic-structure calculations and first-principles molecular dynamics. We'll give an overview on the status of porting VASP to GPUs with OpenACC. Parts of VASP were previously ported to CUDA C with good speed-ups on GPUs, but also with an increase in the maintenance workload, because VASP is otherwise written wholly in Fortran. We'll discuss OpenACC performance relative to CUDA, the impact of OpenACC on VASP code maintenance, and challenges encountered in the port related to management of aggregate data structures. Finally, we'll discuss possible future solutions for data management that would simplify both new development and the maintenance of VASP and similar large production applications on GPUs.
Today we are investigating different technologies and architectures, and we will present the first hardware and software prototype that will evolve into a system able to overcome an unprecedented challenge.
To probe the predictions of the Standard Model of Particle Physics, the Large Hadron Collider at CERN will be upgraded by 2026 to produce 6 billion proton collisions every second at the centre of the Compact Muon Solenoid (CMS) detector. These collisions produce events in which new particles, which did not exist before the collision, are generated.
The CMS experiment will be able to observe and record the most energetic and rare of these events.
Observing the details of all these events requires reading and analyzing almost 100TB of data every second... and CMS is working on a hybrid approach to tackle this challenge: ASICs and FPGAs will be used for the first level of data reduction, while a hybrid cluster of computer servers and GPUs will be used for the full event reconstruction and final online selection.
In 2014, GENCI set up a French technologyï»¿ watch group that targets the provisioning of test systems, selected as part of the prospective approach among partners from GENCI. This was done in order to prepare scientific communities and users of GENCI's computing resources for the arrival of the next "Exascale" technologies.\nThe talk will present results obtained on the OpenPOWER platform bought by GENCI and open to the scientific community. We will present on the first results obtained for a set of scientific applications using the available environments (CUDA,OpenACC,OpenMP,â¦), along with results obtained for AI applications using IBM's software distribution PowerAI.
Legacy, performance hungry and cutting edge deep learning workloads require best of breed cloud services and hardware. Enterprises require low cost and financial flexibility. Learn how Oracle and NVIDIA have partnered together to solve these challenges with a bare-metal NVIDIA Tesla GPU offering to squeeze every ounce of performance at a fraction of the cost. We'll also detail the ability to use NVIDIA GPU CLOUD to streamline the experience for customers to launch and run clusters of GPU Virtual Machines or bare metal instances for AI or HPC workloads. Come see live demos and learn what Oracle Cloud Infrastructure is doing in this space!
Microsoft Azure's N-Series VMs powered by latest NVIDIA GPUs enable a range of new accelerated scenarios. Learn how you can take advantage of GPUs in Azure - from Workstation Graphics and Visualization, to HPC simulation, to training models for artificial intelligence. This session will delve deep into today's exciting offerings with live examples and offer a view of what's to come in the future.
Learn how to develop an Artificial Intelligence system to localize and recognize food on trays to generate a purchase ticket in a check out process.
(1) Solving a real business problem using Deep Learning advanced technology based on object detection and localization.
(2) Combining a pipeline of models to improve accuracy, precision and with reasonable recall levels.
(3) Discovering how to develop and train a model in the cloud to be used embedded in an NVIDIA Jetson TX1 device.
Detecting road users in real-time is key to enabling safe autonomous driving applications in crowded urban environments. The talk presents a distributed sensor infrastructure being deployed in the city of Modena (Italy) at the heart of the Italian 'Motor Valley'. Modena's Automotive Smart Area (MASA) connects hundreds of smart cameras, supporting embedded GPU modules for edge-side real-time detection, with higher performance GPU (fog) nodes at block level and low latency wireless V2X communication. A distributed deep learning paradigm balances precision and response time to give autonomous vehicles the required sensing support in a densely populated urban environment. The infrastructure will exploit a novel software architecture to help programmers and big data practitioners combine data-in-motion and data-at-rest analysis while providing Real-Time guarantees. MASA; funded under the European project CLASS, is an open testbench where interested partners may deploy and test next-generation AD applications in a tightly connected setting.
Modern computing hardware and NVIDIA Jetson TX1 / TX2 performance create new possibilities for smart city applications and retail, parking lot, and drone industries. We'll present on how the PIXEVIA system covers vision processing and AI tasks using deep neural networks; learning using computer generated images for number plate recognition; and self-supervised learning for vehicle detection. We will explore methods for orchestrating and combining information from different type of neural networks (from SSDs, Mask-RCNNs to attention based RNNs). Real-world use cases for parking lots (empty parking space detection, number plate recognition) and retail industries (amount of stock on the shelf calculation, people counting with age and gender recognition) will also be presented.
Experience how to make spaces aware of the situation of people and objects. Explore new techniques to build real-time systems that can understand scenes with the help of hemispherical point clouds and AI at the edge. The goal of this session is to learn new ways of developing scene understanding needed for action and interaction in public spaces or smart homes. The capture, recognition and understanding of all external and internal degrees of freedom of persons and objects and of their respective states give the full information of the observed space.
While hemispherical vision provides advantages for wide-area coverage from a single point of observation, it also introduces new challenges due to its distinct projection geometry. At the example of 3-dimensional people detection and posture recognition, we explain different approaches to use deep neural networks to extract information from hemispherical RGB-D data. The talk focuses on providing an overview over methods, which attendees can be apply to custom projects and run on Jetson in real-time.
Recent developments in artificial intelligence, advances in GPU computing hardware and the availability of large scale medical imaging datasets allows us to learn how the human brain truly looks like from a biological, physiological, anatomical and pathological point-of-view. This learning process can be augmented by Electronic Healthcare Record data, cognitive examinations, and diagnostic/radiological report data, thus providing an integrated view of the human interpretation of neurological diseases. This talk will present how AI models can learn from big and unstructured neurological and neuroradiological data and be used as tools for precision medicine, with the aim of translating advanced imaging technologies and biomarkers to clinical practice, streamline the clinical workflow and improve the quality-of-care. It will also explore the technological translational process, requiring full clinical support, deep algorithmic integration into the radiological workflow, and the deployment of a high-throughput hospital-integrated GPU computational platform
DLTK is an open-source toolkit providing baseline implementations for efficient experimentation with deep learning methods on biomedical images. DLTK builds on top of TensorFlow, and its high modularity and easy-to-use examples allow for a low-threshold access to state-of-the-art implementations for typical medical imaging problems. Automatic downloading and pre-processing of example datasets allow for running and testing example applications, including medical image segmentation, regression, classification, representation learning, super-resolution and training generative models on biomedical images.
A comparison of DLTK's reference implementations of popular network architectures for image segmentation demonstrates new top performance on the publicly available challenge data "Multi-Atlas Labeling Beyond the Cranial Vault". Additionally, DLTK contains a medical model zoo with downloadable pre-trained models for medical image analysis problems, enabling transfer-learning and direct deployment of evaluated deep learning methods.
We believe that medicine will be more precise and affordable. Physicians will integrate relevant patient data and insights at the point of decision for precise diagnostics. Therapy will be tailored to the characteristics of both the patient and disease ? resulting in the right treatment for the right patient at the right time. AI-powered decision support could help to balance the need for personalization when it matters and standardization to reduce unwarranted variations.
Radiographic image data uniquely represent the spatio-temporal course of disease progression in patients. This talk will present current methods in medical image computing that aim to systematically extract, measure and utilize this information to optimize diagnosis and treatment. The focus will be on suitable machine learning methods, which have become a key technology for successful image processing. Special challenges in the medical context are limitations in terms of data availability, inherent data noise and the lack of high quality annotated training data. The lecture will discuss current ideas and approaches to solving these problems, with relevance potentially not only in medicine but also in other applications.
The latest operating systems like Windows 10 or Server 2016 include graphically rich features in its user interface. In addition to this, contemporary applications built to run on top of these modern operating systems further contribute to GPU consumption. Therefore it is important to have a GPU for your traditional VDI or any cloud enabled desktop deployment to satisfy the number one user demand, which is the best in class user experience. But,
- How can we measure user experience?
- How can we find out the perfect protocol?
- What is that perfect combination of remote protocol policies?
- What is the perfect codec that best fits to your use case?
Don't Feed These Animals is Nebula Studios first short animation film. It's a fully independent project that started from one single character by Jose Alves da Silva. A bespoke international team was built and we were lucky enough to have global supporters joining us in what has become quite a ride. We'll share the full behind the scenes & set up story since DFTA's inception, including how Nvidia was essential to our decision of shifting from CPU to GPU render.
Ergonomics is an important aspect engineering of manufacturing systems andmaintenance procedures of new products for both physical and virtual envronments . A specific case is the prediction of operator or technician visibility under real world lighting conditions. For example, when simulating how efficiently a worker will be able to operate a planned system, it is necessary to evaluate early on how this would be affected by a given illumination situation. The ability to digitally design for manufacturability or serviceability, while taking into account human factors, directly influences the potential profitability of new products. Â In this session, we will talk about how NVIDIA's OptiX and RTX ray tracing technologies can be leveraged to simulate the propagation of light in environments with complex geometric topology on GPUs. A special focus will be on the OptiX AI denoiser, which masks Monte-Carlo noise that is stemming from the underlying numerical integration methods. We will show how this has been integrated into ESI's Helios visualization framework, before we demonstrate the system through practical examples
We'll discuss the basics of NVIDIA's MDL, showing how a single material can be used to define matching appearances between different renderers and rendering techniques. End users will learn how physically based definitions can be defined, while developers will learn what's entailed in supporting MDL within their own products or renderers.
Arnold is a high quality production renderer for visual effects in film and feature animation used by more than 300 studios worldwide on projects such as Blade Runner 2049 and Game of Thrones. Arnold was instrumental in the shift toward physically-based light transport simulation in production rendering; in fact, this role was recognized with an Academy Award in 2017. Arnold's success is rooted in its ability to efficiently produce artifact-free images of dynamic scenes with massive complexity while simplifying the user's workflow.
Autodesk will be demonstrating GPU acceleration inside Arnold using NVIDIA OptiX.
We'll give an introduction to mesh shaders, a new programmable geometric shading pipeline first exposed in the Turing architecture. These new shaders bring the compute programming model to the graphics pipeline as threads are used cooperatively to generate compact meshes (meshlets) directly on the chip for consumption by the rasterizer. Applications and games dealing with high-geometric complexity benefit from the flexibility of the two-stage approach, which allows efficient culling, level-of-detail techniques as well as procedural generation.
The 'Speed of Light' is a fully dynamic ray-traced automotive showcase developed by Epic Games and Nvidia in collaboration with Porsche, running on Unreal Engine 4 and Nvidia's Quadro RTX. This in-depth presentation is split into two separate sessions. Join us for Part 2 as Epic wraps-up their deep-dive by covering shading, rendering and performance optimizations in a real-time ray-traced environment using production examples from the Speed of Light. Make sure you register for the first session E8518!
The presentation is an introduction to NVIDIA OptiX - a sophisticated library for performing GPU ray tracing. You'll get an overview of the NVIDIA OptiX ray-tracing pipeline and learn about its programmable components for high-performance ray tracing on the GPU. NVIDIA OptiX is used in many domains, from rendering to acoustic modeling to scientific visualization. We'll review its features and present code samples to demonstrate best practices for writing a high-performance ray tracer using its programming model.
Globally, chronic lung diseases are the fourth leading cause of death. Clinical parameters for chronic obstructive pulmonary disease (COPD) are of paramount importance in determining clinical parameters for the identification of biomarkers for treatment and therapy monitoring. Present a robust Deep Learning pipeline that allows for the prediction of voxel-wise labels of lung lobes in 4D MRI perfusion maps, trained end-to-end without manual interaction. Training can either be accomplished in less than a week on a consumer-scale dual socket workstation with two attached Tesla P40 accelerators or within a day on a DGX-1 Volta. The resulting segmentation maps and derived clinical parameters exhibit high concordance with the ground truth.
Learn how our world can be understood better and faster by our robotic companions thanks to embedded GPUs. In this talk, we will present developments done since last year thanks to the Jetson TX2 embedded in SoftBank's Pepper, the world's leading affordable humanoid robot. This work revolves around human gesture recognition and better human robot interaction. A live demo of the autonomous Pepper Robot embedding the Jetson TX2 and interacting on stage will be done during the session.
This talk shows how Deep Metric Learning is used to reduce the data hunger of Deep Learning models for object recognition and gives them the ability to learn on-the-fly. Deep Learning for automated object recognition has shown to be highly effective and useful for various applications. However, practitioners need a large amount of labelled data to train robust models. On top, the time ocnsuming training process has to be repeated every time an object class is added to the portfolio. This talk proposes applying Deep Metric learning to overcome these two problems. Instead of learning to recognize a particular set of objects. We train the model to learn similarities between them. This fine task modification results in models which are able to compare known objects with unknown ones. Adding new objects to the portfolio only requires adding a few addiotonal images to the database. No retraining is required. As a use case, we demonstrate an automated stock monitoring system which recognises the products going in and out of a store. The data collected can be used for various applications including stocktaking, computing statistics and worker guidance.
The talk presents the current developments of the NUMBERS project highly interdisciplinary research program in Cognitive Developmental Robotics to construct a novel artificial cognitive model of mathematical cognition that imitates human-like learning approaches for developing number understanding.
The project aims to provide a proof-of-concept and the scientific and technological bases for novel robots capable of abstract and symbolic processing, which is required for improving their cognitive performance and their social interaction with human beings.
During the talk, the current experimental results will be reviewed to give evidence of improved performance thanks to the embodiment.
True holography, or "dynamically sculpted light", allows for the reproduction of a full 3D light field in a display, complete with colour and depth. Hence, it is perhaps the ultimate goal for the display and VR/AR industries. With the advent of powerful GPU computing, it is now for the first time becoming possible to compute holograms - in this full 3D sense- in real time. This is what we are working on at VividQ.
Audience members will be given a brief introduction to how 3D holograms work, using diffractive optics, and why they are so expensive to compute. We will cover the standard approach to computation, and our progress in speeding up the computation using CUDA, taking the calculation time of a holographic frame from many minutes down to milliseconds. Since a full light field is created, instead of using stereographic techniques, the visual conflicts which contribute to headaches in many VR and AR devices are avoided. There will also be the opportunity to see videos of AR holograms, demonstrating how the images focus and defocus at different focal depths, and exhibit expected optical effects such as parallax.
This talk will feature an update on what's happening in the professional VR space at NVIDIA. We first introduce OpenGL and Vulkan VR functionality, and then will talk about how to drive dual-input HMDs from two GPUs efficiently.
Holodeck is NVIDIA's advanced technology platform for Virtual Reality. We will begin this session by reviewing the motivation and goals of Holodeck. We will then highlight the new features from our most recent Holodeck release. The latest version adds support for architectural design review, including model review at scale, teleporting to different floors, and specifying points of interest. As a feature of this session, Hilda Espinal, CTO of CannonDesign, will join us to share how her firm is using Holodeck in customer projects, how customers are reacting to this new technology, and her perspective on how VR will impact architectural design in coming years. CannonDesign is a leading global design firm that was one of the earliest adopters of Holodeck. Come hear this talk to learn all about the technology, and then visit the VR Village to experience the new Holodeck features firsthand!
This talk examines the convergence of different technologies like AR, VR, AI, 3D printing and more to solve problems in the arts space and lower the entry of barrier for aspiring artists. Rick Treweek, Director of Technology, and Gareth Steele, Art Director at The Digital Foundry, will share their knowledge of how these technologies can be used to promote technology across varied audiences, particularly in the African ecosystem. The merging of technology is opening up a new toolset to utilise and expand art practices and allows creation without extensive technical knowledge or experience.
This knowledge is not only applicable in art, but is valuable for traditional and emerging tech companies as well: Custom hardware creation thanks to 3D printing allows us to experiment with alternative delivery mechanisms for Virtual and Augmented Reality, and to rapidly prototype new tools or components. Coupled with the powerful ability to design and visualize within VR for 3D print, we are exploring new ways of bridging the gap between digital & physical and also create digital twins by using technologies like 3D scans and photogrammetry, that are all profiting from GPU acceleration.
Vive X is HTC's global program to build the XR ecosystem by investing in startups. With over 80 portfolio companies and local hubs in the US and Asia, the program is now active in Europe, where over 150 startups have been looked at in the last six months. This talk will share the insights gained from this activity as well as discussing the latest investment trends in the sector generally. Who is investing in AR and VR and what types of ventures are most successful in getting funded? Where do the biggest opportunities lie and what are some key challenges that still need to be addressed? This talk is for you whether you're a startup founder, investor or just interested in the latest innovations happening in XR.
The session will explain how ultra-low latency techniques for video encoding and transmission can be applied for creating advanced XR experiences anywhere in the world, just using a mobile phone connected to a 5G network. The main challenge for creating a proper XR experience through telecommunications networks is to achieve a total latency (" motion to photon") low enough for avoiding desynchronizations between the brain expectations and the signals that the eye is receiving. The use of NVIDIA NVENC plus other protocols like HTTP Chunked Transfer Encoding make possible to achieve a very low latency over a universal protocol like HTTP that is easily transmitted across mobile networks, making possible XR experiences without a physical connection between render engine and the terminal.