VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server High-Performance Visual Computing in the Data Center NVIDIA RTX Server Do your life’s work from anywhere
Creators, designers, data scientists, engineers, government workers, and students around the world are working or learning remotely wherever possible. They still need the powerful performance they relied on in the office, lab, and classroom to keep up with complex workloads like interactive graphics, data analytics, machine learning, and AI. NVIDIA RTX Server gives you the power to tackle critical day-to-day tasks and compute- heavy workloads – from home or wherever you need to work. Visual Computing Today Increasing daily workflow challenges
~6.5 Billion Render Hours Per Year 30,000 New Products Launch Each Year 2.5 Quintillion Bytes of Data Created Each Day
80% of Applications Utilize AI by 2020 $12.9 Trillion in Global Construction by 2022 $13.45 Billion Simulation Software Market by 2022
American Gods image courtesy of Tendril NVIDIA RTX Server High-performance, flexible visual computing in the data center
Highly Flexible Reference Design Delivered by Select OEM Partners I Scalable Configurations Ii Cost Effective and Power Efficient
Powerful Virtual Workstations
Accelerated Rendering
Data Science
CAE and Simulation NVIDIA Turing The power of Quadro RTX from desktop to data center
Physical Workstations Virtual Workstations
NVIDIA RTX Technology
Ray Tracing AI Visualization Compute NVIDIA RTX Server Spans Industries Near universal applicability
AEC Education Energy Financial Services
Government Healthcare Manufacturing Media and Entertainment NVIDIA RTX Server Workloads Covers all major solutions segments
Virtual Workstations Rendering Data Science Simulation
On-Demand Workstations Design and Visualization Offline Workstations for Workstations for Viewport and Render Workload Workstations Rendering Data Science Sim and Sci Viz Rendering Nodes
Applications Hypervisor Applications Hypervisor Hypervisor ISV Software Renderer Renderer Applications Renderer Data Science Software Sim and Sci Viz Apps Hypervisor
Quadro vDWS Quadro vDWS Quadro vDWS Quadro vDWS CUDA-X AI OptiX NVIDIA Software CUDA-X AI CUDA-X AI CUDA-X AI CUDA-X AI OptiX NGC Containers OptiX OptiX OptiX NGC Containers NVIDIA RTX Server Built on Quadro RTX 8000 and RTX 6000 active or passive
RTX 8000 and 6000 Active RTX 8000 and 6000 Passive
Form Factor Dual Slot PCIe Dual Slot PCIe
GPU Memory 48 GB GDDR6 ECC | 24 GB GDDR6 ECC 48 GB GDDR6 ECC | 24 GB GDDR6 ECC
NVLink 100 GB/Sec Bi-Directional (With Bridge) 100 GB/Sec Bi-Directional (With Bridge)
CUDA Cores 4608 4608
RT Cores 72 72
Tensor Cores 576 576
Rays Cast 10 Giga Rays/Sec 10 Giga Rays/Sec
Peak FP32 Performance 16.3 TFLOPS 14.9 TFLOPS
Peak FP16 Performance 32.6 TFLOPS 29.9 TFLOPS
Deep learning TFLOPS 130.5 TFLOPS 119.4 TFLOPS
Total Graphics Power 260 W 250 W
Thermal Management Active Fansink Passive
Auxiliary Power 8-pin PCIe + 6-pin PCIe 8-pin CPU Power Connector
Display Outputs 4x DisplayPort 1.4 + VirtualLink (USB-C) Not Applicable
Quadro Sync II Compatibility Yes Not Applicable NVIDIA Quadro RTX 8000 and RTX 6000 Active vs. passive performance
RTX 8000 and RTX 6000 Active RTX 8000 and RTX 6000 Passive Performance Delta
RTX-OPS 84 T 84 T None
Rays Cast 10 Giga Rays/Sec 10 Giga Rays/Sec None
Peak FP32 Performance 16.3 TFLOPS 14.9 TFLOPS Passive -8.60%
Peak FP16 Performance 32.6 TFLOPS 29.9 TFLOPS Passive -8.20%
Peak INT8 Performance 261.0 TOPS 238.9 TOPS Passive -8.40%
Deep Learning TFLOPS 130.5 Tensor TFLOPS 119.4 Tensor TFLOPS Passive -8.50%
Display Connectors 4x DisplayPort 1.4 + VirtualLink Not Applicable Passive requires Quadro vDWS
Max Power Consumption TGP 260 W | TBP 295 W TGP 250 W Passive TGP 10 W less
Power Connectors 8-pin + 6-pin PCIe (included) 8-pin CPU (included) Passive utilizes 8-pin CPU cable NVIDIA RTX Server Exponential power at a fraction of the cost
Rendering Artificial Intelligence Simulation
6x Dual 18-Core Skylake Servers 8x Dual 18-Core Skylake Servers 60x Dual 12-Core Skylake Servers
1x Eight GPU RTX Server 1x Single GPU RTX Server 1x Single GPU RTX Server
RTX Server Solution 1/4th the Cost RTX Server Solution 1/5th the Cost RTX Server Solution 1/7th the Cost VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server Virtual Workstations NVIDIA Quadro vDWS Delivering complete value to every workflow
Advanced Features Highest Performance Application Certification Enterprise Reliability
Datacenter Security Resource Optimization Data Proximity IT Management NVIDIA RTX Server for Virtual Workstations Workload configuration
Virtual Workstations for Design and Visualization
Hypervisor | ISV Applications
Quadro vDWS | CUDA-X AI | OptiX
RTX 8000 Active or Passive RTX 6000 Active or Passive
NVIDIA RTX Server Reference Design OEM Partners NVIDIA Quadro RTX Virtual Workstations Positioning and recommendations
Light users Medium Users Heavy Users Type of User Small to medium models, scenes or Large assemblies with simple parts or small Massive datasets, very large 3D models, assemblies with simple parts assemblies with complex parts complex designs, large assemblies
NVIDIA T4 or P6 NVIDIA T4 or P6 NVIDIA Quadro RTX 8000 or 6000 Recommended Solution (Perf/$) Quadro Virtual Data Center Workstation Active or passive, Tesla V100S, Quadro vDWS Quadro vDWS software (Quadro vDWS) software software
GPU Memory 16 GB 16 GB 48 GB | 32 GB | 24 GB
Equivalent Performance Multiple P1000 Up to Quadro P4000 Up to Quadro RTX 8000
Replaces K2, M60, P40, P4, M6 K2, M60, P40, P4, M6 M60, P40 Quadro Virtual Workstation Performance Work faster with larger models
1.5 Quadro Virtual Workstations
1.4x improved performance with Quadro RTX 8000 or RTX 6000 for virtual workstations I 1.0 2x GPU memory with Quadro RTX 8000 for larger model sizes I NVLink high-speed GPU interconnect pools GPU memory and scales performance 0.5
Added ray tracing and AI support with RT and Tensor Cores
P40 RTX 8000 or 6000 RTX 3D Graphics 1.4x Faster1
1SPECviewperf13 Geomean NVIDIA RTX Server for Virtual Workstations Broad industry support and NVIDIA NVENC adoption
Thin Clients Soft Clients
Hypervisor Platforms
Infrastructure Providers VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server Rendering NVIDIA RTX Server for Rendering Workload configuration options
Option One Option Two Option Three
Workstation by Day Offline Rendering (Final Frame) On Demand Viewport Rendering Render Node by Night
GPU Renderer ISV Applications | GPU Renderer | Hypervisor ISV Applications | GPU Renderer | Hypervisor
CUDA-X AI | OptiX Quadro vDWS | CUDA-X AI | OptiX Quadro vDWS | CUDA-X AI | OptiX
RTX 8000 Active or Passive RTX 6000 Active or Passive
NVIDIA RTX Server Reference Design OEM Partners 25x Accelerated Rendering for Netflix Renders in a fraction of the time using NVIDIA RTX Server
CPU Node NVIDIA RTX Server Performance (Dual Skylake) (4x RTX 8000) Improvement
Single Frame Render Time 38 Minutes 6 Minutes 6x
Total Render Time (120 frames) 76 Hours 3 Hours 25x
Number of Render Nodes 25 1 25x
Power Requirement (kW) 13.2 1.9 7x
Acquisition Cost $188,000 $25,000 7x
5 Year Cost of Power $68,000 $10,000 7x
Total Cost $256,000 $38,000 7x
6x Faster for a Single Frame
25x Faster for the Entire Shot
Render courtesy of Image Engine | Copyright Netflix | NVIDIA RTX Server was not used in actual Lost in Space production NVIDIA RTX Server Dramatically boosts rendering workload performance
Blender Cycles V-Ray Next GPU Autodesk Arnold SOLIDWORKS Visualize
RTX up to 8x faster than CPU RTX up to 18x faster than CPU RTX up to 17x faster than CPU RTX up to 17x faster than CPU
CPU vs. 2x Quadro RTX 6000 boards CPU vs. 2x Quadro RTX 6000 boards CPU vs. 8x Quadro RTX 8000 boards CPU vs. 2x Quadro RTX 6000 boards running Blender 2.811 running V-Ray Next GPU1 running Autodesk Arnold 6.0.11 running SOLIDWORKS Visualize 20201
1Performance results may vary depending on the scene NVIDIA RTX Server Rendering ISV Support 30+ RTX accelerated applications are available now
DaVinci Resolve VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server Data Science Data Science is Everywhere Use cases in every industry
Retail Financial Services
▪ Supply chain and inventory management ▪ Claim fraud
▪ Price management and markdown optimization ▪ Customer service chatbots and routing
▪ Promotion prioritization and ad targeting ▪ Risk evaluation
Telecom Manufacturing
▪ Detect network and security anomalies ▪ Remaining useful life estimation
▪ Forecast network performance ▪ Failure prediction
▪ Network resource optimization (SON) ▪ Demand forecasting
Healthcare Energy
▪ Improve clinical care ▪ Sensor data tag mapping
▪ Drive operational efficiency ▪ Anomaly detection
▪ Speed up drug discovery ▪ Robust fault prediction
Consumer Internet Automotive
▪ Ad personalization ▪ Personalization and intelligent driver interaction
▪ Click through rate optimization ▪ Connected vehicle predictive maintenance
▪ Churn reduction ▪ Forecasting, demand and capacity planning NVIDIA RTX Server for Data Science and AI CUDA-X AI end-to-end GPU accelerated workflow
Data Science Workstation Software Data Science Server Software Package
Frameworks Cloud ML Services Deployment
Amazon Google Amazon SageMaker Cloud ML SageMaker Neo
Machine Learning Serving
NVIDIA CUDA-X AI
DA GRAPH ML DL TRAIN DL INFERENCE
NVIDIA CUDA
NVIDIA Data Science Workstation OEM Partners NVIDIA RTX Server Reference Design OEM Partners Cloud NVIDIA’s GPU-Accelerated Data Science Workflow Built on CUDA-X AI and featuring RAPIDS
DATA
Data Data Preparation (ETL) Model Training Visualization Predictions
GPU accelerated compute for in- GPU acceleration of most Effortless exploration of billions memory data preparation popular ML algorithms such as of records in milliseconds Simplified implementation using XGBoost, PCA, K-means, k-NN, Dynamic data interaction, faster familiar data science tools DBScan, tSVD and others ML model development Data visualization ecosystem (Graphistry, OmniSci) integrated RAPIDS with Quadro RTX 8000 Unprecedented data science performance (shorter is better)
End-to-End
XG Boost
Data Prep
2x RTX 8000 RTX 8000 CPU
0 100 200 300 400 500 600
End-to-End time = ETL + conversion + training + validation | CPU Xeon 6140 at 3.2 GHz, 3.7 GHz Turbo, 384 GB RAM, Ubuntu 16.04.4, NVIDIA driver 410.93 TensorFlow with Quadro RTX 8000 FP16 large batch size
ResNet50 Batch Size 1024
ResNet152 Batch Size 512
InceptionV3 Batch Size 512
InceptionV4 Batch Size 512
VGG16 Batch Size 512
NASNet RTX 8000 2x RTX 8000 Batch Size 512
0 200 400 600 800 1000 1200
End-to-End time = ETL + conversion + training + validation | CPU Xeon 6140 at 3.2 GHz, 3.7 GHz Turbo, 384 GB RAM, Ubuntu 16.04.4, NVIDIA driver 410.93 Gauge-Equivariant CNNs Training neural networks on curved surfaces with fewer examples
Deep Ties to Physics, Significant Medical Implications – Including COVID-19
Gauge equivariance – quantities and their relationships don’t depend on ▪ arbitrary frames of reference (gauges) When converted different gauges must remain consistent and preserve the ▪ underlying relationships between things Equips the neural network with data assumptions in advance – a COVID-19 ▪ lung lesion is still a lesion, even if rotated or reflected within an image Ideal for detecting patters in data gathered from the irregularly curved ▪ surfaces of brains, hearts, lungs or other organs Identified visual evidence of lung abnormalities using just one-tenth of the ▪ data required to train other networks
3D G-CNNs for Pulmonary Nodule Detection
https://arxiv.org/abs/1804.04656 Deep Learning Drug Discovery Insilico Medicine designed DDR1 kinase inhibitors in 21 days
Generative Tensorial Reinforcement Learning (GENTRL) AI
Designed six novel inhibitors of DDR1, a kinase target implicated in fibrosis ▪ and other diseases, in 21 days Four of the compounds were found to be active in biochemical assays and two ▪ were validated in cell-based assays One lead candidate DDR1 kinase inhibitor was tested and demonstrated ▪ favorable pharmacokinetics in mice From identification to in vitro assays only took 46 days – 15 times faster than ▪ any other publicly disclosed pharmaceutical company R&D processes
Deep learning enables rapid identification of potent DDR1 kinase inhibitors
nature.com/articles/s41587-019-0224 (behind paywall) Quadro RTX GPUs are Shaping the Future of Cryo-EM AI assisted near-atomic scale resolution of biomolecules
Understand disease, develop new drugs, and administer medical treatments
Allows direct observation of proteins in native and near-native states without ▪ dyes or fixatives, providing molecular detail Requires thousands of images and complex computing, making it difficult to ▪ deliver high-resolution structures Traditionally required expert intervention, prior structural knowledge, and ▪ weeks of calculations on expensive computer clusters Topaz, an open-source DL application powered by NVIDIA RTX GPUs ▪ drastically reduces the amount of data that needs to be manually labelled
NVIDIA Quadro RTX graphics and deep learning performance are ideal
science.sciencemag.org/content/367/6483/1260 (behind paywall) VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server Computer Aided Engineering and Simulation NVIDIA RTX Server Simulation within the product design workflow
Conceptual Design Design Simulation Prototyping
Common simulation product design challenges include:
▪ Complex and lengthy process
▪ Computationally intensive
▪ High level of expertise required
▪ Often a time or resource bottleneck NVIDIA RTX Server For CAE and simulation
In-silico CAE and Simulation Product Development Workflow
Hypervisor | CAE Applications
OptiX | NGC Containers
RTX 8000 Active or Passive RTX 6000 Active or Passive
NVIDIA RTX Server Reference Design OEM Partners NVIDIA RTX Server On-demand power for engineering simulation
▪ Accelerate 32-bit simulation applications ▪ Faster model simulation and pre-processing ▪ Orders of magnitude speedup compared to CPU-only systems ▪ Visualize and evaluate physically accurate simulation results faster NVIDIA RTX Server Altair ultraFluidX accelerated CAE at scale on Quadro RTX 8000
8x 1 RTX 8000 2x RTX 8000 4x RTX 8000 6x RTX 8000 8x RTX 8000
7.3
6x 6.1 5.7
5.1
4x 4.0 4.0 4.1
3.1 3.2 3.2 2.9 2.8
2x 1.9 1.9
1.0 1.0
36 Million Voxels 36 Million Voxels with NVLink 230 Million Voxels 230 Million Voxels with NVLink NVIDIA RTX Server CAE and simulation GPU accelerated applications
For NVIDIA RTX Server Virtual Workstations For NVIDIA RTX Server Multi-GPU Compute
▪ Altair AcuSolve ▪ DASSALUT SYSTEMES SIMULIA CST ▪ Altair nanoFluidX ▪ Altair Hyperworks ▪ Impetus AFEA Solver ▪ Altair ultraFluidX ▪ Altair nanoFluidX ▪ META POST PROCESSOR ▪ Ansys HFSS SBR+ ▪ Altair ultraFluidX ▪ MSC Apex ▪ Ansys VREXPERIENCE ▪ Ansys Discovery Live ▪ ParaView ▪ AUTODESK Generative Design ▪ Ansys HFSS SBR+ ▪ ptc creo simulate ▪ DASSAULT SYSTEMES SIMULA CST ▪ Ansys SPEOS ▪ REMCOM Wireless InSite ▪ ParaView ▪ Ansys VREXPERIENCE ▪ REMCOM XF dtd ▪ SIMSCALE Pacefish ▪ Ansys Workbench ▪ SIEMENS FEMAP ▪ ANSA PRE PROCESSOR ▪ SIEMENS STAR-CCM+ VR ▪ AUTODESK Generative Design ▪ Simcenter ▪ DASSAULT SYSTEMES 3DEXPERIENCE ▪ SIMSCALE Pacefish ▪ DASSAULT SYSTEMES ABAQUS/CAE ▪ tecplot NVIDIA RTX Server High-performance, flexible visual computing in the data center
Highly Flexible Reference Design Delivered by Select OEM Partners I Scalable Configurations Ii Cost Effective and Power Efficient
Powerful Virtual Workstations
Accelerated Rendering
Data Science
CAE and Simulation NVIDIA Data Center Solutions NVIDIA RTX Server plays an essential role
NVIDIA RTX Server Quadro RTX 8000 and 6000 Active T4
Quadro RTX 8000 and 6000 Passive V100S
NVIDIA DGX
Quadro GV100 NVIDIA RTX Server Available from select PNY ecosystem partners | www.pny.com/rtxserver
[email protected] VR/AR Enterprise Value Propositions Data immersion and 3D conceptualization
NVIDIA RTX Server Interactive Q&A Session