Slide 1: NVIDIA Investor Call
TITLE NVIDIA Investor Call May 28, 2020
Slide 2: Challenges: Accelerating Big and Small
CHALLENGES: ACCELERATING BIG AND SMALL Computing For Training AI AI Interactions Per Day 1E+03 3000x Megatron-BERT Turing NLG 1E+02 Megatron-GPT2 GPT-2 Millions of Billions of Petaflop/s - Days 1E+01 10s Billions of Ecom Millions of Medical BERT Interactions Searches Recommendations Scans 1E+00 1E-0
Slide 3: Modern Cloud Data Center
MODERN CLOUD DATA CENTER Diverse Applications | Scale-Up & Scale-Out Workloads | Insatiable Demand
Slide 4: Reimagining the GPU
REIMAGINING THE GPU Three Breakthroughs to Fuel the Next Era of Modern Accelerated Data Centers 20X UNIFIED AI TRAINING AND INFERENCE 1-50 SCALABILITY FOR THE ELASTIC A GIANT PERFORMANCE LEAP ACCELERATION DATACENTER
Slide 5: Announcing NVIDIA A100: Greatest Generational Leap – 20X Volta
ANNOUNCING NVIDIA A100 GREATEST GENERATIONAL LEAP – 20X VOLTA Peak Vs Volta FP32 TRAINING 312 TFLOPS 20X INT8 INFERENCE 1,248 TOPS 20X FP64 HPC 19.5 TFLOPS 2.5X MULTI INSTANCE GPU 7X GPUs
Slide 6: Announcing NVIDIA A100: Greatest Generational Leap – 20X Volta
ANNOUNCING NVIDIA A100 GREATEST GENERATIONAL LEAP – 20X VOLTA 54 BILLION XTORS 3RD GEN TENSOR CORES SPARSITY ACCELERATION MULTI INSTANCE GPU 3RD GEN NVLINK & NVSWITCH 54B xtors | 826mm2 | TSMC 7N | 40GB Samsung HBM2 | 3rd gen Tensor Core GPU | 600 GB/s NVLink
Slide 7: New TF32 Tensor Cores
NEW TF32 TENSOR CORES Range of FP32 and Precision of FP16 | Input in FP32 and Accumulation in FP32 | No Code Change Speed-up for Training FP32 8 BITS 23 BITS TENSOR 8 BITS 10 BITS FLOAT32 FP16 5 BITS 10 BITS BFLOAT16 8 BITS 7 BITS
Slide 8: New Tensor Core Acceleration for Sparsity
NEW TENSOR CORE ACCELERATION FOR SPARSITY Optimized For Sparse AI Tensor Ops | 2X Faster Execution | Supported on TF32, FP16, BFLOAT16, INT8 and INT4 2X Effective Dense Matrix Sparse Matrix A100 Sparsity Optimized Tensor Core
Slide 9: New Multi-Instance GPU for Elastic GPU Computing
NEW MULTI-INSTANCE GPU FOR ELASTIC GPU COMPUTING 7x Higher Throughput of V100 with Simultaneous Instances per GPU
Slide 10: Unified AI Training and Inference Acceleration
UNIFIED AI TRAINING AND INFERENCE ACCELERATION BERT Training 7x BERT Inference 6x 1x 0.6x 1x V100 A100 T4 V100 A100 (7 MIGS) BERT Pre-Training Throughput using Pytorch including (2/3)Phase 1 and (1/3)Phase 2 | Phase 1 Seq Len = 128, Phase 2 Seq Len = 512 V100: DGX-1 Server with 8xV100 using FP32 pre
Slide 11: Announcing NVIDIA DGX A100: 3rd Generation Integrated AI System
ANNOUNCING NVIDIA DGX A100 9x Mellanox ConnectX-6 VPI 200Gb/s Network Interface 3 GENERATION INTEGRATED AI SYSTEM RD Dual 64-core AMD Rome CPU 5 PetaFLOPS of Performance in a Single Node 1TB RAM Unified System for End-to-End Data Science and AI Fully Accelerated Stacks — Spark 3.0, RAPIDS, TensorFlo
Slide 12: A100 Lighthouse Customers
ANNOUNCING NVIDIA A100 LIGHTHOUSE CUSTOMERS Elastic Data Center Accelerator Choice of Industry Leaders CLOUD SYSTEMS
Slide 13: Today's AI Data Center
TODAY’S AI DATA CENTER 50 DGX-1 Systems for AI training 600 CPU Systems for AI Inference $11M $11M 630 kW 25 Racks 630 kW
Slide 14: DGX A100 AI Data Center
DGX A100 AI 5 DGX A100 Systems for AI Training and Inference $1M 1 Rack $1M 28 kW 28 kW 1/10th COST 1/20th POWER
Slide 15: Announcing NVIDIA DGX A100 SuperPOD
ANNOUNCING NVIDIA DGX A100 SUPERPOD 140 DGX A100 Systems (1,120 A100) 170 Mellanox Quantum 200G InfiniBand Switches 280 TB/s Network Fabric - 15km of Optical Cable 4 PB of All-Flash Networked Storage 700 PFLOPS of AI Performance Built in under 3 Weeks
Slide 16: NVIDIA Expands SaturnV
NVIDIA EXPANDS SATURNV Before Expansion 1,800 DGX Systems 1.8 ExaFLOPS Adding 4 DGX SuperPODs 560 DGX A100 = 2.8 ExaFLOPS 4.6 ExaFLOPS Total Capacity
Slide 17: Smart Everything Revolution
SMART EVERYTHING REVOLUTION ALWAYS-ON | INSTANT SENSE-INFER-ACT | DISTANT | TRILLIONS
Slide 18: Announcing NVIDIA EGX A100 with Mellanox CX6 DX
ANNOUNCING NVIDIA EGX A100 WITH MELLANOX CX6 DX NVIDIA Ampere GPU 3rd generation Tensor Core New Security Engine for Confidential AI NVIDIA Mellanox ConnectX-6 DX Secure, Authenticated Boot Dual 100 Gb/s Ethernet or InfiniBand Line-speed TLS/IPSec Crypto Engine Time Triggered Transmission Tech for T
Slide 19: NVIDIA EGX Ecosystem
NVIDIA EGX ECOSYSTEM 5G & CloudRAN CONVERSATIONAL AI + SECURITY & NETWORKING INTELLIGENT VIDEO ANALYTICS ROBOTICS INFRASTRUCTURE MEDICAL SYSTEMS INDUSTRIAL CLOUD INDUSTRY LEADERS
Slide 20: GTC 2020 Announcements
GTC 2020 ANNOUNCEMENTS Merlin JARVIS Recommender Conversational System AI Data Center-Scale Omniverse NVIDIA AI DGX A100 EGX A100 Computing RTX Server Powered by A100 ISAAC, and BMW
Slide 21: Slide 21
Slide 22: Announcing NVIDIA Merlin: Deep Recommender Application Framework
ANNOUNCING NVIDIA MERLIN — DEEP RECOMMENDER APPLICATION FRAMEWORK NVIDIA Merlin USERS EMBEDDING 1TB Ads Dataset O(10) O(1000) Recommendation CPUs NVTabular HugeCTR CANDIDATES GENERATION RANKING ETL 2 HR TRAINING 1 ½ Days … O(Billions) ITEMS EMBEDDING User Query GPUs RAPIDS cuDNN TensorRT ETL 3 Min T
Slide 23: Announcing NVIDIA Jarvis: Multimodal Conversational AI Services Framework
ANNOUNCING NVIDIA JARVIS — MULTIMODAL CONVERSATIONAL AI SERVICES FRAMEWORK PRE-TRAINED MODEL RE-TRAIN NVIDIA JARVIS Transfer Learning Dialog Manager NeMo Multi- Gesture Speaker Chatbot Look to Talk Recognition Transcription Service Maker NLU Vision NVIDIA GPU CLOUD NVIDIA AI TOOLKIT Speech JESSICA:
Slide 24: Announcing NVIDIA Accelerates Spark 3.0
ANNOUNCING NVIDIA ACCELERATES SPARK 3.0 ETL TRAINING INFERENCE USERS EMBEDDING O(10) O(Peta) O(Tera) TensorFlow O(1000) Recommendation CANDIDATES RANKING PyTorch GENERATION O(Billions) User Query ITEMS EMBEDDING RAPIDS cuDNN TensorRT Magnum IO Magnum IO Triton Inference Server Data Lake 100’s PB