Nvidia Corporation · NASDAQ:NVDA
Nvidia — Piper Sandler Webinar: Networks for AI (June 2023)
11 slides from Nvidia's conference presentation, Jun 28, 2023
Slide 1: Networks for AI
Networks for AI Gilad Shainer — SVP, Networking
Slide 2: Forward-Looking Statements
FORWARD-LOOKING STATEMENTS Except for the historical information contained herein, certain matters in this presentation including, but not limited to, statements as to: our markets and market opportunity; our growth and growth drivers; and the benefits, impact, and performance of our products and te
Slide 3: NVIDIA platform: application frameworks, software and hardware stack
APPLICATION FRAMEWORKS PLATFORM NVIDIA NVIDIA AI OMNIVERSE RTX CUDA-X CUDA ACCELERATION LIBRARIES SYSTEM Magnum IO DOCA Base Command Forge SOFTWARE RTX DGX HGX EGX OVX AGX MLNX HARDWARE GPU CPU DPU NIC SWITCH SOC
Slide 4: The Data Center is The Computer
The Data Center is The Computer The network defines the data center AI FACTORY • Cloud NVLink + InfiniBand • Multi-tenant • Variety of small-scale workloads • Traditional ethernet network can suffice AI CLOUD • Generative AI Cloud InfiniBand • Multi-tenant • Variety of workloads including larger sca
Slide 5: The NVIDIA AI Network Advantage
The NVIDIA AI Network Advantage Hardware & software accelerated in-network computing NCCL Performance With vs Without SHARP (In-network Computing) Software Acceleration NCCL — NVIDIA Collective Communication Library The SDK library for AI communications - connects the GPUs 1.7X HIGHER and the networ
Slide 6: In the Cloud — The Two Worlds of Ethernet
In the Cloud — The Two Worlds of Ethernet East-West for distributed and disaggregated processing | North-South for user-to-cloud communications Control / User Access Network (North-South) AI Fabric (East-West) Traditional Ethernet Spectrum-X Loosely Distributed Coupled Applications Tightly-Coupled P
Slide 7: NVIDIA Spectrum-X Ethernet AI Platform
NVIDIA Spectrum-X Ethernet AI Platform The ethernet fabric for AI Spectrum-4 Switch DOCA SONiC Cumulus NetQ Air Services 100 billion transistors, TSMC 4N 51.2T bandwidth, 100G SerDes 64 X 800G Ports, 128 X 400G ports SAI/SPSDK DOCA Magnum IO / NCCL 8K GPUs in 2-level, 500K in 3 level BlueField-3 Spe
Slide 8: NVIDIA Quantum InfiniBand Platform
NVIDIA Quantum InfiniBand Platform The gold standard for large scale AI Quantum-2 Switch DOCA NVOS UFM Air Services 57 billion transistors, TSMC 7N 64 X 400G ports, 100G SerDes 64 In-Network Computing AI Engines DOCA Magnum IO / NCCL 65K GPUs in 3-level, 2M in 4-level BlueField-3 Quantum-2 BlueField
Slide 9: The Network "Pays for Itself"!
The Network “Pays for Itself”! Performance gains far outweigh the entire cost of the network NDR 400G InfiniBand (64-port switches) Relative AI Fabric Performance (NCCL Allreduce) 3 switch level network topology (up to 65K nodes) 4 switch level network topology (up to 2M nodes) >2X 1.6X Traditiona
Slide 10: Strong Growth and Large Opportunity
Strong Growth and Large Opportunity Data Center Networking Growth Turbocharged by AI Revenue more than doubled since Mellanox acquisition Data Processing Units — In every server NVIDIA InfiniBand revenue more than tripled Data Center Switching — InfiniBand & ethernet Spectrum-X to boost cloud AI eth
Slide 11: Slide 11
Nvidia — Piper Sandler Webinar: Networks for AI (June 2023) on 6ix