英伟达2026 GTC大会 黄仁勋演讲_第1页
英伟达2026 GTC大会 黄仁勋演讲_第2页
英伟达2026 GTC大会 黄仁勋演讲_第3页
英伟达2026 GTC大会 黄仁勋演讲_第4页
英伟达2026 GTC大会 黄仁勋演讲_第5页
已阅读5页,还剩71页未读 继续免费阅读

付费下载

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

2026AnnouncingDLSS

53D-GuidedNeuralRenderingStructuredDataistheGroundTruthofAI$120BStructuredDataEcosystem-TheGroundTruthofEnterprisesEnterprisestbcroageP

。tforrsC

P

Engine

ccommercicl

css

户lotformCUDFroprietry

Enginess

EnginesSource:Gartner

DBMS,2025Unstructured

Data

is

the

Context

of

AI100’sofZettabytesPerYearofUnstructuredData-GrowingExponentiallySource:IDC

Global

DataSphere,

2025“Working

with

IBM

and

NVIDIA,early

results

show

we

can

refresh

global

operationsdata

in

minutes

atreducedcost—makingacapabilitythatcanbeturned

intotangiblebusinessimpactinareassuchasmanufacturingorwarehousing.

”―ChrisWright,ChiefInformationandDigitalOfficerof

NestléIBM

watsonx.dataNVIDIA

A100IBM

Storage

ScaleData

Engines:Open

Formats:

多parquetDataGovernanceandAccessControlThroughout1,0007505002500NVIDIAcuDFOrder-to-CashDataMartUpdates3min

vs

15min83%CostSavingsCPU

GPU“DeIIAIDataPIatformwithNVIDIAheIpsusbuiIdscaIabIeandrepeatabIedatapipeIines

that

drive

automation—processing

massive

data

voIumes

inminutes

instead

of

hours

and

deIivering

transformative

vaIue

for

our

cIients.”―AbhijitDubey,CEOof

NTT

DATADellAI

Data

PlatformDataOrchestration

EngineData

EnginesNVIDIAcuVS

NVIDIAcuDFHigh

PerformanceScale-OutStorage

NVIDIA

CMX

Data

ProtectionandCyber

ResilienceNVIDIARTX

PRO6000

BlackwellDellAI

Factorywith

NVIDIAEnterpriseDataProcessing

FromHours

to

Minutes10TB

Dataset100806040200CPU

GPU3XFaster“Our

collaboration

with

NVIDIA

and

Google

Cloud

helps

us

innovate

faster

for

more

than

abillion

Snapchatters

worldwide.By

lowering

costs

and

scaling

experiments

across

petabytesof

data,we’re

delivering

AI-powered

experiences

more

quickly

and

efficiently.

”―SaralJain,CIO

ofSnap

GoogleCloudAI

HypercomputerOpenSoftwarePerformance-optimized

Hardware50,00040,00030,00020,00010,0000FlexibleConsumptionA/B

Experimentation

Platform

45kCPUsto

1k

GPUsNVIDIAcuDFNetworkingJupiter

NetworkComputeG2VM

(NVIDIA

L4GPUs)

G4VM

(NVIDIA

RTX

PRO

6000)StorageGCS

RapidGCS

LustreApacheSparkJobsApacheSpark

Engine76%CostSavingsGoogleKubernetes

EngineControl

plane,APIs,Scheduling

and

monitoringDataprocManagedApacheSpark

and

HadoopCPU

GPUAIOptimizedInfrastructureAIDataScience

andAnalyticsJAXOpenSoftware,FrameworksXLAvLLMPlatformsLiquid

CoolingStorageNetworkingGoogleDistributedCloudVertexAIGoogleKubernetesEngineCloudRun

GoogleCompute

EngineBigQuery

Dataproc

Dataflow

BigtableSystemBuildersandOEMsNVIDIAAIPlatformAmazonBraketAmazonEKSAWSNitro

EnclavesAWSBatchAmazonEMRAmazonBedrockAmazonSageMakerAIAmazonEC2AmazonEFAAWSParallelClusterSystemBuildersandOEMsNVIDIAAIPlatformMicrosoftFoundryWindows365MicrosoftFabricAzureContainer

AppsAzureKubernetesServiceFoundryLocalMicrosoft365

CopilotMicrosoftBingSystemBuildersandOEMsNVIDIAAIPlatformOCISuperclustersOracleAIDatabaseOracleAutonomousAILakehouseOracle

AIDataPlatformOCIAIAcceleratorPacksOCIGenerativeAIOCIDataScienceOCIOKESystemBuildersandOEMsNVIDIAAIPlatformW&BModelsW&BTrainingW&BInferenceW&BWeaveCoreWeaveInferenceCoreWeaveSlurm

onKubernetesCoreWeaveKubernetesServiceCoreWeaveComputeCoreWeaveStorageCoreWeaveMissionControlSystemBuildersandOEMsNVIDIAAIPlatformAIP

LogicModelCatalogWorkshopAI

FDERubix

Ontology

Pipeline

Builder

Fusion

Agent

Studio

Model

StudioDellNVIDIAAIPlatformDL

FrameworksAgentFrameworks/ProtocolsFrontierModelBuildersModeltoProductionInferenceFrameworks

--

.-

---

-,r..r.,

-it.

-

i-

s-

AI

for

Auto

AI

for

Customer

Support

AI

for

Engineering

AI

for

Healthcare

AI

for

Robotics

AI

for

Search

AI

for

Software

Development

AI

NativesNVIDIAAIPlatformInferenceInflectionArrives10,000X

ChatGPT

ComputeInferenceInflectionDrivesStrongGrowthNVIDIA

Full-StackExpandingAItoAllRegionsandIndustriesANTHROPICMETA

SLMU儿TIP

LE

OSSTokensperWattDrives

FactoryRevenuenceXy$4.50$4.00$3.50$3.00$2.50$2.00$1.50$1.00$0.50$0.00PerformanceDrivesTokenCost9,000,0008,000,0007,000,0006,000,0005,000,0004,000,0003,000,000b2,000,0001,000,00000

50

100

150

200H200NVL8Competition35XLowerCostGB300NVL720

50

100

150

200Interactivity(TPS/User)Interactivity(TPS/User)NVIDIAExtremeCo-DesignRevolutionizedTokenCost“GB

NVL72

Inference

King”DeepSeekR10528·FP4·

1K/1K·Source:SemiAnalysis

InferenceX50XHigherPerf/

WattInfereCompetitionGB300NVL72H200NVL8NVIDIAistheGlobalStandardforAIInferenceatScaleAllLeadingInference

End-PointsRun

on

NVIDIAKimi

K2.5ReasoningAI

FactoriesaretheIndustrialInfrastructureoftheAIEraInferenceistheWorkloadTokensaretheNewCommodityComputeis

RevenueInferencePerformanceandEfficiencyDriveCompanyResults1.8

M1.6

M1.4

M1.2

M1.0M0.8

M0.6

M0.4

M0.2

M50

1002004006008001,000+InteractivityTPS/UserInferencePerformanceandEfficiency

DriveCompanyResults$6

$45

$150Free

MediumQwen

3

Kimi

K2.5

235

Billion

1

Trillion32K

128K$0

$3PremiumGPT

MoE2Trillion

400KUltraGPT

MoE2Trillion400KHighGPT

MoE2Trillion128KThroughput

TPS/MW1.8

M1.6

M1.4

M1.2

M1.0M0.8

M0.6

M0.4

M0.2

M

HopperInferencePerformanceandEfficiency

DriveCompanyResults100InteractivityTPS/UserMediumKimi

K2.51Trillion128K$3FreeQwen

3235Billion32K$0Throughput

TPS/MW20050InferencePerformanceandEfficiency

DriveCompanyResultsBlackwellNVL72Hopper50$0

$3

$60.8

M0.6

M0.4

M0.2

M1.8

M1.6

M1.4

M1.2

MMediumKimi

K2.51Trillion128KFreeQwen

3235Billion32KInteractivityTPS/UserHighGPT

MoE2Trillion128KThroughput

TPS/MW1.0M35X200InferencePerformanceandEfficiency

DriveCompanyResultsBlackwellNVL72Hopper0.8

M0.6

M0.4

M0.2

M1.8

M1.6

M1.4

M1.2

MPremiumGPT

MoE2Trillion

400K$45200InteractivityTPS/UserMediumKimi

K2.51Trillion128K$3FreeQwen

3235Billion32K$0HighGPT

MoE2Trillion128K$650

100Throughput

TPS/MW1.0M35X400RubinNVL722X2XBlackwellNVL723XHopperInferencePerformanceandEfficiency

DriveCompanyResults1.4

M1.2

M1.0M0.8

M0.6

M0.4

M0.2

MPremiumGPT

MoE2Trillion

400K$45200InteractivityTPS/UserMediumKimi

K2.51Trillion128K$3FreeQwen

3235Billion32K$0HighGPT

MoE2Trillion128K$61.8

M1.6

MThroughput

TPS/MW40010010X50$150B$100BAnnual

RevenuePerGigawatt$50B$0BNVIDIAVeraRubin

Unlocks$150BRevenueOpportunityPremiumGPT

MoE2Trillion400K$45MediumKimi

K2.51Trillion128K$3FreeQwen

3235

Billion32K$0HighGPT

MoE2Trillion128K$6 Blackwell

RubinTotalAnnualRevenue

5X

$150B$30BRubinNVL722X2XBlackwellNVL723XHopperInferencePerformanceandEfficiency

DriveCompanyResults1.4

M1.2

M1.0M0.8

M0.6

M0.4

M0.2

MPremiumGPT

MoE2Trillion

400K$45200InteractivityTPS/UserMediumKimi

K2.51Trillion128K$3FreeQwen

3235Billion32K$0HighGPT

MoE2Trillion128K$61.8

M1.6

MThroughput

TPS/MW40010010X50RubinNVL722X2XBlackwellNVL72Hopper50

100InferencePerformanceandEfficiency

DriveCompanyResults1.8

M1.6

M1.4

M1.2

M1.0M0.8

M0.6M0.4

M0.2

M0.0M3X200InteractivityTPS/UserPremiumGPT

MoE2Trillion

400K$45MediumKimi

K2.51Trillion128K$3FreeQwen

3235Billion32K$0HighGPT

MoE2Trillion128K$635X10XThroughput

TPS/MW40050

1002004006008001,000+InteractivityTPS/UserHighGPT

MoE2Trillion128KPremiumGPT

MoE2Trillion400KUltraGPT

MoE2Trillion400K3XHopper

35X35XPerformanceEfficiencyDrivesCompanyResults1.8

M1.6

M1.4

M1.2

M1.0M0.8

M0.6

M0.4

M0.2

M$6

$45

$150Free

MediumQwen

3

Kimi

K2.5

235

Billion

1

Trillion32K

128K$0

$3BlackwellNVL72Throughput

TPS/MWRubinNVL72Rubin

+

LPX35X2X2XNVIDIAVeraRubin

+

LPXUnlocks$300B

RevenueOpportunity$300B$150B$30B$300B$250B$200B$150B$100B$50B$0BPremiumGPT

MoE2Trillion400K$45MediumKimi

K2.51Trillion128K$3 Blackwell

Rubin

VR+LPX10XFreeQwen

3235

Billion32K$0HighGPT

MoE2Trillion128K$6UltraGPT

MoE2Trillion400K$150TotalAnnualRevenueAnnual

RevenuePerGigawattRubinGPUGroq

3

LPU500MB

SRAM150TB/s

SRAM

Bandwidth

1.2

PFLOPs

(FP8)98B

TransistorsUnitingProcessorsofExtremePerformances288GB

HBM422TB/s50PFLOPs

(NVFP4)336B

Transistors

+

2.5T

(HBM4)RubinGPUGroq

3

LPU4GB

SRAM1,200TB/s

SRAM

Bandwidth

55X

9.6PFLOPs

(FP8)784B

TransistorsUnitingProcessorsofExtremePerformances288GB

HBM422TB/s50PFLOPs

(NVFP4)336B

Transistors

+

2.5T

(HBM4)UnitingProcessorsofExtremeFLOPSand

BandwidthNVIDIADynamoVeraRubin

NVL72Groq3

LPXRemaining

TokensInput

1st

TokenKV

CacheDecodePrefill/

/

LPU

C2C

Spine

Connectors

./

8

LPUsFPGA

Host

CPU

\

/

BF4LPUC2C

Links

AIInferenceComputeSRAM

CapacityMemoryBandwidthScale-UpDensityScale-UpBandwidth315PFLOPS

128

GB40PB/s256

Chips640

TB/sNVIDIAGroq

3

LPXAvailable2H26●AI

LabsAnnouncingNVIDIAVeraRubinNVL72LaunchPartners10X

Perf/W3.6EF

NVFP4|

1.6

PB/s

HBM4

|

260TB/s

NVLink6OEMsandODMsCloudOEMsandODMsAnnouncingNVIDIAVeraCPULaunchPartnersCloud256Vera

CPUs|300

TB/s

LPDDR5X

ETL

Spine6.5X

ThroughputAnnouncingNVIDIABlueField-4STXLaunchPartnersCloud5x

Tokens/sec|50

Tb/s

Networking

BW

|

16TB

Shared

Context/

GPUOEMsRubinGPUVeraCPUCX9BF4NVLinkSwitchSpectrumCPOGroq

3

LPUVera

Rubin

Compute

Tray

NVLink

Switch

Tray

Vera

Compute

Tray

BF4STX

Server

Spectrum

Switch

Groq3Compute

TrayNVIDIAVeraRubin7Chips-5

RackSystemsAI

FactoryfortheAgenticAI

Frontier1GW

AI

Factory#of

GPUsAI

FLOPSAll-to-AllScale-upMemoryBW-per-Domain

(GROQSRAM)Tokens

per

SecondX86+

Hopper600K1.2ZFLOPS7.2TB/s2

EB/s2MVera

Rubin300K16ZFLOPS260

TB/s100EB/s700MNVIDIAExtremeCo-DesignDeliveringX-FactorsEveryYearFromChipstoRacks

toAI

FactoriesFeynmanRubinLP40LP30

LP35

NVLinkNVFP4GraceCPU2026Oberon

OberonKyberNVL72

ETL256NVL144NVL576BlueField-3BlueField-4Spectrum5

51T2024OberonNVL72NVLink8

CPOSpectrum7

204T,CPONVLink

6Switch

3600GB/sSpectrum6

102T,

CPONVLink

7Switch

3600GB/sCX91600GFeynmanDieStacking

CustomHBMKyberNVL144NVL11522028OberonETL256OberonNVL72Rubin

UltraHBM4eBlueField-5Blackwell

UltraHBM3eBlackwellRosaCPUNVLink

51800GB/sRubin

HBM4Blackwell

HBM3eVeraCPUCX10CX8800GReference

Designs,Best

Practices,

MethodologiesNVIDIADSXAI

FactoryPlatformExtremeCo-DesignatInfrastructureScalePower

LiquidCoolingDSX

Max-QDSX

FlexDSX

ExchangeDSXSimChips,Systems,

FacilitiesDSXLibraries,APIs,SoftwareRubin

DSXReferenceDesignsOmniverseDSX

BlueprintDSXEcosystemNVIDIADSXAI

FactoryPlatformAcceleratesScalable,Energy-EfficientAIFactoryDeployment

DSX

Max-Q

DSX

Exchange

DSX

Flex

DSXSim

AnnouncingNVIDIASpace-1VeraRubin

ModuleInferenceInflectionArrivesAnnouncingNVIDIANemoClawforOpenClawAgents-A

New

Computing

PlatformEnterpriseITRenaissancefromSaaS

toAgent-as-a-Service$3T

EnterpriseITRenaissancefromSaaS

toAgent-as-a-Service$3T

$30TCU

DF,

CVSVGPWAnnouncingNVIDIANemoClawRef

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论