infrastructure choices for machine learning and deep...
TRANSCRIPT
Infrastructure choices for Machine Learning and Deep Learning
Vikas Ratna, Ravi Mishra
Cisco Product Management
Session ID: E8356
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
How Important is AI & ML?
By 2035,
AI technologies are
projected to increase
business productivity
by up to
40%
By 2020, insights-driven
businesses will steal
$1.2T
per annum from their
less-informed peers
8 out of 10
businesses have already
implemented or are
planning to adopt AI as
a customer service
solution by 2020
• $1.2T https://go.forrester.com/wp-content/uploads/Forrester_Predictions_2017_-Artificial_Intelligence_Will_Drive_The_Insights_Revolution.pdf
• 8 out of 10 - Oracle - https://www.oracle.com/webfolder/s/delivery_production/docs/FY16h1/doc35/CXResearchVirtualExperiences.pdf
• Accenture https://www.accenture.com/us-en/insight-artificial-intelligence-future-growth
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public• 277X data created by IoE devices vs. end users – source: 2014 Cisco® Global Cloud Index
• By 2020, there will be 5200 GB of data for every person on earth – source: 2012 Digital Universe Study conducted by IDC and sponsored by EMC
(see: http://www.computerworld.com/article/2493701/data-center/by-2020--there-will-be-5-200-gb-of-data-for-every-person-on-earth.html)
• 180 billion mobile app downloads by 2015 – source: 2011 IDC Study: https://www.smaato.com/blog-180billiondownloads/
277XData created by IoT devices vs.
end users
30MNew devices connected every week
180BMobile apps downloaded
in 2015
40%Of all data will
come from sensor
data by 2020
5TB+Of data
per person by 2020
4.2BWeb
filtering blocks per
day
Why Now?First…The Data Deluge - Volume, Velocity, Heterogeneity
Unrealistic to leverage traditional analytics to mine meaningful information at speed.
AI activates the potential of raw data into powerful competitive advantage.
Why Now?
Second –Perfectly timed Infra, Compute and ML Framework Innovation
Improved
Data Collection Infrastructure
Increased
Computing Power
Advancement in
ML Frameworks
Hadoop w/Apache Spark Faster GPUs and CPUs TensorFlow, PyTorch, Caffe, Theano
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Quick Recap: What is AI
Act of Artificial
Intelligence
Machine
Learning
Deep
Learning
Machines Making
Decisions
Learn From
Traditional Data and
Make Decisions
without Explicit
Programming
Use Artificial Neural
Networks to Learn
From Complex Data
and Make Decisions
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Business Drives AI
RetailSecurity and
DefenseMedia and
EntertainmentHealthcareFinanceAI Task
Business
QuestionManufacturing
Is "it" present
or not?Detection
Identify
Access
Anomalies
Indication of
Anomaly in
Scan
Content
Based Search
Identify
Security
Breaches
Events in
Store
Surveillance
Detect
Manufacturing
Flaws
What type of
thing is "it"?Classification
Fraud
detection
Diagnostics –
Tumor?
Content
Labeling
Facial
Recognition
Returning vs
New
Shoppers
Robots to
Track Objects
To what
extent is "it"
present?
SegmentationSentiment
Analysis
Condition
Analysis
Improved
Product
Placement
Crowd
Analytics
Segment by
Customers
Actions
Sort
Components
by Quality
What is the
interpretation?
Natural
Language
Processing
Chatbot
Advisors
Expert
Diagnosis
from Notes
Video
Captioning
Real Time
Language
Translation
In Store
Personal
Assistants
Assembly Build
Instruction
Translation
What is the
likely
outcome?
PredictionCredit
Profiling
Length of Stay
Forecasting
Targeted
Content
Generation
Equipment
Health
Assessment
Customer
Churn and
Retention
Proactive
Machine
Maintenance
What will
satisfy the
objective?
Recommenda
tions
Algorithmic
Trading
Treatment
Recommenda
tions
Content
Recommendati
ons
Risk
Management"Magic Mirror"
Assembly
Process
Improvements
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Built-In AI vs AIaaS vs Purpose Built
Tetration Analytics
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
An Example Of Purpose Built Infrastructure
Big Data DL Framework
TrainingInference
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Popular DL Frameworks To Choose From
• Python Based
• Popular for Production
• Static Graph
• More Control and Work
• Apache 2.0 License
• Great Documentation
• Community Supported
• Python Based
• Popular for Research
• Dynamic Graph
• Less Control and Work
• BSD License
• Easy to Debug and
Prototype
• Community Supported
• JVM Based
• Popular in Big Data
• Static Graph
• Integrates well with Kafka,
Spark and Hadoop
• Apache 2.0 License
• High Performance
• Commercially Supported
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Data Science Tools To Choose From
Environment Libraries Algorithms Frameworks PlatformUser
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
High Level View Of Overall Workflow
Data
InfrastructurePrepare Data Train a Model
Evaluate
the Model
Deploy &
Improve
More Data Trumps
Better Algorithms
Data and Business
Questions will
Determine ML
Algorithm(s)
Data Can Come
From Anywhere and
is Not Usually in a
State Where it is
Ready to Use for
Training ML Models
Build a Model and
Feed the Model with
Prepared Training
Data so that it Can
Learn to Make
Inferences
Test the Trained
Model Performance
and Accuracy by
Analyzing Inference
Feedback
Improve Performance
by Either Selecting
Different Algorithms
and/or Retraining
Model with More Data
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
• Hadoop native support for GPU
YARN 2.0 /Resource Manager Support for GPU and CPU
• First class support for Docker on YARN
Docker based application/assemblies support on Hadoop
• AI/ML Frameworks on Hadoop
Tensorflow/Caffe/etc docker containers on YARN accessing data on Hadoop, scheduling resource (GPU/CPU)
• HDFS Federation (Federated Namenodes)
• HDFS with Erasure encoding
(1.7 copies instead of 3) Primarily for Cold data
• cblock (volume driver) for kubernetes
Technology Trends with HadoopHadoop 3.0 with GPU
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
An Example: Edge-To-Core AI/ML With Big Data
IOT Data
Sources
Web Data Sources
Streaming Data Sources
Gateway with Kafka
HDFS
No SQL
ML Training & Inference
Se
rvin
g L
aye
r
Speed Layer Batch Layer
Spark Streaming
Compute Nodes
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
59%
25%
6% 6%
Knowledge gathering/developing strategy
Piloting Implementing Deployed/in use today
Investigating Using
The majority of organizations are still gathering
information to inform their AI adoption strategy
It’s Still Early Days….
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Customer AI Challenges
• C-Suite wants an AI strategy
• Marketing wants to include AI in everything
• Worried about falling behind and competition passing
• Not sure what problems to solve, Not enough expertise
• Everyone is pitching them and they are overwhelmed
Looking For Guidance, Simplification
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Key Challenges for ITMassive & Active
Data Sets
Volume, Velocity, and Variability of
AI Workloads at Scale Demand
New Data Center Architectures
Uncharted
Territory
Rapidly Evolving AI/ML
Ecosystem and Requirements;
Skill Shortages in Data
Science and IT
Infrastructure
Challenge
Distributed Data Sources and
Technologies Risk Operational
Silos and Complexity
Looking For DC Infrastructure Which is: Silo-less, Simple to manage at scale, Validated for new
workload
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Cisco Strategy Product Portfolio
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Recap: Cisco Unified Computing System (UCS) Momentum
Source: 1 IDC, 2017 Q4, Sep 2017, Vendor Revenue Share
Source: As of Cisco Q4FY17 earnings results. Data Center Revenue is defined as Cisco UCS and Nexus 1000V
Integrated Infrastructure
(Cisco UCS, Nexus, FlexPod)
#1
HyperCoverged vendor
#3
Global Revenue
Market Share in x86 Blades
#2
World Record Performance
Benchmark
150+
Exabyte Total Storage
Deployed
1.4
Big Data Revenue
Growth in 4 Years
18x
Enterprise customers
64,000+
of Fortune 500 Have
Invested in UCS
>85%
HyperFlex Customers
3000+
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Component
Cisco UCS Portfolio: Any Workload. Any Scale. One System.
HyperconvergedInfrastructure
Converged InfrastructureROBO
Fifth Generation UCS
HyperFlex Systems
UCS MiniE-Series
Data Intensive
S-SeriesStorage Optimized
C-Series Rack Servers
UCS Integrated Infrastructure
Solutions
CoreEdge Cloud
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Eliminating Operational Silos Demystifying AI/ML/DL
StacksCurating top-to-bottom SW and HW
stacks with leading ecosystem partners to
ensure a faster and more predictable
deployment
Full array of accelerated
computing options for test/dev,
training and inference, all unified
by cloud-based management
Integrating changing data sources as part of a
dynamic data pipeline
Powering the Full AI Data Lifecycle
Introducing: Cisco Computing Solutions for AI/ML
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Inference in Regional and Micro Data Centers
Cisco AI/ML - Computing Platforms, Partners, and Solutions
UCS C220 Servers
HyperFlex C220 Servers
UCS C240 Servers
HyperFlex GPU Nodes
Test/Dev and ML in Private Cloud
Deep Learning in DC Core
Accelerated Computing ISV Partners Solution Partners
NEW!
UCS C480 ML
UCS C480 ML
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Cisco Portfolio Alignment
Use Case
Conception
Feasibility
Study
Model
Design
Model
Deployment
Model
Maintenance
Optimized Big Data
Solutions for Data
Collection and
Preparation
Distribute and
Scale AI Inference
from the Data
Center to the Edge
Offload and
Accelerate AI
Training at Scale
with Performance
Optimized Systems
V100 P4
NVLink
C240 M5 C480 ML M5 HX220c M5
Quick Installation
of Hardware and
Software for AI
Exploration and
Experimentation
V100
HX240c M5
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Cisco AI/ML Compute Portfolio
Test & Dev and Model Training
C240
2 x P4
6 x P4
HyperFlex
240
Deep Learning/ Training
C480
Inferencing
C/HX 220
C/HX 240
Unified Management
Simplified Management, Customer Choice, Cisco Validated Design
Option of GPU Only Nodes
2 x P100/ V100 2 x P100/ V100 Per Node
6 x PCIe P100/ V100 8x V100 with NVLink
C480 ML
Cisco IMC XML API
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
DC Infrastructure AI/ML/DL Platforms
Data / BigData
Libraries
FrameworkAI-ML
Server / Appliances
(CPU or GPU)
Simplified DC Infra To Support AI/ML Workload & Cisco
AI/ML
Apps/
APIs
Services
Manageability
Network
DC Servers and
Storage
Model Deploy/Mgmt
Existing UCS Business
Several CVDs exist
Strategic Partnerships / Cisco AS
UCSM, Intersight
Exists, Build New Partner
Build CVD
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Cisco UCSCisco HyperFlex Cisco ACICisco Multicloud Portfolio
Cisco AI/ML- A Holistic Approach: Edge to Core
Accelerated Computing
for Inference at the Edge
Accelerated Computing
for AI/ML in the Core
Big Data/Analytics
and ERP
AI/ML Stack
Partnerships
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Specification• 8 x SMX2 V100 32G Module with NVLink interconnect
• Support for Dual Intel Xeon Scalable Processor Family
• Support for 2666 MHz DDR4 DIMMS with capacity point of 128GB
• Supports up to 24 SAS/SATA HDD/SSD
• Supports up to 6 NVMe drives
• Supports up to 4 x100G VIC
• Modular Internal Flex Storage Option: M.2 SATA
• 4 x 1600W PSUs in 3+1 Redundancy
• 1 x 10/1000 dedicated management Ethernet Port
• Two 10 Base-T Gbps Ethernet Ports
Accelerate AI/ML Projects
Validated with Popular Machine
Learning Software
Industry leading Spec
Fully Integrated Platform Designed
to Accelerate Deep Learning
Simplified
Management
Cisco Intersight or
UCSM/CIMC managed
Prevents Operation Silos
Extends Existing UCS
Environments with Consistent,
Cloud-Based Management
Cisco UCS C480ML Rack Server for Deep LearningA No-Compromise Purpose Built Server for Deep Learning
40,000+ CUDA Cores
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
System Overview
CPU Module
CPU Blank
Module
KVM Access
Storage
2 x PCIe Slots(part of GPU
module)
For 10/25 or
40/100G VIC or
3rd party adapter
2 x PCIe Slots (part of GPU
module)
For 10/25 or
40/100G VIC or
3rd party adapter
4x1600W PSU Slots
Hot swappable and redundant
3+1 Redundancy
IO ModuleTwo 10 Base-T Gbps shared LoM
Display VGA port, Serial port
Dedicated CIMC mgmt. port
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
UCS C480ML - Overview
Raid Controller
NetworkChoice of 10/25 or 40/100G Four PCIe SlotsTwo 10G Base-T shared LOM on IO Module
GPUs8 X V100 32GB: 1st GPU to break 100 teraflops NVLink Interconnect: >300GB/s bandwidth
Redundant Fans
StorageUp to 24 SAS/SATA SSD/HDDUp to 6 NVMe DrivesM.2 SATA
CPUs2 * Intel® Xeon® Processor Scalable Family(Up to 28 cores per socket)24 DDR4 DIMMS—up to 3 TB Memory
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Storage Options
• Most powerful and flexible storage option in the Industry for Deep Learning workloads
• 24 Front Panel Drive Slots in 3 cages
Cage 1 Cage 2 Cage 3
NVME
slots
Cage 1 Cage 2 Cage 3
SAS/SATA drives 4 7 7
NVME/SAS/SATA 4 1 1
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Networking Options
VIC 1455 VIC 1495
Speed 10/25G 40/100G
Ports 4 2
Form Factor PCIe PCIe
FI 6200/6332/6454 6332/6454
UCS FI
Direct Connect
Single cable for both management
traffic and data traffic from VIC and
connected directly to FI
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Network Connectivity Options
Dual 10G
LoM
10G Bandwidth
Dual 10G for UCS Management and
Data
Dual 10G LoM
for MGMT and
VIC for Data
10/25/40/100G Bandwidth
Data
(10/25/40/100G)
MGMT (10G)
Dual 10G for UCS management and
multiple10/25/40/100G VIC for data
10/40G Bandwidth
FEX – 2232,
2348
Data
MGMT
UCS FI
VIC and
Dual 10G LoM
Dual 10G for UCS management and
multiple10/40G VIC for data
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Centralized Management
Across Domains and Geographies:
Up to Thousands of Servers
Global Policies
Centralized Templates
Service Profiles
Policy-based Infrastructure
No Silos
Racks, Blades, and
Hyperconverged
Embedded Management
Integrated ’Agent-less’ out of Band
Remote Management
Scriptable Interface
Large scale Management with XML
API, Python SDK, Redfish
UCS C480ML Management Options
UCS Manager Cisco Intersight (SaaS) Cisco IMC
Data Center 1 Data Center 2UCS Manager UCS Manager
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
• PCI switch info from UCS management
• GPU and PCI adapter and link information connected to that PCI switch
GPU Module Info and Management
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
6542
37783208 2830
22035
12161
6423
3735 32182793
21078
11438
Tensorflow Training: 8GPU (V100)
Synthetic Data Real Data
C480 ML Performance DataIm
ag
es
/se
c
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Data Centric Approach: Expanding Big Data to AI/ML SolutionsCisco Validated Designs
Cloudera Data Science Work Bench Kubeflow
Hortonworks Hadoop 3.1 Data Lake
Hadoop Coupled with GPU Nodes for Deep Learning with
Jupyter Notebook
Portable, Scalable ML Stack Enabling Rapid Development
and Deployment
Integrate Hadoop and AI/ML: YARN Scheduling CPU and GPU with Docker Application Support
YARN Scheduler
HDFS Hot Tier
HDFS Cold TierHDFS
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Big Data CVD with Hortonworks and ClouderaUCS C480ML as Datanode and GPU node on Hadoop
Nvidia CUDA containers
orchestrated with YARN for
Deep Learning
ML/DL with data stored on Hadoop
GPU acceleration for DL workloads
All popular ML/DL
frameworks
01
05
10
15
20
25
30
35
40
02
03
04
06
07
08
09
11
12
13
14
16
17
18
19
21
22
23
24
26
27
28
29
31
32
33
34
36
37
38
39
41
42
01
05
10
15
20
25
30
35
40
02
03
04
06
07
08
09
11
12
13
14
16
17
18
19
21
22
23
24
26
27
28
29
31
32
33
34
36
37
38
39
41
42
01
05
10
15
20
25
30
35
40
02
03
04
06
07
08
09
11
12
13
14
16
17
18
19
21
22
23
24
26
27
28
29
31
32
33
34
36
37
38
39
41
42
01
05
10
15
20
25
30
35
40
02
03
04
06
07
08
09
11
12
13
14
16
17
18
19
21
22
23
24
26
27
28
29
31
32
33
34
36
37
38
39
41
42
CISCO UCS-FI-6332
ENV
LS
STS
BCN
1
2
3
4
L1 L2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 31 3229 3027 28
CISCO UCS-FI-6332
ENV
LS
STS
BCN
1
2
3
4
L1 L2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 31 3229 3027 28
UCS
C480 M5
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
ReW ritabl eRECORDER
M U L T IDVD+ReWritable
S
X
X
UCS
C480 M5
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
ReW ritabl eRECORDER
M U L T IDVD+ReWritable
S
X
X
UCS
C480 M5
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
ReW ritabl eRECORDER
M U L T IDVD+ReWritable
S
X
X
UCS
C480 M5
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
XX
NVME SSD
800 GBNVMEHWH800
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X
2 TBHD2T7KL6GN
SATA HDD
X X
NVME SSD
800 GBNVMEHWH800
X
NVME SSD
800 GBNVMEHWH800
ReW ritabl eRECORDER
M U L T IDVD+ReWritable
S
X
X
2 x Cisco UCS 6332
Fabric Interconnect
16 x C240 M5
datanode
8 x C240 M5
datanode
4 x Cisco UCS Modoc each withCompute:
8x V100 Nvidia GPU
2x6132 CPU
Storage:
24 x 2.4 TB SFF drives
Networking:
2 x40G
GPU node + datanode
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Joint Solutions
Cisco Container
Platform on
HyperFlex
KubeFlow
ML/AI
Framework
References: https://cloud.google.com/blog/big-data/2018/03/simplifying-machine-learning-on-open-hybrid-clouds-with-kubeflow
https://blogs.cisco.com/datacenter/cisco-ucs-and-hyperflex-for-ai-ml-workloads-in-the-data-center
Cisco UCS and Cisco HyperFlex
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Cisco/Google Open Hybrid Cloud Solution
Governance
Cisco UCS
or
HyperFlex
CiscoCloudCenter
Cloud Platform
Kubernetes
Kubernetes
Engine
SecurityCisco
CSR1000vCisco
SteathWatch
Cluster Management
Cisco Container Platform
Networking
ContivCisco ACI
Storage
Cisco HyperFlex FlexVolume Driver
API / Mesh ManagementOpen
Service Broker
GA
Now
Kubeflow
Kubeflow
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Orchestrate AI Workloads with Kubernetes and Cisco HyperFlex
• Support for NVIDIA Tesla V100, the most advanced data center GPU ever built
• Leverage Kubernetes on NVIDIA GPUs to automate and orchestrate AI workloads on Cisco HyperFlex
• Maximize utilization across hyperconverged infrastructure
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Analytical Solutions with GPU Acceleration
GPU Accelerated Databases
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
UCS: One System for All Workloads
VDI, VSI, Distributed Databases, CI, Enterprise
Applications: Oracle and SAP HANA, Big Data and
Analytics, AI/ML with GPUs
Mainstream Workloads B200 M5
C220 M5 and C240 M5 HyperFlex
Cloud Computing, Microprocessor Design and Simulation,
Computational Fluid Dynamics (CFD),
High Frequency Trading (HFT,) Fraud Detection,
Online Gaming
Scale Out and Compute Intensive
Applications C4200 and C125
6400 Series Fabric Interconnects
AI/ML With Dense GPUs, and Memory-Intensive Mission-
Critical Enterprise Applications: In-Memory Databases
AI/ML and Scale up Workloads
C480 ML M5 and B480 M5
Data Protection, SDS, Scale-Out Unstructured Data
Repositories, Media Streaming, and Content Distribution
Data Intensive Workloads
S3260 M5
NEW
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Public
Visit Cisco Booth
• UCS C480 ML
• Machine learning solutions
on Cisco UCS
• GPU accelerated virtual
workspaces
• Quadro® Virtual DC
Workstations
For more info
http://cisco.com/go/ai-compute