Collaboration
with your
Organization
Industry Centers of Excellence
Centers of Excellence for Innovation and Expertise
IBM/OpenPOWER is teaming with universities,
startups, ISV’s and industries to help develop
Several Centers of Excellence for Innovations and
developing expertise for real-world opportunities
AI & HPC Solutions development
Data Science Solution development
Industry 4.0 Solution development
3
Background and Motivation
The IBM CoE Labs will play a major role in the research and
development commercial and industrial development of
emerging technologies
There is a strong need for research and development activity
in these domains:
– Encouraging academic-industry partnerships
– Cross-disciplinary and collaborative research
– Making technology accessible to non-technical
business students
– Enabling faculty-technologist interaction and learning
– Enabling startups ,ISVs and industries to use the platform
to innovate in ways that improve the World condition
.
Technologies and Partners
4
5
AI TECHNOLOGIES curriculum
1
4
3
2
5
6
AI cloud setup and specifications - Hardware
College Ethernet Network
4
4
College Ethernet Network
2 IBM AC922 System
NVLINK-2 nVidia GPUs – 8
1 IC922 System with 2 T4 GPUs
OpenCAPI features
2 Raptor POWER 9 systems
The AC922 has 2 POWER9 sockets, each providing extreme levels of IO
and memory bandwidth. As an example, in the configuration proposed, each
socket will communicate with two Nvidia V100 GPUs directly utilizing the
300GB per second NVLink2 bus connections on each POWER9 socket. In
addition, sockets have high memory bandwidth, PCIe Gen4 bandwidth,
and a high bandwidth SMP interconnect.
7
The Power + GPU
advantages
8 Italy - 2019
– The AC922:
• Faster I/O - up to 5.6x more I/O bandwidth
than x86 servers
• The best GPUs - 2-6 NVIDIA® Tesla® V100
GPUs with NVLink
• Extraordinary CPUs - 2x POWER9 CPUs,
designed for AI
• Simplest AI architecture - Share RAM
across CPUs & GPUs
• Enterprise-ready - PowerAI DL frameworks
with IBM support
• Next Gen PCIe - PCIe Gen4
– 2x faster vs PCIe Gen3 in x86
• Built for the world's biggest AI challenges
– Summit #2 supercomputer!
AI on IBM Power Systems
The Power + t4 GPU advantages
9 Italy - 2019
– The IC922
– An accelerated inference server
engineered to put your AI models
to work
– Engineered for AI inference, IBM
Power System IC922 provides the
compute-intensive and low-latency
infrastructure needed to unlock
business insights from trained AI
models. POWER9-based, Power
IC922 provides advanced
interconnects (PCIe Gen4,
OpenCAPI) to support faster data
throughput and decreased latency.
Accelerated, the Power IC922
supports up to six NVIDIA® T4
GPUs.
AI on IBM Power Systems
Watson Machine
Learning
Community
Edition
Deep Learning Impact
(DLI) Module
Data & Model
Management, ETL,
Visualize, Advise
IBM Spectrum Conductor with Spark
Cluster Virtualization,
Dynamic Resource Orchestration,
Multiple Frameworks, Distributed Execution Engine
PowerAI base: Open Source ML Frameworks
Large Model Support (LMS)
Distributed Deep Learning
(DDL – 1000s of nodes)
Auto Hyper-parameter
Tuning
WML Accelerator
(formerly PowerAI
Enterprise)
PowerAI
Vision
Auto-DL for
Images &
Video
Accelerated
Infrastructure
Accelerated Servers Storage (Spectrum Scale) and Metadata
management/enhancement (Spectrum Discover)
AI for
Data Scientists and
non-Data Scientists
H2O
Driverless AI
Auto-ML for Text &
Numeric Data, NLP
Systems AI Portfolio in Perspective ( Software )
Distributed Deep Learning
(up to 4 nodes)
Watson Studio, Machine
Learning
Model management, large scale
projects, links to other analytics,
data catalog, data preparation
IBM Power/OpenPOWER System infra available
AI/HPC Systems for solutions development
IBM Power/OpenPOWER hardware for porting/evaluation/solution dev activities
§ Two IBM ppc64le systems with 2 K80 GPUs
§ Two IBM ppc64le systems with 4 V100 NVIDIA Volta GPUs with 32 GB RAM each
§ Two Raptor based POWER 9 System
§ One Mihawk POWER 9 server with OpenCAPI enabled as well as T4 GPUs
§ Voltar Intel Xeon CPU based system with v100 GPU for performance comparison between Power and x86
§ Mellanox 200 Gbps Connect-X 6 interconnect between nodes
§ 50 TB NFS
§ 256 GB RAM, 1.5 TB NVMe and 1.8 TB SSDs on each node
§ •Software: IBM Spectrum Scale MPI, MVAPICH2-GDR, XL Compilers, TAU, PAPI, GNU compilers, IBM PowerAI
Vision, AL and ML software stacks
11
Thank you!
Ganesan Narayanasamy
OpenPOWER global leader in Education &
Research
IBM Systems
12
Past years I have been enabling and engaging worldwide researchers , partners
and developers into our Disruptive Technologies and Open Technologies . What
excites me is receiving back insights on assignments/activities . According to me ,
One should never stop learning in life. The imagination of connected partners ,
researchers and students as they contemplate the applications and fusion of
emerging technologies in society and security is an inspiration.

AI/Cloud Technology access

  • 1.
  • 2.
    Centers of Excellencefor Innovation and Expertise IBM/OpenPOWER is teaming with universities, startups, ISV’s and industries to help develop Several Centers of Excellence for Innovations and developing expertise for real-world opportunities AI & HPC Solutions development Data Science Solution development Industry 4.0 Solution development
  • 3.
    3 Background and Motivation TheIBM CoE Labs will play a major role in the research and development commercial and industrial development of emerging technologies There is a strong need for research and development activity in these domains: – Encouraging academic-industry partnerships – Cross-disciplinary and collaborative research – Making technology accessible to non-technical business students – Enabling faculty-technologist interaction and learning – Enabling startups ,ISVs and industries to use the platform to innovate in ways that improve the World condition .
  • 4.
  • 5.
  • 6.
    6 AI cloud setupand specifications - Hardware College Ethernet Network 4 4 College Ethernet Network 2 IBM AC922 System NVLINK-2 nVidia GPUs – 8 1 IC922 System with 2 T4 GPUs OpenCAPI features 2 Raptor POWER 9 systems
  • 7.
    The AC922 has2 POWER9 sockets, each providing extreme levels of IO and memory bandwidth. As an example, in the configuration proposed, each socket will communicate with two Nvidia V100 GPUs directly utilizing the 300GB per second NVLink2 bus connections on each POWER9 socket. In addition, sockets have high memory bandwidth, PCIe Gen4 bandwidth, and a high bandwidth SMP interconnect. 7
  • 8.
    The Power +GPU advantages 8 Italy - 2019 – The AC922: • Faster I/O - up to 5.6x more I/O bandwidth than x86 servers • The best GPUs - 2-6 NVIDIA® Tesla® V100 GPUs with NVLink • Extraordinary CPUs - 2x POWER9 CPUs, designed for AI • Simplest AI architecture - Share RAM across CPUs & GPUs • Enterprise-ready - PowerAI DL frameworks with IBM support • Next Gen PCIe - PCIe Gen4 – 2x faster vs PCIe Gen3 in x86 • Built for the world's biggest AI challenges – Summit #2 supercomputer! AI on IBM Power Systems
  • 9.
    The Power +t4 GPU advantages 9 Italy - 2019 – The IC922 – An accelerated inference server engineered to put your AI models to work – Engineered for AI inference, IBM Power System IC922 provides the compute-intensive and low-latency infrastructure needed to unlock business insights from trained AI models. POWER9-based, Power IC922 provides advanced interconnects (PCIe Gen4, OpenCAPI) to support faster data throughput and decreased latency. Accelerated, the Power IC922 supports up to six NVIDIA® T4 GPUs. AI on IBM Power Systems
  • 10.
    Watson Machine Learning Community Edition Deep LearningImpact (DLI) Module Data & Model Management, ETL, Visualize, Advise IBM Spectrum Conductor with Spark Cluster Virtualization, Dynamic Resource Orchestration, Multiple Frameworks, Distributed Execution Engine PowerAI base: Open Source ML Frameworks Large Model Support (LMS) Distributed Deep Learning (DDL – 1000s of nodes) Auto Hyper-parameter Tuning WML Accelerator (formerly PowerAI Enterprise) PowerAI Vision Auto-DL for Images & Video Accelerated Infrastructure Accelerated Servers Storage (Spectrum Scale) and Metadata management/enhancement (Spectrum Discover) AI for Data Scientists and non-Data Scientists H2O Driverless AI Auto-ML for Text & Numeric Data, NLP Systems AI Portfolio in Perspective ( Software ) Distributed Deep Learning (up to 4 nodes) Watson Studio, Machine Learning Model management, large scale projects, links to other analytics, data catalog, data preparation
  • 11.
    IBM Power/OpenPOWER Systeminfra available AI/HPC Systems for solutions development IBM Power/OpenPOWER hardware for porting/evaluation/solution dev activities § Two IBM ppc64le systems with 2 K80 GPUs § Two IBM ppc64le systems with 4 V100 NVIDIA Volta GPUs with 32 GB RAM each § Two Raptor based POWER 9 System § One Mihawk POWER 9 server with OpenCAPI enabled as well as T4 GPUs § Voltar Intel Xeon CPU based system with v100 GPU for performance comparison between Power and x86 § Mellanox 200 Gbps Connect-X 6 interconnect between nodes § 50 TB NFS § 256 GB RAM, 1.5 TB NVMe and 1.8 TB SSDs on each node § •Software: IBM Spectrum Scale MPI, MVAPICH2-GDR, XL Compilers, TAU, PAPI, GNU compilers, IBM PowerAI Vision, AL and ML software stacks 11
  • 12.
    Thank you! Ganesan Narayanasamy OpenPOWERglobal leader in Education & Research IBM Systems 12 Past years I have been enabling and engaging worldwide researchers , partners and developers into our Disruptive Technologies and Open Technologies . What excites me is receiving back insights on assignments/activities . According to me , One should never stop learning in life. The imagination of connected partners , researchers and students as they contemplate the applications and fusion of emerging technologies in society and security is an inspiration.