Adopt Open AI, Compute, Storage and Total Solution
Alan ChangDeputy GM & Sr. Director of Server Product Line, Inspur
Path to Adoption
Marketplace
FSI CoSP
E&Ge-Commerce
San JoseGoose LakeJDM
Productize
CSP
Crane MountainOlympus StorageOAM AI NoteMission Bay
OAI
Rack and
Power
ODCC
ServerOpenRMC Storage
High Density Compute on Tencent Cloud
2U 4S OCP Compute Node
Jason WaxmanIntel CVP in 2019 OCP Global Summit
High Performance
Run computing-intensive services
High Frequency
Solve insufficient core density
High I/O
Reducecommunication delay
Powering Computing Services for a FinTech Cloud Customer
2 0 0 % p e r f o r m a n c e i n s a m e s p a c e
2 3 % c o s t r e d u x
c o m p a r e d t o 2 d u a l - s o c k e t s e r v e r s
4CPU & 48RAM
Customer Needs: • Larger capacity and higher
performance cloud DB• More VMs for cloud ECS
OCP Solution:• Inspur NF8260M5
Application:• Cloud ECS (high reliability,
manageability and flexibility)
SAP HANA Value Prop
7
With Intel® Optane™ Persistent Memory
Simpler Queries Lower Platform TCOLandscape Consolidation
Customer Data in 4 CPU Systems
Before & After Persistent Memory
DRAM used to store HANA Main Data
DRAM used for Working Memory
PMEM used to store HANA Main Data
HA
NA
Dat
a in
TB
0Haswell &Ivy Bridges
Broadwell Skylake Cascade Lake
3
4
6
1.5
9
3
3
2
2.1.5
1.53
3
Intel Persistent Memory
6
3
12
3
HANA Transactional Workloads4-CPU system: DRAM vs. Intel® Optane™ persistent memory DIMM
DRAM used for Working Memory PMEM used to store HANA Main DataDRAM used to store HANA Main Data
Total:3,072 GB
Storage
DRAM:3,072 GB
Total:4,608 GB
DRAM:1,536 GB
64 128
64 128
64
12864
128
64 128
64 128
64 128
64 128
64
12864
128
64 128
64 128
64 128
64 128
64
12864
128
64 128
64 128
64 128
64 128
64
12864
128
64 128
64 128
Persistent Memory:3,072 GB
Storage
64 64
64 64
64
6464
64
64 64
64 64
64 64
64 64
64
6464
64
64 64
64 64
64 64
64 64
64
6464
64
64 64
64 64
64 64
64 64
64
6464
64
64 64
64 64
Existing HANA 1 / HANA 2 Systems Future Landscape HANA 2 SP04+
Storage
Deliver More for Less
SAP HANA Use Case With Intel® Optane™ Persistent Memory
SaveMore
SimplifyConsolidate
MoreCapacity 100%
Lower Cost/System
Scale-out21 Node
Scale-up8 Node
40% + TCO Saving
Adopting OCP Universal Baseboard (UBB) and OAI (Open Accelerator Infra) Design
OCP CTO Bill Carter showing Inspur UBB in 2019 OCP Amsterdam Summit
Announced in OCP summit, Amsterdam
Inspur 21” OAM reference Platform
NF5468M6 6U 8x AI Acceleration Module
UBB Designed to Support Flexible Topology Designs
Consider a Grid of Planar OAM sites
Standard Volumetric
Protocol Agnostic Interconnects
Common Components and Connectors
OAMUBB
AI Use Case 1: Open Rack with Dual San Jose Nodes Connecting 8xOAMs
System Topology
4x Inspur 3OU OAM systems
8x San Jose Nodes
4x Inspur 3OU OAM systems
4x 4 SocketOlympus Nodes
AI Use Case 2: OCP Accepted JBOG and Compute NodesSupport retail / shopper analytics
Closed Division Times much improved
4x Inspur 3OU ON5388M5V100 SXM 2GPU Expansion
8x ON5263Head Nodes
Head Node Head Node
GPU 5 GPU 4
GPU 7 GPU 6
GPU 1 GPU 0
GPU 3 GPU 2
P20
P12
P8
P16
P0
P4PEX9707_B
P16
P8
P0
P12
P4 P20
PEX9707_A
P0
P12
P8
P16
P20
P4PEX9707_C
Up Stream Post Slot B2 Slot B1 Slot B0 Slot A2 Up Stream PostSlot A0 Slot A1
PCIe x16
NVLINK V100
NVLINK V100/P100
Open Storage Platform Collaboration with Samsung
Disaggregated and Shared
NVMeoF Interface Cascades with the JBOF in 1U Chassis
Contributing Hardware
Inspur Switches on SONiC
Join the SONiCcommunity
Contribute Hardware
Ongoing Commitment
25GE high-density TOR switch100GE Spine-Leaf switch