Appro XtremeX Supercomputers APPRO INTERNATIONAL INC Company Overview

Appro Xtreme-X Supercomputers APPRO INTERNATIONAL INC

Company Overview : : Corporate Snapshot • Leading developer of high performance servers, clusters and supercomputers TX – Manufacturing Hardware in Asia – Global Presence via Strategic and Channel Partners – 72% Profitable CAGR over past 3 years APPRO HPC PRESENTATION SLIDE | 2 – Established in 1991 – Headquartered in Milpitas, CA – Sales & Service office in Houston,

Strategic Partnership APPRO HPC • NEC has a strong presence in the EMEA HPC Market with over 20 years of experience • This is a breakthrough for Appro’s entry into the EMEA HPC market • Provides sustainable competitive Formal Press Announcement will go advantages enabling both out on Tuesday, 9/16/08 companies to participate in this PRESENTATION SLIDE | 3 : : Appro & NEC Join Forces in HPC Market

HPC Experience : : Past Performance History • • • 2006. 9 2006. 11 NOAA Cluster LLNL Atlas Cluster 1, 424 Cores 2. 8 TB System Memory 15 TFlops • • • 9, 216 Cores 18. 4 TB System Memory 44 TFlops 2007. 6 2008. 2 LLNL Minos Cluster DE Shaw Research Cluster • • • 4, 608 Cores 9. 2 TB System Memory 49 TFlops SLIDE | 4 • 6, 912 Cores 13. 8 TB System Memory 33 TFlops APPRO HPC PRESENTATION

HPC Experience : : Past Performance History • • • 2008. 4 2008. 6 TLCC Cluster Tsukuba University Cluster 48, 384 Cores LLNL, LANL, SNL 426 TFlops • • • 10, 784 Cores Quad-rail IB 95 TFlops 2008. 7 2008. 8 Renault F 1 CFD Cluster LLNL Hera Cluster 4, 000 Cores Dual-rail IB 38 TFlops • • 13, 824 Cores 120 TFlops SLIDE | 5 • • • APPRO HPC PRESENTATION

HPC Challenges : : Changes in the Industry Petascale deployments (4000+ node deployment) – Balanced Systems (CPU/Memory/Network) – Scalability (SW & Network) – Reliability (Real RAS: Network, Node, SW) – Facilities (Space, Power & Cooling) • Integrated exotics (GPU cluster) – Solutions still being evaluated SLIDE | 6 • APPRO HPC PRESENTATION

: : Based on a Scalable Multi-Tier Architecture : Infini. Band for Computing : 10 Gb. E Operation : Gb. E Operation Network (10 Gb. E) External Network Controllers Storage Firewall Router FC or Gb. E Management Bridge or Servers 4 X IB Controllers Storage Parallel File System 4 x IB Node Node Compute Node Compute Compute (Gb. E) Network Operation 2 x Gb. E per node (Gb. E) Network Operation Group Server Compute Node I/O IO Node I/O IO Infini. Band Network Gb. E or 10 Gb. E APPRO HPC 2 x 10 Gb. E PRESENTATION 2 x Gb. E Node I/O IO SLIDE | 7 2 x 10 Gb. E

Petascale Deployments : : Scalable cluster management software S t a t e l e s s Op e r a t i o n Mi d d l e W a r e - H o o k s “Appr Virtual Cluster Manoager Cluste r Remote Lights out Manage. Engin m e nt IB-Subnet Manae™ ge r softw are turns a cluste r of Serve rs into a, ” functi onal, usabl e, reliabl APPRO HPC PRESENTATION A CE B I OS S y n c h r o n i z a t i o I ns t a nt S W P r o v an ild ov SFta ae rdr & Lin. Ruexco. Ov. Ser. Syu SLIDE | 8 J o b S c he d ul i ng D u a l Ra i l N e t w o r k s

Petascale Deployments : : Innovative Cooling and Density needed Top View • Up to 30% Improvement in Density with Greater Cooling Efficiency • Delivers Cold Air directly to the equipment for optimum cooling efficiency. • Delivers comfortable air temperature to the room for return to Chillers encloses the Cold isles inside the racks • FRU and maintenance is done from the front side of the rack cabinet APPRO HPC PRESENTATION SLIDE | 9 • Back-to-Back Rack configuration saves floor space in the datacenter and

Petascale Deployments : : Path to Peta. FLOP Computing Appro Xtreme-X Supercomputer - Modular Scalable Performance Number of Racks 1 2 8 48 96 192 Number of Processors 128 256 1024 5, 952 11, 904 23, 808 Number of Cores 512 1, 024 4096 23, 808 47, 616 95, 232 Peak Performance 6 TF/s 12 TF/s 49 TF/s 279 TF/s 558 TF/s 1. 1 PF/s Memory Capacity 1. 5 TB 3 TB 12 TB 72 TB 143 TB 286 TB Memory BW Ration GB/s per GF/s - 0. 68 GB/s per GF/s Memory Capacity Ratio GB per GF/s - 0. 26 GB per GF/s IO Fabric Interconnect – Dual-Rail QDR IO BW Ratio GB/sec per GF/s - 0. 17 GB/s per GF/s Usable Node-Node BW GB/s - 6. 4 GB/s Node-Node Latency - <2 us Performance Numbers are Based on 2. 93 GHz Intel Nehalem Processors and Includes only Compute Nodes SLIDE | 10 . . APPRO HPC PRESENTATION

Xtreme-X Supercomputer : : Possible Path to Peta. FLOP GPU Computing Cluster – Solution still being evaluated Number of Racks 3 5 10 18 34 Number of Blades 64 128 256 512 1024 Number of GPUs 32 64 128 256 512 Peak GPU Performance 128 TF 256 TF 512 TF 1 PF 2 PF Peak CPU Performance 6 TF 12 TF 24 TF 48 TF 96 TF Max Memory Capacity 1. 6 TB 3. 2 TB 6. 4 TB 13 TB 26 TB Bandwidth to GPU – 6. 4 GB/sec Node Memory Bandwidth – 32 GB/sec Max IO Bandwidth (2 QDR X 4 IB) - 6. 4 GB/sec Node to Node Latency – 2 us SLIDE | 11 . . APPRO HPC PRESENTATION

Appro Xtreme-X Supercomputers Thank you Questions? APPRO INTERNATIONAL INC