Changing the Game with Gen Z Kurtis Bowman

Changing the Game with Gen -Z Kurtis Bowman President, Gen-Z Consortium 4 -9 -2019 12/6/2020 Gen-Z Consortium

About The Gen-Z Consortium • The Gen-Z Consortium launched in October of 2016 to create an open, industry standard for a high speed, low latency, scalable, memory centric fabric • A growing group with 60+ member companies covering all of the disciplines required to create an ecosystem based on our standard • Gen-Z is a modern fabric specifically architected to increase performance, enhance security, spur innovation, and enable a memory centric model for cloud, datacenter, and edge environments • Gen-Z is the fabric needed to usher in full composability • Released and selected draft specifications are available on www. Gen. ZConsortium. org for public review and comment Gen-Z Consortium

Open Consortium With Broad Industry Support Consortium Members System OEM Cisco Cray Dell EMC H 3 C Hitachi CPU/Accel AMD Arm IBM Qualcomm Xilinx Mem/Storage Everspin Micron Samsung Seagate SK hynix Silicon Broadcom IDT Marvell Mellanox Microsemi IP Avery Cadence Intelliprop Mentor Mobiveil Connect Aces AMP FIT Genesis Jess Link Software Redhat VMware Govt/Univ ETRI HP Smart Modular Sony Semi PLDA Lotes Oak Ridge HPE Spintransfer Synopsys Luxshare Simula Huawei Lenovo Net. App Toshiba WD Eco/Test Allion Labs Molex Samtec Senko UNH Yonsei U ITT Madras Nokia Yadro Technology SP Google Keysight Teledyne Le. Croy TE 3 M Triad Microsoft Node Haven 12/6/2020 Gen-Z Consortium 3

Insights From Data: A Problem To Solve Explosive Growth of Data Need Answers … FAST! 2015 2005 0. 1 ZB 2010 1. 2 ZB 2012 2020 40 ZB 2. 8 ZB Value of Analyzed Data $ 8. 5 ZB 10 -2 More than 37% of total data generated in 2020 (40 ZB) will have significant business value! Gen-Z Consortium 100 102 104 Time to Result 106 Seconds Increasing amounts of data to be analyzed But businesses demanding real-time insight! 4

Gen-Z In A Nutshell FPGA SCM DRAM Provides strong hardware-enforced isolation and security So. C GPU – So. C Accelerator Pools SCM No stranded resources or single-point-of-failures Transparently bypass path and component failure Enables highly-resilient data (e. g. , RAID / erasure codes) DRAM – – – • Reliable Server FPGA High Bandwidth, Low Latency, Scalable Eliminates protocol translation cost / complexity / latency Eliminates software complexity / overhead / latency SCM – – – Gen-Z speaks the language of compute DRAM • High Performance GPU • Secure Gen-Z • Compatible – Use existing physical layers, no OS modifications required – – – Multiple topologies, component types, etc. Supports multiple use cases using simple to robust designs Thorough yet easily extensible architecture Point-to-point or switch-based, Any Topology • Flexible DRAM SCM DRAM Memory Pools • Economic • Lowers CAPEX / OPEX, unlocks / accelerates innovation Gen-Z Consortium SCM Network Storage I/O Pools

Gen-Z Encourages Innovation Media Module Gen-Z Memory. Fabric Bus Semantic DRAM Gen-Z Logic Media Controller Gen-Z Logic Processor DRAM 2 -84 -8 High-speed Serial Links Memory Channels Low Latency, High performance 17 -25 GB/s / Channel Split 288 pins Memory Controller / DIMM Asynchronous Interface Synchronous Interface Processor is media agnostic Gen-Z Consortium

Gen-Z Is Designed To Be Secure Access Keys Link & device partitioning Similar to VLANs or Zones Enforced by Gen-Z switches Memory Region Keys Crypto Authentication Component memory partitioning Enables multiple requesters Enforced by target component Crypto-signed packets (HMAC) Multi-tenant, Gov, Financial Enforced by target component Anti-Replay Tags Prevents “in-the-middle” attacks HMAC protected sequence #’s Enforced by target component Dropped packets are reported to fabric management (to identify misconfigurations or malicious attacks) Gen-Z Consortium 9

Next Generation Composability CPU CPU (Large) CPU CPU (Large) (Small) Accel- Accel. CPU (Small) Accelerator Networks Network e. NIC Gateway Gen-Z Fabric NVMz Disk Drive DRAM SCM Disk Drive CPU – Or subparts or subregions of components (e. g. memory/storage) CPU – Logical systems match exact workload requirements – No stranded resources overprovisioned to workloads CPU CPU Eliminates data movement: Do more with less, reduces cost e. NIC CPU – Facilitates data-centric computing via shared memory 12/6/2020 Disk Drive SCM – Logical systems composed of physical components – Disk Drive Msg Logical Web Server © Gen-Z Consortium 2019 Msg Logical Database Server Msg DRAM Storage Gateway Logical Analytics Server Key-Value Store 10

Gen-Z Enables Composability Compose servers with resources app requires INCREASE AGILITY Increase RAS Unlock trapped resources Avoid overprovisioning OPERATE EFFICIENTLY Repurpose retired resources Gen-Z Consortium Purchase resources independently UNLOCK VALUE Technologies can evolve – and deployed independently

Gen-Z Resources • Gen-Z demos at DCD NY 2019 • Gen-Z demos at FMS 2019 (August) • Gen-Z demos at Super Computing 2019 (November) linkedin. com/ company/genz-consortium/ Gen-Z educational material, membership details and links to related information at www. Gen. ZConsortium. org Twitter. com/ Gen. ZConsortium Gen-Z Consortium Channel