The Future of Intel Optane™ Persistent Memory€¦ · Baseline: 1 -node, 1x Intel® Xeon® 8280L...

The Future of Intel® Optane™ Persistent MemoryAlper Ilkbahar

Vice President, Data Platforms Group & General Manager, Intel Optane Group

Intel Corporation

2Memory & Storage Moment 2020 Under Embargo until December 16, 2020, 6:00 AM PST

Data is Fueling Digital Transformation

175 ZB globally by 20251 Compound annual growth rate (CAGR)2 in data26%

CAGR

Deliverbusiness valuein real time

Minimize risk andincrease future opportunities

Highly parallelprocessingon largedatasets

Low latency for mission-criticaltransactions

New capabilities from transformation

Artificial intelligence (AI) and machine learning (ML)

Real-timeanalytics

Predictiveanalytics

Businessintelligence

Data security

Massivedatasets

Transactionalprocessing

Growing and new workloads that are driving transformation

1 IDC. “The Digitization of the World From Edge to Core.” Sponsored by Seagate. November 2018. seagate.com/files/www-content/our-story/trends/files/idc-seagate-dataage-whitepaper.pdf.2 IBM. “3D NAND Technology – Implications to Enterprise Storage Applications.” 2015 Flash Memory Summit. 2014. flashmemorysummit.com/English/Collaterals/Proceedings/2015/20150811_FM12_Yoon.pdf.

https://seagate.com/files/www-content/our-story/trends/files/idc-seagate-dataage-whitepaper.pdfhttps://www.flashmemorysummit.com/English/Collaterals/Proceedings/2015/20150811_FM12_Yoon.pdf


Compute Demand Accelerating1

Den

sity

DRAM is Not Scaling2

4Mb

16Gb

1Mb

64Mb

16Mb

128Mb256Mb

512Mb1Gb

2Gb4Gb

8Gb

Phase 3Density increase~2X/4years

Phase 2Density increase~2X/2years

1980 1985 1995 2000 2005 2010 2015 2020

Density

Phase 1 Density increase~4X/3years

Gap Between Data and Memory Capacity is Increasing3

Cap

acity

Sca

ling

(nor

mal

ized

)

10

9

8

7

6

5

4

3

2

1

01 2 3 4 5 6 7 8 9 10

DRAM ScalingPhase 32X/4yr

Memory NeedGrowth G

AP

The data economy is fundamentally changing the workloads and infrastructure in the data center

Adapting to the Changing Data Landscape

1. Source for “Compute Demand Accelerating” Intel Market Intelligence Group2. Source for “DRAM is Not Scaling”: 3D NAND Technology – Implications for Enterprise Storage Applications” by J.Yoon (IBM), 2015 Flash Memory Summit3. Source: Data Age 2025, sponsored by Seagate with data from IDC Global DataSphere, Nov 2018


Memory

Storage

Performance Tier: Capacity Tier:

DRAM

SSD Storage

Memory Bus

PCIe Bus

Intel® Optane™persistent memory

Intel® Optane™persistent memory

Vision of 2-Tier Memory / Storage for PMem

Intel® Optane™ persistent memory can help memory and storage applications!

DRAMHOT TIER

3D NAND SSD

HDD/TAPECOLD TIER


More to be Gained by Being on Memory Bus

Average Lightly Loaded Random Read Latency1

0

25

50

75

100

Storage With Intel® 3D NAND SSD

P5600

Storage with Intel® Optane™ SSD

P5800X w/SPDK

Hardware latency Software latency

Memory Subsystem with Intel® Optane™

persistent memory 200 Series

Average~6µsfor 4kB

Memory SubsystemAverage

~100ns to ~340nsfor 64B2

Performance results are based on testing as of September 25, 2020. Performance varies by use, configuration and other factors, for details, see Slide 12. Learn more at www.Intel.com/PerformanceIndex.

http://www.intel.com/PerformanceIndex%E2%80%8B


A Strong Memory and Storage Future

3rd Generation Intel Xeon

Scalable processors(Ice Lake)

Intel Optane Persistent Memory

200 series

Intel® Optane™ Persistent Memory

100 series

2nd Generation Intel® Xeon® Scalable

processors

3rd Generation Intel Xeon

Scalable processors(Cooper Lake)

Next GenerationIntel Optane

Persistent Memory(Crow Pass)

Next Generation Intel Optane

Persistent Memory

Future2019 2020 2021

Future Intel Xeon

Scalable processors(Sapphire Rapids)

FutureIntel Xeon

Scalable processors


Affordable memory capacity for many applications

Maximize capacity with new persistent memory tier

Intel® Optane™ Persistent Memory (PMem) 200 Series

Intel® Optane™ Persistent Memory – a Solution on a DIMM

Supporting 3rd Gen Intel® Xeon® Scalable processors

6 TB Total memory per socketup to

3 W Typical power reduction, watts per module2

approx.

Support toimprove appperformance

with eADR

1 Performance results are based on testing as of March 31, 2020. Performance varies by use, configuration and other factors, for details, see Slide 12. 2 Source: Intel® Optane™ Persistent Memory Series 200 Data Sheet, https://www.intel.com/content/www/us/en/products/docs/memory-storage/optane-persistent-memory/optane-persistent-memory-200-series-brief.html Learn more at www.Intel.com/PerformanceIndex.

App Direct Mode

PMem Aware Software

PMem Aware OS

Intel Xeon Scalable Processors

Memory Controller

Cores

DRAM Memory Persistent Memory

DRAM

Memory Mode

Unmodified Application

Unmodified OS/VMM

Intel Xeon Scalable Processors

Memory Controller

Cores

DRAM as Cache Volatile Memory

DRAM

25% Higher memory bandwidth vsprior generation1

avg.

https://www.intel.com/content/www/us/en/products/docs/memory-storage/optane-persistent-memory/optane-persistent-memory-200-series-brief.htmlhttp://www.intel.com/PerformanceIndex%E2%80%8B


Enhanced Asynchronous DRAM Refresh (eADR)

System Level Optimization

CPU

eADR automatically flushes caches eliminating CLFlush runtime use/overhead

(Requires: Barlow Pass, Energy Store, App support)

MLC/LLC

IIOMemory

Controller

eADR Persistency Visibility ADR Option ADR Persistency Visibility

Persist the CPU caches No cache flush; stores are considered persistent as soon as they are visible Free up resources, avoid waiting for flushes to complete

Applications with mapped data in PMem will benefit Databases Transactional processes Mission critical data handling High performance apps using lock-free or non-blocking algorithms

Immediately available with PMDK

Available on Intel® Optane™ persistent memory 200 series with 3rd Gen Intel® Xeon® Scalable processors


A Growing Ecosystem for Intel® Optane™ Technology

Other names and brands may be claimed as the property of others.

OEMs & SIs

CSPs & COSPS

Software


Intel® Optane™ Persistent MemoryDelivering Real World Benefits

22.5%-48% improvementin TCO (Redis)

~2XVM instantiation for 5G

Multi-access Edge (Redis)

13.7X accelerated database startup

(SAP HANA)

30% reductionin recommendation system

& Redis service

2.78X increasein games hosted on a

single server

80% latency reduction &3X accelerated indexing

(Elasticsearch)

41% reductionon infrastructure cost

~40% more VMs & containers within

same budget (Redis)

up to 17X fasterstorage applications

(RocksDB, MongoDB, mySQL)

15X reductionin memory for image processing

1.1X increasein CPU utilization

vs. DRAM-only

15X fasterdatabase data load startup

(SAP HANA)

1.3X improvementin TCO (Redis)

~3X improvementin jobs per physical host ratio

8X faster solver run compared to

Lustre filesystemStudioCloud

TCO Savings Increased Throughput Faster Time to Insights

Over 200Fortune 500

Over 85%POC to Sale Conversion

Over 475Production Wins

Customer TractionSince Launch

Performance results are based on testing as of dates in configuration. Performance varies by use, configuration and other factors, for details, see Slide 13 & Slide 14. Learn more at www.Intel.com/PerformanceIndex.

http://www.intel.com/PerformanceIndex%E2%80%8B


Legal Notices and DisclaimersAll product plans and roadmaps are subject to change without notice.

Intel optimizations, for Intel compilers or other products, may not optimize to the same degree for non-Intel products.

Intel technologies may require enabled hardware, software, or service activation.

Performance varies by use, configuration, and other factors. Learn more at www.intel.com/PerformanceIndex.

Performance results are based on testing as of dates shown in the configurations and may not reflect all publicly available security updates. See backup for configuration details. No product or component can be absolutely secure.

Your costs and results may vary.

Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.

Results have been estimated or simulated.

© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.

http://www.intel.com/PerformanceIndex


P5800X 1.6TB: CPU: Intel® Xeon® Gold 6254 3.10GHz 30MB 160W 18 cores per socket, CPU Sockets: 2, BIOS: SE5C620.86B.02.01.0009.092820190230, RAM Capacity: 32G, RAM Model: DDR4, RAM Stuffing: NA, DIMM Slots Populated: 4 slots, PCIe Attach: CPU (not PCH lane attach), Chipset: Intel C610 chipset, Switch/ReTimer Model/Vendor: Intel G4SAC switch (PCIe Gen4), OS: CentOS 7.5.1804, Kernel: 4.14.74, FIO version: 3.5; NVMe Driver: Inbox, C-states: Disabled, Hyper Threading: Disabled, CPU Governor (through OS): Performance Mode. EIST (Speed Step), Intel Turbo Mode, and P-states = Disabled; IRQ Balancing Services (OS) = Off; SMP Affinity, set in the OS; QD1 utilizes I/O Polling Mode with ioengine=pvsync2/hipri. Tested by Intel on 9/25/20.

P5600 6.4TB: CPU: Intel® Xeon® Gold 6139 2.30GHz, BIOS: SE5C620.86B.00.01.0014.070920180847 (Intel Server Board S2600WFT), CPU Sockets: 2, RAM Capacity: 32G, RAM Model: DDR4-2137, RAM Stuffing: 1 of 4 channels, DIMM Slots Populated: 4 slots, PCIe Attach: CPU (not PCH lane attach), Chipset: Intel C612 chipset, Switch/ReTimer Model/Vendor: Intel G4SAC switch (PCIe Gen4), NVMe Driver: Kernel 4.17.74 (native), C-states: Disabled, Hyper Threading: Disabled, CPU Governor (through OS): Performance Mode, OS: CentOS 7.5, Kernel: 4.14.74, FIO version: 3.5. Tested by Intel on 3/16/20.

Intel Optane PMem 100 Series - Tested by Intel on single DIMM configuration; Test date 09/25/2020. Intel® 2nd Gen Xeon® Scalable Processor Platform ; Chipset LBG B1; CPU CLX QRL1 B0 28 Core (8276), 1S; DDR speed 2666 MT/s; Intel Optane PMEM 256GB module,15W Memory configuration 1 channel, 32GB DDR4 (6 per socket); Intel Optane PMEM FW 1.2.0.5435; BIOS 0601.D02; BKC version WW02 BKC, Linux OS Fedora release 31 5.5.7-200.fc31; Spectre/Meltdown patched (1,2,3,3a); Performance Tuning QoS Disabled IODC=5(AD). Intel® Optane PMem uses DDRT memory interface and MLC for performance measurements.

Intel Optane PMem 200 Series - Tested by Intel on single DIMM configuration; Test date 09/25/2020. Intel® 3rd Gen Xeon® Scalable Processor Platform; Chipset LBG B1; CPU CPX-6 QUC9 26 Core, 1S; DDR speed 2666 MT/s; Intel Optane PMEM 256GB module, 15W Memory configuration 1 channel, 32GB DDR4 (6 per socket); Intel Optane PMEM FW 2.2.0.1516; BIOS 0017.P23; BKC version WW38 BKC, Linux OS Fedora release 29 4.20.6-200.fc29.x86_64; Spectre/Meltdown patched (1,2,3a,4); Performance Tuning QoS Disabled IODC=5(AD). Intel® Optane PMem uses DDRT memory interface and MLC for performance measurements.

Average of 25% higher memory bandwidth vs prior gen:

Baseline: 1-node, 1x Intel® Xeon® 8280L 28C @ 2.7GHz processor on Neon City with Single PMem module config (6x32GB DRAM; 1x{128GB,256GB,512GB} Intel Optane PMem 100 Series module at 15W) ucode Rev: 04002F00 running Fedora 29 kernel 5.1.18-200.fc29.x86_64, and MLC ver 3.8 with App-Direct. Source: 2020ww18_CPX_BPS_DI. Tested by Intel, on 27 Apr 2020. New configuration: 1-node, 1x Intel® Xeon® pre-production CPX6 28C @ 2.9GHz processor on Cooper City with Single PMem module config (6x32GB DRAM; 1x{128GB,256GB,512GB} Intel Optane PMem 200 Series module at 15W), ucode pre-production running Fedora 29 kernel 5.1.18-200.fc29.x86_64, and MLC ver 3.8 with App-Direct. Source: 2020ww18_CPX_BPS_BG. Tested by Intel, on 31 Mar 2020.

Typical power reduction of approximately 3 Watts per module:

Intel® Optane™ Persistent Memory Series 200 Data Sheet, https://www.intel.com/content/www/us/en/products/docs/memory-storage/optane-persistent-memory/optane-persistent-memory-200-series-brief.html

Configuration details –Intel® Optane™ P5800x, Intel® SSD D7-P5600, Intel® Optane™ PMem

https://www.intel.com/content/www/us/en/products/docs/memory-storage/optane-persistent-memory/optane-persistent-memory-200-series-brief.html


Intel Optane Persistent Memory: Delivering Real World Benefits

Kingsoft Cloud REDIS service* (self-defined workload); OS: Red Hat Enterprise Linux* 7.5 4.18.8-x86_64. Testing by Intel and Kingsoft Cloud completed on Jan 10, 2019. Security Mitigations for Variants 1, 2, 3 and L1TF in place. BASELINE: 2nd Gen Intel® Xeon® Platinum 8260 processor, 2.3 GHz, 24 cores, turbo, and HT on, BIOS 1.018, 1536GB total memory, 12 slots / 64GB / 2666 MT/s / DDR4 LRDIMM, 1 x 480GB / Intel® SSD DC S4500 + 1 x 1TB / Intel® SSD DC P4500. NEW: 2nd Gen Intel® Xeon® Platinum 8260 processor, 2.3GHz, 24 cores, turbo and HT on, BIOS 1.018, 1536GB total memory, 12 slots / 16GB / 2933 MT/s / DDR4 LRDIMM and 12 slo/ 128 GB / Intel® Optane™ DC persistent memory, 1 x 480GB / Intel® SSD DC S4500 + 1 x 1TB / Intel® SSD DC P4500. For more complete information about performance and benchmark results, visit: https://www.intel.com/content/www/us/en/processors/xeon/scalable/software-solutions/kingsoft-cloud-redis-service.html.Kuaishou Technolgy: Test results are based on Kuaishou’s internal tests and evaluation. For more details, please contact Kuaishou https://www.intel.cn/content/www/cn/zh/architecture-and-technology/kuaishou-recommendation-system-and-redis-services-storage-upgrade.html.Max Planck: 2x 6248 CPUs with 2-2-2 128GB Apache Pass modules configured in memory mode. 32GBx12 DDR4 2666MHz RAM, CentOS* 7.6. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/max-planck-institute-customer-story.html.Ping An Cloud Total Cost of Ownership (TCO): This cost reduction data is derived from the joint calculation by of Ping An Cloud and Intel. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/ping-an-cloud-customer-story.html.SoftBank Results of the validation at SoftBank: Intel® Xeon® Silver 4114 processor: 40 cores with Intel® Hyper-Threading Technology enabled, 512 GB 1VM resource at 30VM capacity: 1.3 cores, 17.0 GB. Intel® Xeon® Gold 6222V processor: 80 cores with Intel® Hyper-Threading Technology enabled, 1536 GB 1VM resource at 60VM capacity: 1.3 cores, 25.6 GB. The information was described as of 13th December 2019. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/ping-an-cloud-customer-story.html.CDW Canada StudioCloud. Test results are based on StudioCloud internal tests and evaluation as of December 2019. For more complete information about performance and benchmark results, visithttps://www.cdw.ca/content/cdwca/en/industries/studiocloud.html.GPORTAL: 1 Baseline Configuration: Dell EMC PowerEdge R640 server; 2x Intel® Xeon® Gold 6154 processor @ 3.0 GHz (18 cores/36 threads); 768 GB DDR4; BIOS = 2.3.10; OS = Linux Results: 180 Minecraft game instances DUT Configuration: Dell EMC PowerEdge R640 server; 2x Intel® Xeon® Platinum 8268 processor @ 2.90 GHz (24 cores/48 threads); 12 x 32 GB DDR4 + 12 x 128 GB Intel® Optane™ DC persistent memory modules; BIOS = 2.3.10; OS = Linux Results: 500 Minecraft game instances. Testing by GPORTAL as of 5 December 2019.Nitrado: Testing by Nitrado as of February 7, 2019. All-DRAM configuration: Dual-socket Intel® Xeon® Gold 6148 processor (8x 64 GB DDR4-2666 DRAM), total memory installed = 512 GB. System memory available = 512 GB. Number of Minecraft* instances: 182. CPU utilization: 40%. DRAM + Intel® Optane™ DC persistent memory configuration: Dual-socket Intel® Xeon® Gold 6252 processor (12x 128 GB (1.5TB) Intel® Optane™ DC persistent memory plus 12x 16 GB (192 GB) DDR4-2600 DRAM), total memory installed = 1,692 GB. System memory available = 1,536 GB. Number of Minecraft instances: 500. CPU utilization: 85%. Final results were extrapolated from Nitrado’s testing data. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/nitrado-online-gaming-customer-story.html.SK Telecom: Testing conducted by SKT and Intel as of June 7, 2019. For more complete information about performance and benchmark results, visit https://builders.intel.com/docs/networkbuilders/case-study-of-scaled-up-skt-5g-mec-reference-architecture.pdf.ZTO Express: For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/zto-express-improves-infrastructure-video.html.EPCC: Performance results provided by EPCC and may not reflect all released security updates. No product can provide absolute security. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/edinburgh-parallel-computing-center-customer-story.html.phoenixNAP and Panzura configurations: Up to 3x indexing and 80% cache latency decrease – based on phoenixNAP and Panzura testing as of March 2019 on Elasticsearch: Intel® Xeon® Gold 6230 processor, Total Memory 256 GB RAM, 1.5TB of Intel® Optane™ DC persistent memory, HyperThreading: Enabled, Turbo: Enabled, ucode: 0x043, OS: (‘centos-release-7-5.1804.el7.centos.x86_64’), Kernel: (3.10.0-862) vs. AWS i3xlarge (Intel) Instance, Elasticsearch, Memory: 30.5GB, Hypervisor: KVM, Storage Type: EBS Optimized, Disk Volume: 160GB, Total Storage: 960GB, Elasticsearch version: 6.3. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/phoenixnap-panzura-customer-story.html.


Intel Optane Persistent Memory: Delivering Real World Benefits (CONT)

Siemens: 15X faster database data load at startup performance results are based Siemens testing in April 2019 and may not reflect all released security updates. No product can provide absolute security. For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/siemens-in-memory-processing-customer-story.html.

T-Systems: Testing by T-Systems as of March 18, 2019. Baseline Configuration Hardware: HPE Superdome Flex* server with 4x CPU sockets (Intel® Xeon® Platinum processor Beta 8276M 2.20 GHz ; Memory = 4x6 256 GB Intel® Optane™ DC persistent memory (6 TB) - DEACTIVATED and 4x6 64 GB DDR4 Memory (1.5 TB) for a total memory configuration of 1.5 TB Software: Database: 4 TB SAP S/4HANA* database in App Direct Mode; OS: Standard SUSE Linux Enterprise Server* 12 Service Pack 4 microcode = 0xb00002e; kernel = Linux 4.12.14-95.16, standard NetApp cDot*-based storage used for persistence; SAP HANA 2.0 SPS4 rev. 40 installation with BW-Benchmark workload Re-start time: 10, 248 seconds (approximately 2.85 hours) Proof of Concept Configuration Hardware: HPE Superdome Flex* server with 4x CPU sockets (Intel® Xeon® Platinum processor Beta 8276M 2.20 GHz ; Memory = 4x6 256 GB Intel® Optane™ DC persistent memory (6 TB) and 4x6 64 GB DDR4 Memory (1.5 TB) for a total memory configuration of 7.5 TB Software: Database: 4 TB SAP S/4HANA* database in App Direct Mode; OS: Standard SUSE Linux Enterprise Server* 12 Service Pack 4 microcode = 0xb00002e; kernel = Linux 4.12.14-95.16, standard NetApp cDot*-based storage used for persistence; SAP HANA 2.0 SPS4 rev. 40 installation with BW-Benchmark workload Re-start Time: 748 seconds (approximately 12.47 minutes). For more complete information about performance and benchmark results, visit https://www.intel.com/content/www/us/en/customer-spotlight/stories/t-systems-in-memory-database-customer-story.html.

UC San Diego: Testing conducted by UC San Diego as of August 8, 2019. For more complete information about performance and benchmark results, visit https://arxiv.org/pdf/1903.05714.pdf.

The Future of �Intel® Optane™ �Persistent MemoryData is Fueling Digital TransformationAdapting to the Changing Data LandscapeVision of 2-Tier Memory / Storage for PMem More to be Gained by Being on Memory BusA Strong Memory and Storage FutureIntel® Optane™ Persistent Memory (PMem) 200 SeriesEnhanced Asynchronous DRAM Refresh (eADR)A Growing Ecosystem for Intel® Optane™ TechnologyIntel® Optane™ Persistent Memory�Delivering Real World BenefitsLegal Notices and DisclaimersSlide Number 12Slide Number 13Slide Number 14Slide Number 15

The Future of Intel Optane™ Persistent Memory€¦ · Baseline: 1 -node, 1x Intel® Xeon® 8280L...

Documents

Transcript of The Future of Intel Optane™ Persistent Memory€¦ · Baseline: 1 -node, 1x Intel® Xeon® 8280L...