Mais conteúdo relacionado Semelhante a Rightsizing Your Silicon Design Environment: Elastic Clusters for EDA Workloads (MFG401) - AWS re:Invent 2018 (20) Mais de Amazon Web Services (20) Rightsizing Your Silicon Design Environment: Elastic Clusters for EDA Workloads (MFG401) - AWS re:Invent 20182. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Rightsizing Your Silicon Design Environment:
Elastic Clusters for EDA Workloads
Kristopher King
Engineering Manager
AWS
M F G 4 0 1
3. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Workshop prerequisites:
1. Laptop(Windows orMacOS) withSSHorPuTTY accesstoAmazonElasticCompute Cloud
(AmazonEC2)Instances
2. AWSAccount –administrator accessincluding IAMandsuccessfully launched Non-free Tier
instance(s) (Non-T2/T3)
3. Checkinstance limits bygoing toConsole ->EC2->Limits->”Running On-Demand m4.large
instance”. Limitmust be20ormore
Toimmediately increaseyour instance limit,pleasesubmit your AWSAccount number through this
website: http://bit.ly/InstanceLimitI
Rightsizing Your Silicon Design Environment:
Elastic Clusters for EDA Workloads
4. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Agenda
Workshop—Provisioning
EDA Development on AWS Cloud
Internal AWS silicon development
AWS Managed Services for EDA
AWS EDA Infrastructure
Workshop
AWS credits will be provided
5. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Introductions and workshop setup
Workshop staff: Solution Architects and EDA Infrastructure engineers
Workshop overview
1. Provision an EDA infrastructure running in the cloud
2. Login a development desktop
3. Run simulations at scale
4. Explore design tools
6. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Provision an EDA infrastructure running in the cloud
Deploying the AWS CloudFormation templates will take around 20 minutes.
Please start the workshop:
http://bit.ly/EDAworkshop
Please complete workshop sections:
1. Introduction
2. Prerequisites
3. Deploy cluster
7. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
A complex industry …
Design and
verification
Wafer
production
Chip
packaging
Assembly Product
integration
Product
distribution
… with many opportunities for cloud-accelerated innovation
8. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
On premise EDA infrastructure
Shared file storage
Is cloud secure enough?
Capacity Planning
Throughput limited by compute and storage quota
Network Latency
How do we protect IP?
IOPSThird-party EDA licenses
Which servers should we buy? How many?
9. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Fabless Semiconductor
Company
• A semiconductor “branded” company that designs, but does not manufacture its own ICs
• Examples: Qualcomm, NVIDIA, Broadcom, Apple, Mediatek, Mellanox, Xilinx, Amazon
Semiconductor Foundry
• A highly automated facility that creates silicon wafers for use in IC products
• Or a company (“pure-play foundry”) that offers wafer foundry services
• Examples: TSMC, UMC, Global Foundries
Integrated
Semiconductor Company
• A semiconductor company that designs and manufactures its own ICs
• Examples: Intel, Samsung, SK Hynix, Micron
Electronic
Design Automation
• Software provided by “EDA Vendors” Synopsys, Cadence, Mentor, Ansys, and others
• Used to design, simulate, synthesize/compile, and verify integrated circuits (ICs)
• Very expensive software! Cost of EDA licenses is much higher than cost of underlying IT
Semiconductor
Intellectual
Property
• Silicon design components that are licensed for use by major semiconductor companies
• Major providers of IP include Arm, Synopsys, Cadence
CHIP = INTEGRATED CIRCUIT = SILICON DEVICESome vocabulary
10. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
11. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Development flow challenges
Design Specification
Design
Verification
Synthesis Physical Layout
Physical
Verification
Power/Signal
Analysis
Tape out/
Manufacturing
Silicon
Validation
12. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
EDA vendor and foundry support?
“The cloud is pervasive and will fundamentally influence silicon design. TSMC
is the first foundry to collaborate with design ecosystem partners and cloud providers to
enable designs in the cloud,” said Cliff Hou, vice president of Technology Development at
TSMC.
“
”
13. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Cloud products for Cadence customers
Cloud products today for the engineering challenges of tomorrow
Cloud-enabled products to run in your
cloud environment with Cloud Passport
CLOUD READY
Cloud-optimized products that run in a
fully supported and Cadence-managed,
ready-to-go cloud design environment.
HDS CLOUD
Cadence® HDS Cloud includes:
• Licensed software and support
• Cloud-optimized services
• CAD and IT infrastructure support
• PDK and foundry expertise
• Complete security support
Customer
managed
Cadence
managed
14. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Synopsys cloud solutions
▪ Cloud scaling for peak usage and full-flow workloads
▪ Robust environment to accelerate project cycle times
https://www.synopsys.com/solutions/cloud.html
15. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
16. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2017, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
C4
• Fabless semiconductor
company inside AWS
• Multiple internal start-ups
• Multiple product lines,
multiple sites
• Missionary, passionate, and
uniquely talented team
17. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Our own journey—Our own digital transformation
20152014
Hybrid Model:
Native AWS (New Projects)
Native AWS for overflow
(Existing Projects)
Shorten Si development
time
AWS
“One Team”
Acquisition of
Annapurna
On-prem data center
On-prem Labs
All-in on AWS:
Lift & shift architecture
Four different teams
multiple regions
Multiple end-to-end
silicon projects using AWS
2016 2017 Today
Multiple
Teams
Multiple
EDA
Environments
All-in on AWS:
Increased productivity via
native AWS services:
• Containers
• Batch
• Amazon EFS and
Amazon S3
• z1d, R5, C5, X1e
18. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
19. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Solution components for cost-optimizing EDA
Storage
Amazon EBS
Amazon EFS
Amazon S3
Networking
Enhanced
networking
Placement
groups
Automation &
Orchestration
AWS Batch
AWS ParallelCluster
NICE EnginFrame
Visualization
NICE DCV
Amazon
AppStream 2.0Compute
EC2 Instances
(CPU, GPU,
and FPGA)
EC2 Spot
AWS Auto Scaling
Partners
ISVs and Sis
IP Providers
20. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
General purpose
and burstable
Compute
optimized
Storage and
I/O optimized
Memory optimized GPU
graphics
GPU and
FPGA compute
AWS compute instances for EDA workloads
T2
G3
EG
G2
F1
P3
P2
M5
M4
I3 H1
D2C5
C4
Z1d X1
R5
Z1d
21. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
General purpose
and burstable
Compute
optimized
Storage and
I/O optimized
Memory optimized GPU
graphics
GPU and
FPGA compute
AWS compute instances for EDA workloads
T2
G3
EG
G2
F1
P3
P2
M5
M4
I3 H1
D2C5
C4
Z1d X1
R5
Z1d
License Servers Physical Synthesis
Formal Verification
RTL Simulations Power/
Timing/
DRC Analysis
Emulator Build
PCB Layout IP Emulation
22. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
z1d instances—Optimized for EDA
• Sustained all core frequency of up to 4.0 GHz
• Six different instance sizes with up to 48 vCPUs per instance
• Includes local NVME storage
• Optimized for EDA and other high performance workloads
• 30-40% saving in runtime over previous generation ➔ Productivity boost
Z1d
Model vCPU Memory (GiB) Instance storage (GiB) Networking performance EBS bandwidth
z1d.large 2 16 1 x 75 NVMe SSD Up to 10,000 Mbps Up to 2,333 Mbps
z1d.xlarge 4 32 1 x 150 NVMe SSD Up to 10,000 Mbps Up to 2,333 Mbps
z1d.2xlarge 8 64 1 x 300 NVMe SSD Up to 10,000 Mbps 2,333 Mbps
z1d.3xlarge 12 96 1 x 450 NVMe SSD Up to 10,000 Mbps 3,500 Mbps
z1d.6xlarge 24 192 1 x 900 NVMe SSD 10,000 Mbps 7,000 Mbps
z1d.12xlarge 48 384 2 x 900 NVMe SSD 25,000 Mbps 14,000 Mbps
23. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Optimize using EC2 pricing models
per-second billing
substantially less
?
discount
24. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Automation & orchestration: AWS Batch
• AWS Batch dynamically provisions resources
• Plans, schedules, and executes workload
• No batch software to install and manage
Focus on your applications and results!
25. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Storage: Amazon EBS/Amazon EFS/Amazon S3
Ephemeral/EBS EFS S3
26. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Visualization: Amazon WorkSpaces, nice DCV
Secure cloud Linux desktops for remote engineers
Efficient PC-over-IP (PCoIP) technology from Teradici
Clients for Windows, Mac, and mobile devices
MATE desktop environment on Amazon Linux 2
27. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
EDA White Paper bit.ly/aws-eda-whitepaper
28. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
29. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Workflow
AWS Cloud
Workshop Venue
Submit job Provision
capacity and
dispatch job
AWS Batch Amazon EC2
Instances
Amazon Elastic File
System
Internet
Remote Desktop
Server
AWS Directory
Service
Authenticate user
Log in
Access project data Access project and
scratch data
Lambda function
Amazon DynamoDB
Monitor job status
Update job status
Query job status
30. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Architecture AWS Cloud
AWS Region - us-west-2
Compute Instances
AWS Directory
Service
Availability zone 1
Public subnet
VPC
Workshop
Internet
AWS Batch
Amazon
Elastic File
System
VPC NAT
gateway
Internet
gateway
Private subnet
Remote
Desktop
Server
Amazon DynamoDB Lambda function
31. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Life of an EDA job
JOB
AWS Batch
JOB
Amazon EC2/ECS
SimulationSynthesisPlace and
Route
Containers
32. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
33. Thank you!
© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Kristopher King
Mark Duffield
Matt Morris
Allan Carter
Deep Patel
34. © 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.