z16 Technical Overview 50M KennyStine
z16 Technical Overview 50M KennyStine
Kenny Stine
IBM zSystems Technical Specialist
Washington System Center
kjstine@us.ibm.com
Not all common law marks used by IBM are listed on this page. Failure of a mark to appear does not mean that IBM does not use the mark nor does it mean that the product is not actively marketed or is not
significant within its relevant market.
Those trademarks followed by ® are registered trademarks of IBM in the United States; all others are trademarks or common law marks of IBM in the United States.
• IBM z16 to IBM z16 all remaining Loose Piece MES orders - September 30 2022
IBM z16
Design Principles
GHz / PCI*
4.4 GHz
215,089*
183,267* (+17%)
146,462* +25%
1.7 GHz 111,556* +31%
78,426* +42%
52,286* +50%
31,826* +64%
+72%
2012
2010 2019 2022
zEC12 2015 2017
z196 IBM z15 IBM z16
2005 32 nm SOI z13 IBM z14
2008 14 nm SOI 7 nm SOI
45 nm SOI 101 Cores** 22 nm SOI 14 nm SOI
z9 EC z10 EC 190 Cores** 200 Cores**
80 Cores** OOO and eDRAM 141 Cores** 170 Cores**
90 nm SOI 65 nm SOI OOO core cache improvements Optimized OoO AI Acceleration on CHIP
54 Cores** 64 Cores** eDRAM cache Arch extensions SMT and SIMD Enh. SMT & SIMD
On-Chip Compression Restructured Cache
System level High-freq core RAIM memory for scaling Up to 10TB of Memory Up to 32 TB of
scaling 3-level cache zBX integration Memory Up to 40 TB of Memory Up to 40 TB of Memory
* PCI Tables are NOT adequate for making comparisons of IBM Z processors. Additional capacity planning required
IBM z16 Technical Overview_14 ** Number of PU cores for customer use © 2022 IBM Corporation
IBM zSystems
IBM z16 Full and Sub-Capacity CP Offerings
CP Capacity – Relative to Full Capacity Uni
701 = 100% ≈ 2,253 PCI
§ Subcapacity CPs, up to 39 may be ordered (317 total capacity levels).
601 ≈ 66% ≈ 1,496 PCI If more CPs are ordered, all must be full 7xx capacity.
501 ≈ 41% ≈ 937 PCI § All CPs on an IBM z16 CPC must be the same capacity (except during Recovery Boost
401 ≈ 12% ≈ 280 PCI periods).
§ All specialty engines are full capacity.
§ zIIP to CP ratio – 2:1 and is the same for CPs of any capacity
(except during System Recovery Boost periods).
7xx
6xx
5xx
4xx
PBU
PBU
PCIe 1 PCIe 0
NXU
§ Up to 1 SMP Connection (to other CPC
X-Bus
MCD
Drawers)
X-Bus
Core 1 Core3 Core 5 Core 7 Core 6 Core 4 Core 2 Core 0
AIU
X-Bus
§ ~50 GB/s bandwidth between drawers
X-Bus
L2 L2 L2 L2 L2 L2 L2 L2
X-Bus
M-Bus
32MB 32MB 32MB 32MB 32MB 32MB 32MB 32MB
M-Bus
X-Bus
X-FBC
X-FBC
§ 12 Xbus connections
X-Bus
M-Bus
M-Bus
X-Bus
L2 L2 L2 L2
X-Bus
32MB 32MB 32MB 32MB
X-Bus
chips in a drawer
X-Bus
AIU
Core 0 Core 2 Core 4 Core 6 Core 7 Core 5 Core3 Core 1
X-Bus
MCD
§ ~35 GB/s Bandwidth per link
NXU
PBU
PBU
PCIe 0 PCIe 1
*Speed and performance data based on running at max CP frequency in laboratory test
IBM z16 Technical Overview_19 © 2022 IBM Corporation
IBM zSystems
8-Core Processor Chip Detail (Telum)
X-Bus X-Bus X-Bus X-Bus X-Bus X-Bus § 8 Telum chips per CPC Drawer in 4 DCMs
PBU
X-FBC NXU § Up to 8 active cores per Telum Chip
L2 L2
§ Up to 200 active cores per system
A-Bus
Core 0 Core 1
32MB 32MB
§ Added Integrated Accelerator for AI
PCIe 0
L2 L2
Core 2 Core3
32MB 32MB
§ On Core L1 Cache
§ Private 128K L1I and 128K L1D
MC
L2 L2
Core 4 Core 5
PCIe 1
32MB 32MB
L2 L2 MCU
§ On Core/Chip L2 Cache
Core 6 Core 7
32MB 32MB
§ Each core has access to a private 32 MB cache
PBU
AIU M-Bus M-Bus MCD § Up to 16MB of each cache can be used by other cores as
virtual cache depending on the current activity
§ L2 cache of an inactive core becomes shared virtual L3
§ 7nm FinFET Technology
cache by the active cores of the chip
– 8 cores per CP § L2 cache of an inactive core of another CP can become
– 18.8 miles of wire virtual L4 cache
– ~ 23mm x 22 mm
– Up to 2 PCIe buses, and 8 DDIMMS § I/O buses
– 22.5B transistors versus 9.2B on z15 § Each CP chip will support up to 2 Gen-5 PCIe buses
IBM z16 Technical Overview_20 © 2022 IBM Corporation
IBM zSystems
8-Core Processor Chip Design & Improvements
§ General performance enhancements
§ New Cache/TLB Design
§ Branch prediction improvements
§ System Coherency Fabric Manager
§ On-Chip ring among all cores of a CP
§ Merge/Sort Accelerators
§ Introduced with the z15 CD announcement
On Chip AI Accelerator
Compute Arrays
– 128 processor tiles with 8-way FP-16 FMA SIMD
– Optimized for matrix multiplication and convolution
– 32 processor tiles with 8-way FP-16/FP-32 SIMD
– Optimized for activation functions & complex operations
Factory Only
−Max39, Max82, Max125, and Max168 each adds additional
CPC Drawers
Max168
−Max200 is implemented in 4 drawers
−No MES upgrade to Max168 or Max200
−Additional I/O Drawers
Max125
−Based on available space in current frames and/or I/O
Concurrent Upgrade
expansion frames
Prevent fraud before it happens Insights at unprecedented speed Leveraging AI, in operational
by scoring up to 100% of and scale means every customer processes can proactively
transactions in real-time without interaction can now be a identify and stop outages
impacting SLA’s personalized experience before they occur
Generated
inference program
IBM Deep
Deploy on
Learning IBM z16 and
Compiler IBM LinuxONE
ONNX interchange
format
Build and train model Use ONNX, an opensource tool for The IBM DLC Deploy on IBM z16
in any popular framework interoperability (Deep Learning Compiler), and IBM LinuxONE
framework on any Models are converted optimized for performance and infuse model into
platform of to the ONNX interchange format and new libraries, generates workload application
your choice Leverage zCX and run on a program from the model for
zIIP engines execution on z/OS or Linux®
on IBM z16
IBM z16 Technical Overview_39 © 2022 IBM Corporation
IBM zSystems
We are entering a new cryptographic era
There will be a time when the power Harvest now, decrypt later
of quantum may crack public key schemes are underway to collect data
now for decryption when quantum
cryptographic security protection … computers are powerful enough
Lifetime of data
means that sensitive data generated
today that is not protected with
quantum-safe algorithms is at risk
now
Quantum-safe technology and key management services were developed to help protect data
and keys against a potential future quantum attack like harvest now, decrypt later
Site A Site B
zIIP Configuration
Configuration zIIP CP: 401
CP: 720 zIIPs: 0
zIIPs: 10
0 ICF: 0
ICF: 4
0 IFL: 0
IFL: 4
0 IFL IFL
ACP: 401 ICF ICF
UzIIP: 10 720 720
UICF: 4
UIFL: 4
710 710
401 401
Flex Record Flex Record
CP: 20 CP: 20
IFL: 4 IFL: 4
ICF: 4 ICF: 4
zIIPS: 10 zIIPS: 10
Capacity indicator: 7 Capacity indicator: 7
Flex Capacity
– A3 moved to B2
Note: for up-to-date hardware support and service dates, please visit the End-of-Life pages for Cisco and Broadcom.
Dual-HSM FC 0908
Integrated Coupling Adapter Short Reach (ICA SR) 0172 Carry Forward 2 48
Integrated Coupling Adapter Short Reach 1.1 (ICA 0176 New Build / Carry Forward 2 48
SR1.1)
Coupling Express2 LR 0434 New Build 2 32
IBM z16
No InfiniBand
CF
Coupling Express2 LR
ICA SR 1.1
CF CF CF
IBM z16 Technical Overview_80 © 2022 IBM Corporation
IBM zSystems
Max 39/82/125
• 4 line cords required
• From 39 to 125 CPs
• 2-7 I/O Drawers (112 I/O
Cards Max)
Max 82/125
• 6 line cords required
• Up to 125 CPs
• 7-12 I/O Drawers (192 I/O
Cards Max)
Max 168/200
• 6 line cords required
• 168 or 200 CPs (Factory Build Only)
• 1-4 I/O Drawers (64 I/O Cards Max)
Max 168/200
• 4 frame expansion
• 8 line cords required Max 168/200
• 168-200 CPs • 3 Frame Expansion
(Factory Build Only) • 6 line cords required
• 9-12 I/O Drawers • 168-200 CPs
(192 I/O Cards Max) (Factory Build Only)
• 4-9 I/O Drawers
(144 I/O Cards Max)
• FC 7898 – Top Exit Cabling with Top Hat • FC 7899 – Bottom Exit Cabling
– Includes additional “Top Hat” hardware for strain ! Must be ordered to allow cabling or power in and out
relief, cable organization, or flat sealing surface if of the bottom of the frame for raised floor installation
required for hot/cold data center aisle containment ! Can be used with Fiber Quick Connect brackets at
– Must be ordered when planning to use Fiber Quick bottom of the frame
Connect brackets egressing out the top of the system ! If not ordered, bottom seal plate will ship with system
and I/O and power cabling cannot exit bottom of frame
• FC 7816 – Top Exit Cabling without Top Hat • FC 5827 – Fiber Quick Connect
– Ordered for clients intending to egress cables out of ! Optional feature provides brackets at top and/or bottom
the top of system without Top Hat hardware of system for cable organization and structured cabling
– Does not support Fiber Quick Connect Brackets on ! Does not ship trunking cables; clients will have to work
top of frame with third-party providers to acquire and install cables
Customer Bottom Exit Top Exit Cabling Feature Codes Additional Comments
Environment Cabling to be Ordered
Raised Floor Yes No 7899 only Bottom FQC support only
Raised Floor Yes Yes, no Top Hat 7899 & 7816 Bottom FQC support only
Raised Floor Yes Yes, with Top Hat 7899 & 7898 Top & bottom FQC support
Raised Floor No Yes, no Top Hat 7816 only No FQC support;
ships bottom seal plate
Raised Floor No Yes, with Top Hat 7898 only Top FQC support only;
ships bottom seal plate
Non-Raised Floor No (not supported) Yes, no Top Hat 7998 & 7816 No FQC support;
ships bottom seal plate
Non-Raised Floor No (not supported) Yes, with Top Hat 7998 & 7898 Top FQC support only;
ships bottom seal plate
IBM z16
HMC/SE
IBM z16
Operating Systems
Core WAS
PR/SM
Virtualized I/O
z/VM
•z/VM 7.2 with PTFs IBM cannot legally discuss
•z/VM 7.1 with PTFs IBM z16 exploitation prior
to GA from distributors.
#_Statements of Direction
IBM's statements regarding its plans, directions, and intent are subject to change or
withdrawal without notice at IBM's sole discretion. Information regarding potential future
products is intended to outline our general product direction and it should not be relied on
in making a purchasing decision. The information mentioned regarding potential future
products is not a commitment, promise, or legal obligation to deliver any material, code, or
functionality. Information about potential future products may not be incorporated into any
contract. The development, release, and timing of any future features or functionality
described for our products remain at our sole discretion.
• Removal of support for OSA-Express 1000BASE-T hardware adapters: IBM z16 will be the
last IBM Z server to support OSA Express 1000BASE-T hardware adapters (FC 0426, FC 0446,
and FC 0458). Definition of all valid OSA CHPID types will be allowed only on OSA-Express GbE
adapters, and potentially higher bandwidth fiber Ethernet adapters, on future servers.
• Firmware update process: IBM z16 is planned to be the last server family to support IBM service
support representatives (SSRs) on-site performing firmware updates without an additional
premium service contract. The IBM Z Remote Code Load (RCL) option, which was introduced on
the IBM z15, is available without an additional premium service contract. With the IBM z15, and
now with IBM z16, clients can request an RCL or they can choose the SSR onsite method for their
firmware update. IBM recommends clients to try the RCL option on the IBM z15 or IBM z16 to see
that IBM provides the same quality service through RCL.
• z/OS Containers and Kubernetes orchestration support for IBM z/OS applications and
workloads: IBM has previously announced the intention to provide clients with capabilities that will
help accelerate their transformation to greater portability and agility in a hybrid cloud environment
by delivering z/OS Containers and Kubernetes orchestration support for IBM z/OS applications
and workloads. To deliver on this capability, IBM intends to provide a beta program for z/OS 2.5
clients to begin their container journey with z/OS UNIX applications. These capabilities are
designed to support architecture-independent standards and new containerized deployment
options. The intention is to empower agile development teams to incorporate z/OS applications
into a Kubernetes-based orchestration model utilizing industry standard operations. Future z/OS
container use cases are planned to promote application modernization, new application
development, and API creation with tight integration to core z/OS applications.
• IBM LinuxONE Next release: In addition to the Linux on IBM Z functionality described in
this RFA, which will be available on IBM z16, IBM plans to announce and release a new
generation of LinuxONE systems in the second half of 2022, designed to help enterprises
in their hybrid cloud and AI journey.