Revision as of 01:39, 6 November 2017

z14 was a z/Architecture-based microarchitecture designed by IBM and introduced in 2017 for their z14 processors and mainframes. The z14 microarchitecture replaced the z13.

Process Technology

z14-based microprocessors are manufactured on GlobalFoundries's 14 nm (14HP) FinFET Silicon-On-Insulator (SOI) process. The process was designed by IBM at what used to be their East Fishkill, New York fab which has since been sold to GlobalFoundries.

Release Dates

IBM z14 was announced on July 17, 2017. General availability started on September 13, 2017.

Architecture

Key changes from z13

• 14 nm process (from 22 nm)
  • 6.1B transistors (from 3.99B; 53% increase)
• Higher clock frequency (5.2 GHz from 5 GHz; 4% increase)
• Higher scalability
  • Up to 170-way multiprocessing (from 141-way)
• Core
  • Improved Operand Store Compare (OSC) prediction
  • Faster branch wakeup
  • Improved instruction delivery
  • Reduced execution latency
• Cache
  • New directory design
    • Power efficient
  • L1I$ increased to 128 KiB/core (from 96 KiB/core; 33% increase)
  • L2D$ increased to 4 MiB/core (from 2 MiB/core; 100% increase)
  • L3$ increased to 128 MiB/CP (from 64 MiB/CP; 100% increase)
  • New 672 MiB/drawer of shared L4
• TLB
  • New Translation/TLB2
    • Reduced latency
  • 4 concurrent translation
  • lookup integrated into L2 access pipe
  • 2x Larger CRSTE
  • 1.5x large PTEs
  • New 64-entry 2 GiB TLB2
• BTB
  • 1.33x larger BTB1
  • 1.33x larger BTB2
  • New Hashed Perceptron Predictor
  • New Simple Call Return Stack

• Central Processor Assist for Cryptographic Function (CPACF)
  • Dedicated co-processor for each core
  • Claims 6x faster encryption functions (vs. z13)
    • 4x Advanced Encryption Standard (AES) speedup
  • Support for True Random Number Generator
  • New support for SHA-3 standard

This list is incomplete; you can help by expanding it.

New instructions

• New SIMD instructions for COBOL and analytics applications

Overview

This section is empty; you can help add the missing info by editing this page.

Die

Core

Below is a layout of a single physical core:

• L2 - L2I$ + L2D$
• PC + TP - Core pervasive unit (instrumentation/error collection) + Trap
• LSU - Load-store unit (+ L1D$)
• XU - Translation unit (TLB + DAT)
• ICM - Instruction cache & merge
• COP - Dedicated Co-Processor
• FXU - Fixed-point unit
• VFU - Vector and Floating point Unit
• ISU - Instruction sequence unit
• RU - Recovery unit
• IDU - Instruction decode unit
• IFB - Instruction fetch and branch prediction

Single-chip module (SCM)

IBM's z14 Single-Chip Module (SCM) consists of a multi-layer metal substrate module that includes either:

• 1x Processor Unit (PU)
• 1x System Controller (SC)

Processor Unit (PU) Chip

• IBM's developed (now GlobalFoundries) 14HP Process
  • CMOS FinFET SOI
  • 17 Metal Layers
• deca-core
• 5.2 GHz (192 ps cycle time)
• 6,100,000,000 transistors
• 14.4 miles of copper wire
• 26.5 x 27.8 mm die
  • 736.7 mm² die size
  • 18,581 power pins
  • 1,505 signal pins

System Controller (SC) Chip

• IBM's developed (now GlobalFoundries) 14HP Process
  • CMOS FinFET SOI
  • 17 Metal Layers
• 25.3 x 27.5 mm die
• 695.75 mm² die size
• 7,100,000,000 transistors
• + 2,100,000,000 cells of eDRAM (~2.1B xTors + 2.1B capacitors)
  • 672 MiB shared eDRAM L4 Cache