From WikiChip
Difference between revisions of "sifive/microarchitectures/7 series"
< sifive

(Key changes from {{\\|5 Series}})
(Core)
Line 54: Line 54:
  
 
=== Core ===
 
=== Core ===
7 Series cores are based on a [[dual-issue]] [[in-order]] [[pipelined]] design.  
+
7 Series cores are based on a [[dual-issue]] [[in-order]] [[pipelined]] design.
 +
 
 +
 
 +
:[[File:7 series pipeline.svg|500px]]
 +
 
 +
The pipeline itself is one cycle longer than prior microarchitectures, enabling up to 10% higher frequency. SiFive reworked the execution units. The new design can handle up to two instruction being issued at once, this is double the prior design which brings along substantial performance improvement.
 +
 
 +
==== Memory subsystem ====
 +
The largest change in the 7 Series is the overhaul of the memory subsystem. The data cache and the optional tightly integrated memory (TIM) can now span two cycles, enabling large SRAM/TIM to be included with the core. Additionally, two sets of ALUs have been incorporated into the pipeline in order to allow a zero cycle [[load-to-use latency]] where the first stage is used for the address generation and the last stage can be used to operate on the data.
 +
 
 +
===== Fast I/O Port (FIO Port)=====
 +
7 Series incorporates an optional fast I/O port (FIO port) which is tightly coupled to the core, enabling low-latency core-to-memory/accelerator operations. The FIO port can be used to incorporate much larger [[SRAM]] as well as custom accelerators via the accelerator register interface.
 +
 
 +
:[[File:sifive 7 series core with fio port.svg|500px]]
  
 
=== Cluster ===
 
=== Cluster ===

Revision as of 23:26, 4 November 2018

Edit Values
7 Series µarch
General Info
Arch TypeCPU
DesignerSiFive
ManufacturerTSMC, GlobalFoundries
IntroductionOctober 21, 2018
Core Configs1, 2, 4, 6, 8
Pipeline
TypeSuperscalar, Pipelined
OoOENo
SpeculativeYes
Reg RenamingNo
Stages8
Decode2
Instructions
ISARISC-V
Cores
Core NamesE76,
E76-MC,
S76,
S76-MC,
U74,
U74-MC
Succession

7 Series is a series of high-performance RISC-V IP cores designed by SiFive.

Architecture

Key changes from 5 Series

  • Core
    • 8-stage pipeline (from 5-7)
      • 10% higher frequency
    • dual-issue (from single-issue)
    • Memory subsystem
      • 0-cycle load-to-use latency (down from 1 cycle)
      • 2-cycle SRAM access time worst case (down from 5 cycles)
      • Fast I/O Port (FIO Port)

Block Diagram

Cluster

sifive 7 series cluster.svg

Core

sifive 7 series block diagram.svg

Overview

The 7 Series is based on an improved version of the 5 Series which SiFive claims to provide 60% improvement in CoreMarks/MHz as well as 40% improvement in DMIPS/MHz. Series 7 base design is a cluster of up to 9 cores (8+1). Cores can be a mix of 7 Series cores as well as SiFive other existing cores. Everything in the cluster is cache coherent - including any extended SRAM options as well as any custom accelerator attached to the cores. The cluster can be further scaled up through the use of AMBA or TileLink which allows for up to 64 clusters to be integrated onto a single chip. Multi-chip support is also possible via ChipLink.

Core

7 Series cores are based on a dual-issue in-order pipelined design.


7 series pipeline.svg

The pipeline itself is one cycle longer than prior microarchitectures, enabling up to 10% higher frequency. SiFive reworked the execution units. The new design can handle up to two instruction being issued at once, this is double the prior design which brings along substantial performance improvement.

Memory subsystem

The largest change in the 7 Series is the overhaul of the memory subsystem. The data cache and the optional tightly integrated memory (TIM) can now span two cycles, enabling large SRAM/TIM to be included with the core. Additionally, two sets of ALUs have been incorporated into the pipeline in order to allow a zero cycle load-to-use latency where the first stage is used for the address generation and the last stage can be used to operate on the data.

Fast I/O Port (FIO Port)

7 Series incorporates an optional fast I/O port (FIO port) which is tightly coupled to the core, enabling low-latency core-to-memory/accelerator operations. The FIO port can be used to incorporate much larger SRAM as well as custom accelerators via the accelerator register interface.

sifive 7 series core with fio port.svg

Cluster

The basic cluster comprises up to 9 cores configured as 8+1. The type of cores is fully customizable and may incorporate SiFive previous cores (e.g. the high-efficiency 2 Series cores). All the cores sit on a cache coherent bus.

Cores

Core Perf Comperable ISA
E7 2.3 DMIPS/MHz
4.9 CoreMarks/MHz
Cortex-M7 RV32GCV
S7 2.5 DMIPS/MHz
4.9 CoreMarks/MHz
Cortex-R8 RV64GCV
U7 2.5 DMIPS/MHz
4.9 CoreMarks/MHz
Cortex-A55 RV64GCV
codename7 Series +
core count1 +, 2 +, 4 +, 6 + and 8 +
designerSiFive +
first launchedOctober 21, 2018 +
full page namesifive/microarchitectures/7 series +
instance ofmicroarchitecture +
instruction set architectureRISC-V +
manufacturerTSMC + and GlobalFoundries +
microarchitecture typeCPU +
name7 Series +
pipeline stages8 +