From WikiChip
Vega - Microarchitectures - AMD
< amd‎ | microarchitectures
Revision as of 22:17, 20 December 2018 by Jorgp2 (talk | contribs) (Fixed the arctic islands link)

Edit Values
Vega µarch
General Info
Arch TypeGPU
DesignerAMD
ManufacturerGlobalFoundries
Introduction2017
Process14 nm
Succession

Vega (also known as Graphics core next 5th generation) is both a microarchitecture and GPU family developed by AMD as a successor to the GCN4 microarchitectureand the arctic islands family.

Codenames

amd ryzen radeon.svg
Codename Description
Raven Ridge Mobile processors based on Zen CPUs and a Vega GPU

Models

Vega IGP Models Standards
Name Compute Units Shaders Vulkan Direct3D OpenGL OpenCL
Windows Linux Windows Linux HLSL Windows Linux Windows Linux
Vega 8 8 512 1.1 12 N/A - 4.6 4.6 2.2
Vega 10 10 640
Vega 11 11 704
Vega Dedicated Models Standards
Name Compute Units Shaders Vulkan Direct3D OpenGL OpenCL
Windows Linux Windows Linux HLSL Windows Linux Windows Linux
Vega 56 56 3584 1.1 12 N/A - 4.6 4.6 2.2
Vega 64 64 4096

Performance

The original target of the RX Vega series was to compete with Nvidia's latest Pascal architecture. RX Vega 64 -> GTX 1080 RX Vega 56 -> GTX 1070

The performance is generally good, but when it comes to performance per watt, or PPW, Vega is outperformed by Pascal.

Hardware Accelerated Video

Process Technology

See also: 14 nm process

Vega is manufactured on GlobalFoundries' 14 nm process.

Architecture

amd gpu roadmap.png
Symbol version future.svg Preliminary Data! Information presented in this article deal with future products, data, features, and specifications that have yet to be finalized, announced, or released. Information may be incomplete and can change by final release.

Key changes from Arctic Islands

  • New programmable geometry pipeline
    • Up to 2x throughput
    • Primitive shaders
    • Improved workload balancing
  • Tile based rendering
  • New low precision instructions
    • Half precision floating point with 2x performance
    • 8 bit integer with 4x performance
  • 15% higher clock speeds
  • HBM 2
    • 2x bandwidth
    • 4x memory capacity
  • New memory controller
  • Render back-end is now a client of the L2 cache
  • The instruction buffer has been enlarged


HBM 2

Vega makes use of 2 stacks of HBM 2 (High bandwidth memory).

single stack HBM 1 HBM 2
dies 4 + 1
4 dram die, 1 control die
2-8 + 1
2,4 or 8 dram die, 1 control die
Gb/die 2Gb 8Gb
total Gb 8Gb 16-64Gb
bus width 1024 1024
clock speed 500 MHz 945 MHz
bandwidth 128 GB/s 238 GB/s

References

  • AMD 2017 Financial Analyst Day, May 16, 2017

Documents

See Also

codenameVega +
designerAMD +
first launched2017 +
full page nameamd/microarchitectures/vega +
instance ofmicroarchitecture +
manufacturerGlobalFoundries +
microarchitecture typeGPU +
nameVega +
process14 nm (0.014 μm, 1.4e-5 mm) +