GPU

3 players: Nvidia Geforce, AMD Radeon, Intel ARC Alchemist.

NVIDIA

Geforce

Generations:

  • GEFORCE RTX 40 SERIES (2022H2?): Lovelace architecture.
  • GEFORCE RTX 30 SERIES (2020): Ampere architecture.
  • GEFORCE RTX 20 SERIES (2018): Turing architecture.
  • GEFORCE GTX 16 SERIES (2019):
    • 16 series is based on the same Turing architecture used in thes 20 series, but omitting the Tensor (AI) and RT (ray tracing).
    • SUPER version: with GDDR6.

(R for Ray tracing)

Suffix:

  • (no suffix): standard version.
  • SUPER: super version.
  • Ti: Titanium version.

Data Center

NVIDIA A100 TENSOR CORE GPU: Powered by the NVIDIA Ampere Architecture, A100 is the engine of the NVIDIA data center platform. Used in data centers for AI, data analytics, and HPC.

This GPU has no display drivers.

Architecture

Name Since Process
Ampere 2020 TSMC 7 nm N7
Turing 2018 TSMC 12 nm FinFET
Volta TSMC 12 nm FFN
Pascal TSMC 16 nm FinFET+

Links:

AMD - Radeon

  • Radeon RX 6000 series, RDNA 2 architecture, since 2020
  • Radeon RX 5000 series, RDNA architecture, since 2019

RDNA = Radeon DNA, the microarchitecture.

Suffix:

  • XT: the most powerful

Intel ARC

Graphic solutions: hardware + software + services.

Naming:

  • Alchemist (2022)
  • Battlemage
  • Celestial
  • Druid

Based on the Xe-HPG (high-performance graphics) microarchitecture. Built with Xe-cores featuring 16 vector engines, and 16 matrix engines, with 256 and 1024 bits respectively per engine.

Features: Xe Super Sampling, Deep Link, and Ray Tracing.

Modern Graphics & Compute APIs

Modern GPU programming has shifted from high-overhead driver abstractions (OpenGL, DirectX 11) to explicit, low-overhead hardware interfaces:

  • Vulkan: The cross-platform industry standard maintained by the Khronos Group (unifying OpenGL, OpenGL ES, and OpenCL). Derived from AMD's Mantle API. Gives developers explicit control over memory allocation, command queues, and multithreaded rendering.
  • Metal: Apple’s proprietary low-overhead API for macOS, iOS, and Apple Silicon, replacing OpenGL and OpenCL with deep C++14 and Swift integration.
  • DirectX (Direct3D 12): Microsoft’s low-overhead API for Windows and Xbox.
  • CUDA: NVIDIA's dominant proprietary parallel computing platform for general-purpose computing on GPUs (GPGPU).

Web Standards

  • WebGL: JavaScript API based on OpenGL ES, allowing hardware-accelerated 2D and 3D graphics inside browsers.
  • WebGPU: Modern successor to WebGL, exposing low-level GPU compute and rendering primitives inspired by Vulkan, Metal, and Direct3D 12 directly to web applications.

Shading Languages

GPUs execute specialized shader programs across thousands of parallel SIMD cores:

  • GLSL: OpenGL Shading Language (human-readable C-like syntax).
  • HLSL: High-Level Shading Language for DirectX.
  • MSL: Metal Shading Language (based on C++14).
  • WGSL: WebGPU Shading Language.
  • SPIR-V: Standard Portable Intermediate Representation; binary intermediate bytecode for Vulkan shaders.