← Back to catalogue
Research draft

graphics processing unit

vr.tr.graphics-processing-unit · PHY.OBJ

Let an agent explain GPUs, relay architecture, history, uses and policy debates from computer architecture and industry sources with positions attributed, flag product and bridge chip aliases, and distinguish GPUs from CPUs, systems on chips, AI accelerators such as TPUs, graphics cards as boards and FPGAs, without purchase or investment advice.

Thing Registry Physical world and living systems

Research draft, second pass

A second pass drafted this model: the structure a model of this thing needs, and what is known about it in the world. The line under this one says how the second half was obtained - researched against sources, or recalled without web access, in which case nothing here was read anywhere and every claim is a lead to verify. Unreviewed either way.

written by Claude from model knowledge without web access - no source was read, every claim is a lead to verify

Researched by: Claude

Purpose and description

Let an agent explain GPUs, relay architecture, history, uses and policy debates from computer architecture and industry sources with positions attributed, flag product and bridge chip aliases, and distinguish GPUs from CPUs, systems on chips, AI accelerators such as TPUs, graphics cards as boards and FPGAs, without purchase or investment advice.

A specialised processor designed to accelerate rendering of images and to perform many calculations in parallel, originating in graphics accelerators of the 1980s and 1990s, with Nvidia marketing the GeForce 256 as the first GPU in 1999, now used as discrete graphics cards, integrated GPUs within CPUs and mobile system-on-chip designs, and widely applied to general-purpose computing, scientific simulation, cryptocurrency mining and training artificial intelligence models; the registry aliases name Qualcomm Snapdragon system-on-chip products, which contain Adreno GPUs but are whole chips and instances, and the ATI Rialto, a PCI Express bridge chip, which appear misfiled. GPU export controls and supply constraints are subjects of policy debate, attributed.

What it is for: Parallel computation and graphics rendering.

It can be explain architecture and history; relay uses in graphics and AI; present policy debates with attribution; flag product aliases.

Distinguishing features

Massive parallelism

High memory bandwidth

Graphics and compute

Strategic technology

What it looks like

Chips on graphics cards with heatsinks and fans, or regions within CPUs and mobile chips.

Physical character

GeForce 256 marketed as GPU: 1999 year - Nvidia

CUDA released: 2007 year - general-purpose GPU computing

US AI chip export controls: from 2022 note - check current rules

How it is recognised

Parallel processor for graphics and compute

ATI Rialto, Qualcomm Snapdragon 670, Snapdragon 480 Plus 5G, Qualcomm Snapdragon 820, Qualcomm Snapdragon 660, Qualcomm Snapdragon 630

CPUs handle general sequential work; SoCs combine many components; TPUs are dedicated AI accelerators; graphics cards are boards; FPGAs are reconfigurable

Related models

is a kind of - in registry terms

coprocessor

is programmed with - among others

CUDA

is contrasted with -

central processing unit

is contrasted with - Snapdragon aliases

system on a chip

In practice

Families and kinds

discrete GPUs

integrated GPUs

mobile GPUs in SoCs

data centre GPUs

workstation GPUs

Standards and regulation

US and allied export controls on advanced chips

Graphics APIs such as Vulkan, DirectX and OpenGL

Failure modes and hazards

Treating SoC products as GPUs

Giving investment or purchase advice

Partisan framing of export controls

Also called

ATI RialtoQualcomm Snapdragon 670Snapdragon 480 Plus 5GQualcomm Snapdragon 820Qualcomm Snapdragon 660Qualcomm Snapdragon 630Qualcomm Snapdragon 425Qualcomm Snapdragon 435Qualcomm Snapdragon S1 MSM7225ARCPRadeonPowerVR SGX543MP2PowerVR SGX543MP4PowerVR SGX543MP3PowerVR SGX554MP4PowerVR G6430PowerVR GX6650RIVA TNT2GeForce 3 seriesAdreno 509Standard Television Interface ChipArdeno 630Adreno 305Adreno 650Intel ArcGA102Mali-400 MPAdreno 618Asahi Kasei AK8000Adreno 610Mali-400 MP1Mali-400 MP2Vivante GC7000ULPowerVR SGX544MP2Nvidia TitanATI XenosMali-G52Imagination PowerVR GX6250Imagination PowerVR GX6450PowerVR Series 7XT GT7600

+65

Where this came from

wikidata · CC0 1.0

Drafted structure

Bundle to layer to finding to question, as the second pass will find it: 4 bundles · 8 layers · 8 findings · 16 questions.

Understand What a GPU is.

Definition.

Definition

Definition.

Definition

Definition.

  1. What is a GPU, and how does it differ from CPUs, SoCs, TPUs, graphics cards and FPGAs? definition
  2. Is the question about gaming, AI computing, buying hardware or policy? boundary

Aliases

Product aliases.

Aliases

Aliases.

  1. What are the Snapdragon chips and ATI Rialto, and why are they not GPU kinds? definition
  2. Which entry fits the specific product? action
Architecture Architecture.

Science.

Parallelism

Parallel design.

Parallelism

Parallelism.

  1. Why are GPUs good at parallel workloads such as matrix operations? provenance
  2. Which references are standard? provenance

Memory

Memory and bandwidth.

Memory

Memory.

  1. How do GDDR and high-bandwidth memory affect GPU performance? provenance
  2. Which sources are cited? provenance
Uses Uses.

Sources.

Graphics

Rendering.

Graphics

Graphics.

  1. How do GPUs render 3D graphics through pipelines and shaders? provenance
  2. Which entry fits graphics pipeline? action

AI

AI training.

AI

AI.

  1. Why did GPUs become central to deep learning after 2012? provenance
  2. Which entry fits deep learning? action
Context Policy and industry.

Attribution.

Export controls

Export controls.

Export controls

Export controls.

  1. What do governments, companies and analysts argue about GPU export controls, with positions attributed? provenance
  2. Is the presentation neutral and current? boundary

Mining

Cryptocurrency mining.

Mining

Mining.

  1. How did cryptocurrency mining affect GPU supply, as reported? provenance
  2. Which entry fits cryptocurrency mining? action

What the second pass must settle

  • The registry aliases for Snapdragon products and ATI Rialto should be moved to their own entries
  • Should data centre GPUs be a separate entry?
  • How should export control rules be kept current?