technologybriefs
9:41in productionCh. 1 · Offloading the obvious/ 9:41 · ceiling 15 min
Semiconductors · AI

Graphics processing unit

The GPU didn’t accelerate graphics—it relocated computation, and in doing so, rewrote the rules of what a chip could be used for.

The GPU is a specialised electronic circuit for digital image processing and computer graphics acceleration. It emerged in the 1990s with internal calculation capabilities—unlike earlier video display controllers and blitters—and evolved to support shaders, 3D functionality, and massively parallel computation. Its parallel architecture enabled adoption beyond graphics, notably in AI and scientific computing. The term 'GPU' was coined by Sony in 1994 for the PlayStation's Toshiba-designed chip.

Chapters & takeaways4
  1. 0:53
    Offloading the obvious

    The GPU emerged in the 1990s as a dedicated circuit to accelerate graphics—freeing the CPU from drawing lines, text, and later 3D scenes.

  2. 2:42
    Calculation, not copying

    Unlike earlier controllers, modern GPUs calculate internally—rotating 3D images, running shaders, and handling tasks that video display controllers and blitters could not.

  3. 3:56
    Many small engines

    Graphics tasks are independent by nature—so GPUs pack hundreds or thousands of simple calculation units to exploit that parallelism.

  4. 5:16
    From pixels to parameters

    That parallel structure made GPUs useful beyond graphics—especially in AI, where they handle data-intensive, embarrassingly parallel workloads.

Worth your time?

Yes. Study the whole thing.

4.5/ 5
What works
  • offloading independent graphical calculations from the CPU
  • executing embarrassingly parallel workloads at high throughput
  • enabling real-time 3D rendering and AI model training on commodity hardware
What does not
  • replace general-purpose processors
  • handle serial or latency-sensitive tasks efficiently
  • simplify software development
Study it if
  • AI researchers
  • scientific computing teams
  • real-time 3D application developers
Skip it if
  • developers working primarily with event-driven web applications
  • embedded systems engineers with strict power budgets
  • teams maintaining legacy CPU-bound codebases without parallelisation capacity
The written brief1 min read

What it is and the problem it solves

A GPU is a specialised electronic circuit for digital image processing and computer graphics acceleration. It solves the problem of offloading independent, repetitive graphical calculations from the CPU—freeing it for system tasks while enabling richer visual output.

How it works

Modern GPUs contain hundreds or thousands of calculation units. They perform graphics tasks—like rotating and scaling 3D images—internally, without CPU help. They run custom shader programs. Unlike earlier video display controllers or blitters, they do more than move memory: they calculate.

What works

Graphics functions are independent—and thus suited to separate calculation engines. That independence enables massive parallelism. GPUs excel at embarrassingly parallel problems: matrix operations, pixel shading, neural network inference. Their adoption in AI stems directly from this trait—not from marketing or inevitability.

What does not

GPUs do not replace general-purpose processors. They cannot efficiently handle serial, branching, or latency-sensitive tasks. Their power comes at the cost of programming complexity, memory bandwidth asymmetry, and energy draw that scales with core count—not utility.

What it changes

GPUs changed where computation happens. They shifted massive parallel workloads from CPU clusters to single chips. They enabled real-time 3D rendering, accelerated AI model training, and turned consumer hardware into scientific instruments—without requiring new physics or materials.

Is it worth your time

Yes—if your work involves parallel computation on large datasets, especially in AI or scientific computing. The GPU’s architecture delivers throughput where CPUs stall. But it demands rethinking software design for parallelism, not just swapping hardware.

Same field · Semiconductors4 of 51
9:51
Hardware for artificial intelligenceAI hardware is a set of purpose-built chips—Lisp machines, GPUs, TPUs, NPUs—that accelerate AI workloads by optimising for parallelism, memory bandwidth, and low-precision arithmetic. It delivers real speed and efficiency gains, but at the cost of flexibility, portability, and architectural transparency. Its dominance reflects engineering necessity—not inevitability.
9:20
ChipletChiplets are modular ICs—functional fragments designed for assembly on an interposer. They enable heterogeneous integration, pre-test validation, and IP reuse. But they shift complexity from fabrication to integration, and require standards compliance to deliver on their promise.
9:27
ECC memoryECC memory adds redundancy and decoding logic to catch and fix single-bit memory errors before they propagate. It works via SECDED Hamming or Hsiao codes, using 8 extra bits per word—64+8 for DDR–DDR4, 32+8 per subchannel for DDR5. Mandatory on-die ECC in DDR5 and LPDDR6 addresses rising soft-error rates from shrinking process nodes. It delivers deterministic single-bit correction in servers and critical infrastructure—but offers no protection beyond that, and no benefit where silent corruption is tolerable.
9:33
EUV lithographyEUV lithography replaces transmissive deep-UV optics with reflective Mo/Si mirrors, tin-plasma light sources, and vacuum operation to pattern sub-7 nm semiconductor features. It works — but only after decades of co-development across continents, and only where photon budget, thermal drift and stochastic noise can be managed. It is necessary, not optional, for leading-edge logic — but it does not generalise, simplify or cheapen.
Up next in Technology

Drive Hard

2014 · 9:10

A film that promises action but delivers only speed-limit cruising and neutral dialogue.

9:10