AMD CEO Lisa Su reminisces about designing the PS3's infamous Cell processor

Original Article ↗ Hacker News Discussion ↗

Sony’s PlayStation 3 and its Cell processor are revisited as a bold but notoriously difficult hardware bet: an architecture that promised supercomputer‑class FLOPS, inspired supercomputer clusters and academic courses, yet proved hard to program and hurt game development and backwards compatibility. Commenters contrast Cell’s exotic heterogeneous design with today’s console-on-a-PC approach, arguing that rising asset budgets and cross‑platform requirements pushed the industry toward more conventional, developer‑friendly hardware. The thread also touches on IBM’s role, the Xbox 360’s competing architecture, and how lessons from Cell’s parallelism echo in modern multicore and GPU computing, including AMD’s current accelerator efforts.

Cell, Roadrunner, and Architecture Context

  • Cell co-architects also helped deliver early petaflop systems, e.g., Roadrunner combining PowerXCell and Opteron CPUs.
  • Roadrunner and Cell-based machines delivered strong FLOPS but were notoriously hard to program, with multiple architectures, differing endianness, and manual data movement.
  • Some view Cell and research chips like TRIPS as high-FLOP but wrong direction for general-purpose computing; programming difficulty outweighed raw performance.

Programming the PS3 / Cell

  • Developers describe SPEs as powerful but painful: no direct RAM access, tiny local memories, DMA latency, and awkward synchronization with the main core.
  • Effective use required hand-tuned vector code, double-buffering, careful task scheduling, and often separate codebases for PPE vs SPE.
  • Some argue mastering Cell’s task-based parallelism foreshadowed modern multicore and GPU-style models; others call SPEs fundamentally bad ideas (e.g., lack of caches).

Console Generations, Sales, and “Console Wars”

  • Disagreement over whether PS3 “lost” to Xbox 360:
    • One side: 360 dominated most of the lifecycle, better multiplatform performance, stronger game sales.
    • Other side: PS3 ultimately sold slightly more units globally, with late-cycle surges in regions like South America and SE Asia.
  • Wii is generally seen as winning the 7th gen in unit sales but targeting a different, more family/casual audience.
  • Nintendo is often framed as occupying its own space driven by first-party IP; some debate the extent of direct competition with Xbox/PlayStation.

OtherOS, Clusters, and Backward Compatibility

  • PS3 initially allowed Linux (OtherOS), enabling academic classes and small supercomputers; later removed over security/piracy concerns, prompting criticism and lawsuits.
  • Governments and research labs built PS3 clusters; performance-per-dollar was attractive but RAM constraints and evolution of GPUs limited long-term appeal.
  • Emulation: community projects now run PS3 games well on PCs, leading some to argue PS5 could emulate PS3, but Sony instead prefers remasters or streaming.
  • Xbox is praised for extensive 360 backward compatibility; PS3’s exotic design is blamed for many titles remaining locked to that platform.

From Exotic Consoles to PC-Like Boxes

  • Several commenters lament the loss of wildly different console architectures; modern systems are seen as “disguised PCs” (or tablets).
  • Others argue cost, complexity, and the need for multi-platform AAA releases make commodity-like x86/AMD-style SoCs inevitable.
  • There’s debate over whether unique hardware once forced better optimization versus simply imposing unnecessary pain on developers.

Controllers, Input, and Game Design

  • Discussion of how older generations introduced novel input (analog sticks, motion controls, pressure-sensitive buttons, touchpads), whereas recent Sony/Microsoft consoles are viewed as more iterative.
  • Nintendo is credited with continuing to push new input paradigms (e.g., Switch, Joy-Cons, HD rumble), though not all experiments succeed.
  • Some argue the main value of consoles now is ease-of-use and fixed targets rather than hardware uniqueness.

Game Budgets, Assets, and Aesthetics

  • Rising asset budgets (high-res models, full voice acting, cinematic cutscenes) are cited as a key driver for multi-platform releases and fewer exclusives.
  • Others counter that many successful PC/indie titles thrive without ultra-high-end production values, suggesting the AAA focus is a business choice, not a necessity.

AMD, GPUs, and MI300x

  • Mixed views on AMD’s GPU position:
    • Criticism that AMD underinvests in software (drivers, ROCm), leaving potential on the table versus Nvidia.
    • Counterpoint that MI300x is new, improving rapidly, and already usable for serious compute in some environments.
  • Practical constraints: lack of virtualization support and unstable software stacks currently make renting MI300x instances difficult; some startups are working around this and expect quick progress.

Broader Industry and Legacy

  • Some admire Lisa Su’s career and AMD’s role in mass-market advanced chips, while others think AMD is underperforming vs. Nvidia.
  • Brief digressions on PowerPC/POWER, RISC-V, ARM, and semiconductor supply-chain fragility underscore how much Cell and PS3 sit within a larger arc of CPU/GPU evolution and geopolitics.