The PS3 Licked the Many Cookie

Original Article ↗ Hacker News Discussion ↗

What “many-core” means and whether it survived

  • Several commenters initially equate “many-core” with modern 8–64 core CPUs or mobile big.LITTLE designs; others clarify the article’s meaning: heterogeneous many-core with dissimilar coprocessors exposed directly to programmers.
  • Modern systems still use heterogeneous compute (AI/ML blocks, media and crypto engines, tiny always-on coprocessors), but these are mostly hidden behind APIs, unlike Cell’s explicitly programmed SPEs.
  • Some argue “many-core is alive and well” in homogeneous server/workstation CPUs and mixed P/E cores; others say that’s a different category from Cell-like architectures.

In what sense the PS3 “failed”

  • Repeated clarification: the article’s “PS3 failed” means it failed developers and as an architecture, not that the console bombed commercially.
  • Debate over whether that wording is misleading given ~87M units sold and a narrow win over Xbox 360.
  • Several note that PS3 sold far fewer units than PS2 and that Sony completely abandoned Cell next generation, suggesting internal judgment that the architecture was a failure.

Developer experience on Cell

  • Strong consensus that Cell/SPE programming was painful: separate ISAs/toolchains, tiny 256 KB local stores, manual DMA, no OS/standard library, no memory protection, and extremely slow CPU reads from GPU memory.
  • Anecdotes of IBM’s SDK being confusing and brittle, emblematic of pre-iPhone embedded toolchains.
  • Many argue this was an “expert-friendly” system; only first-party studios with time, money, and access fully exploited it.

Game performance and cross‑platform issues

  • On paper PS3 was more capable, but many multi-platform titles ran worse than on Xbox 360 due to dev complexity and unfamiliarity; some studios reportedly developed primarily for 360/PC and “auto-ported” to PS3.
  • Others counter that by late generation, teams that mastered Cell (especially first‑party) produced games that matched or exceeded 360 versions; task-based/job systems eventually emerged.
  • Cell’s heterogeneity also made ports to PC and other consoles costly, discouraging deep use of SPEs in cross‑platform engines.

Economics, price, and Blu‑ray

  • Launch price ($599) and expensive Cell/Blu‑ray hardware are seen as major commercial handicaps; early on, Wii and 360 dominated mindshare.
  • Over time, cheaper revisions, Blu‑ray value (cheapest player for a while), built‑in HDD, and strong exclusives recovered sales.

Tooling, composability, and legacy

  • Some dispute the article’s thesis that heterogeneous compute is inherently non-composable, arguing that good libraries and middleware (e.g., PS3’s later PlayStation Edge, modern ML stacks) can hide complexity.
  • Others maintain Cell’s specific design (PPE + exposed SPEs + weak GPU) was a poor transistor tradeoff; many‑core success now comes either as homogeneous cores or tightly abstracted accelerators.
  • Commenters link PS3’s pain to Sony’s later pivot: PS4/PS5 adopt far more conventional, developer‑friendly architectures after extensive consultation with devs.

Low-level technical details and anecdotes

  • Noted: ~16 MB/s CPU read bandwidth from RSX memory; 500‑cycle memory/DMA latencies vs very high clocks; SPUs heavily used to compensate for a relatively weak GPU (post‑processing, geometry/vertex work).
  • Some see the era as a mismatch: highly parallel hardware arriving before mainstream engines, tools, and concurrency practices were ready.