Athlon 64: How AMD turned the tables on Intel

Original Article ↗ Hacker News Discussion ↗

Nostalgia and User Experience

  • Many recall Athlon and Athlon 64 builds as huge step-function upgrades: quieter, cooler, cheaper, and often faster than contemporary Pentiums, especially for gaming and Linux.
  • XP x64 and early 64‑bit Linux on Athlon 64 are remembered as surprisingly stable; issues were often artificial OS/vendor blocks rather than hardware limits.

“x86 Is Dead” vs Itanium Reality

  • Commenters recall a period when press and vendors treated x86 as a doomed legacy and Itanium/IA‑64 (EPIC/VLIW) as the inevitable 64‑bit future.
  • Hands‑on reports: Itanium could be very fast for floating point and some Java/HPC workloads, but general-purpose code and everyday tools were often slower than cheap Athlon boxes.
  • Huge complexity was pushed into compilers: static scheduling, predication, register windows, massive register files, hint fields, odd calling/exception models. Multiple compiler teams struggled for years and still couldn’t get broadly good codegen.
  • Some argue Itanium wasn’t intrinsically “turd” so much as badly matched to real-world software and memory behavior; others say it was fundamentally the wrong path.

Why AMD64 Succeeded

  • AMD64 is praised as a pragmatic, well-thought-out extension: good 32‑bit performance first, 64‑bit as “gravy”, with more general‑purpose registers, NX bit, and largely seamless compatibility.
  • Early 64‑bit mode sometimes paid a cost in wider pointers but usually gained more from added registers.
  • Linux on Alpha and other early 64‑bit ports had already flushed out 32‑bit assumptions, making the transition to x86‑64 smoother on open-source stacks than on Windows.

Intel’s 64‑bit Missteps and Politics

  • Intel had its own x86‑64 design (Yamhill) in Pentium 4 but management fused it off to avoid “betraying” Itanium; later re‑introduced it as EM64T/Intel64 once AMD64 was clearly winning.
  • Several posts describe internal and OEM‑market politics: protecting IA‑64, fear of cannibalizing higher‑margin lines, and strong OEM pressure that kept AMD out even when technically superior.
  • Microsoft is cited as a key arbiter: it supported IA‑64 and AMD64, but refused to support an additional Intel‑only 64‑bit ISA.

Death of RISC Workstations and Rise of x86‑64

  • Opteron/AMD64 plus Linux are seen as the combination that finally killed most proprietary RISC/Unix workstations and many high-end servers (Alpha, PA‑RISC, SPARC, most MIPS).
  • Debate on causality: some credit Itanium’s failure, some x86 out‑of‑order designs, some fab economics and volume advantages, with Linux merely “being there” when x86 became “good enough.”

Windows Compatibility and 16‑bit Code

  • Discussion clarifies that in x86‑64 long mode you can’t use v8086 mode; running old 16‑bit DOS/Windows software requires emulation or complex tricks.
  • Microsoft had NTVDM/SoftPC‑based emulation for non‑x86 and internal 64‑bit builds, but chose not to ship 16‑bit support on 64‑bit Windows, likely due to low usage and architectural constraints.