SIGILL crash on AMD Ryzen Zen 3 (no AVX-512 support) — all sessions crash #38986

Activity

1. github-actions

   assigned

   Brendonovich

   on Jul 26, 2026
2. github-actions commented on Jul 26, 2026

   

   github-actionsBot

   on Jul 26, 2026 – with GitHub Actions

   Contributor

   More actions

   This issue might be a duplicate of existing issues. Please check:

   • Bundled OpenCode x64 binaries can crash with SIGILL / invalid opcode; baseline package may not be truly baseline #33595: Bundled OpenCode x64 binaries can crash with SIGILL / invalid opcode; baseline package may not be truly baseline
   • opencode Worker subprocess crashes with SIGILL (illegal instruction) on Intel i5-7200U (Kaby Lake), freezing the system #36280: opencode Worker subprocess crashes with SIGILL (illegal instruction) on Intel i5-7200U (Kaby Lake)
   • macOS x64 "baseline" binary requires AVX2/FMA — crashes on Ivy Bridge CPUs #29039: macOS x64 "baseline" binary requires AVX2/FMA — crashes on Ivy Bridge CPUs
   • zsh: illegal hardware instruction opencode #8345: zsh: illegal hardware instruction opencode

   These issues all cover the same root cause: bundled Bun/native binaries including AVX-512 or AVX2 instructions that crash on CPUs lacking those extensions. Your report adds useful evidence (AMD Zen 3 + desktop AppImage path) that may help narrow down the packaging issue.
3. lqmnah commented on Aug 25, 2026

   

   lqmnah

   on Aug 25, 2026

   More actions

   Confirming this on Intel, not just AMD Zen 3 — and still present 16 releases later.

   Environment

   • opencode 1.18.21 (crashes) / 1.18.22 (currently installed, auto-updated)
   • Bun 1.3.14 (bundled), binary 184,498,304 bytes
   • Ubuntu 26.04 LTS, kernel 7.0.0-30-generic, x86_64
   • CPU: Intel Core i9-14900KF (Raptor Lake), microcode 0x133
   • Installed via npm (~/.opencode/bin/opencode), not the desktop package

   CPU lacks AVX-512

   avx512f  : absent
   avx512bw : absent
   avx512vl : absent
   avx2     : present
   

   This matters for scope: every Intel consumer CPU from 12th gen onward lacks AVX-512
   (disabled because of the E-cores). So this is not an AMD-specific bucket — it covers
   Alder Lake, Raptor Lake, and Meteor Lake desktops/laptops as well as Zen 3.

   Same evidence in the npm binary

   Disassembling .text of the installed binary:

   %zmm register uses    : 6888
   %k1-%k7 mask registers: 2575
   

   Concrete EVEX instructions present:

   vbroadcasti32x4 -0x22b9e5f(%rip),%zmm5
   vinserti64x4    $0x1,%ymm3,%zmm4,%zmm3
   vshufi64x2      $0x0,%zmm4,%zmm4,%zmm4
   

   Kernel traps

   Two crashes, the same instruction pointer, different threads and different model
   providers:

   Aug 23 18:10:22 kernel: traps: HeapHelper[504903] trap invalid opcode
     ip:2607064 sp:7217591ec8d0 error:0 in opencode[21a3064,24c8000+36e6000]
   
   Aug 24 16:04:32 kernel: traps: opencode[3915644] trap invalid opcode
     ip:2607064 sp:7ffd19044c90 error:0 in opencode[21a3064,24c8000+36e6000]
   

   Crash 1 was on a local Ollama provider, crash 2 on a remote OpenAI-compatible endpoint,
   so it is below the provider layer — consistent with what #33595 argues.

   One difference worth noting

   For us this is intermittent, not every session: 2 crashes in roughly 40 hours of
   heavy agentic use, both during long tool-call loops, rather than a crash on every start.
   That fits a rarely-taken code path containing AVX-512 rather than a hot one.

   Possibly related on the same host: opencode run twice hung indefinitely with no output
   and no upstream request ever sent (once ~19 minutes, zero completed turns, model server
   idle with nothing loaded) and had to be killed. No log line after the run started. Not
   claiming it is the same defect, only that it appeared on the same build.

   Ruled out here: system OOM (55 GiB free at crash 1), machine fault (the later reboot was
   a clean systemd-reboot), and the Raptor Lake Vmin-shift errata (microcode 0x133,
   zero MCEs).

   Happy to attach a full backtrace — systemd-coredump is now capturing cores on this host
   with ProcessSizeMax=16G, so the next occurrence on 1.18.22 will produce one.
4. bronger commented on Sep 30, 2026

   

   bronger

   on Sep 30, 2026

   More actions

   I this “illegal instruction” with OpenCode 1.18.33 on an AMD Turion II. Is this the same problem, or are such very old CPUs even another issue?
5. thdxr commented on Oct 11, 2026

   

   thdxr

   on Oct 11, 2026

   Member

   More actions

   Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
6. thdxr

   closed this as not plannedon Oct 11, 2026