BENCHMARKS.md says what a model cost the day it was measured; nothing noticed if a
macOS/torch update or a thermal fault halved a node. perfcheck runs the whole fleet
in ~35s off godcheck's 03:30 cron on the m4mini and reports drift into
GODCHECK_LATEST.md.
Probes matmul (fp16/fp32), memory bandwidth, and SDPA at head_dim 64 vs 56 — the
latter turning CorridorKey's fast-path cliff into a permanent canary: it confirms
the padding win fleet-wide (1.96x-5.26x) and tells us if a future torch closes it.
Runs on venvs/rmbg/bin/python, already identical fleet-wide, so nothing new is
installed (nothing lands on the disk-tight m1max).
First cross-machine capability table for all 6 nodes. The M3 Ultra is ~2.2x the M1
Ultra on fp16 matmul, but they share ~625 GB/s — so bandwidth-bound stages run alike
while compute-bound ones scale. Both Ultras reach only ~78% of spec bandwidth on a
single kernel; the smaller Macs hit ~88%.
Measuring a fleet that is doing real work is the whole problem, and naive
benchmarking here is off by 11x:
- min, not median: a concurrent trellis_mac job dragged a median-of-5 matmul from
~24500 to ~2150 GFLOP/s, which reads exactly like a catastrophic regression.
- n=4096 not 2048: 2048 is dispatch-bound and swung 48% run-to-run; 4096 reproduces
to 0.1% even while contended.
- sdpa 16x2048 not 8x1024: sub-ms probes are dispatch noise — 8x1024 gave ratios of
0.79/3.95/2.35 on three runs of one machine, the first "proving" 56 is faster.
- busy nodes are excluded, not blamed: GPU contention is invisible to load average
(M3 Ultra read load 2.45 with its GPU pinned), so bench.py samples ioreg GPU% before
it touches the GPU — our own matmul pins the device, so ordering is the trick.
- baselines are the median of recent history, not a saved best: the M4 Pro also serves
Ollama and is bimodal (~3200 vs ~5500 fp32), so a best-observed baseline pins to a
lucky outlier and alerts forever.
- a regression must repeat before it is believed ([~] watching -> [!] CONFIRMED).
Validated by re-running the fleet against its own baselines: zero false alarms,
including a sweep where the M3 Ultra read 43 GB/s under load and was correctly
marked BUSY rather than reported as a 93% regression.
Also found: the m4mini is the only node with Tailscale SSH (RunSSH: true) and it does
NOT propagate remote exit codes — `ssh m4mini "exit 7"` returns 0, so any
`if ssh m4mini ...` test silently always passes. Test on output instead, which is
what godcheck already does (and why it is unaffected). It also cannot ssh to itself,
so run_fleet detects its own tailnet IP and benches the local node via the shell.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The gpu lane is now a NODE POOL (this Mac + remote workers from nodes.json) instead
of a single-Metal semaphore. A gpu job runs on whichever node is free:
- server/remote.py: ssh+rsync dispatch — mkdir remote dirs, rsync inputs out, run
the operator's run.py over ssh with repo-relative paths (HF_TOKEN sourced from the
node's .env.remote, off the process table), rsync outputs back, clean up. Cached
health checks; per-node operator allowlist.
- runner: _acquire_gpu_node picks the first free node that supports the op (local
runs anything; a remote must list it + be reachable → auto-fallback to local).
_run_job branches local/remote; the M1 never touches the M3's DB.
- sysinfo/Dashboard: gpu lane limit = pool size; per-node status cards.
- nodes.json (gitignored, primary-only): M1 worker, allowlist excludes trellis_mac
(bf16 unverified on M1) + brush_train (not installed there).
Verified: two FLUX jobs from one queue split M3(local, 10.3s) + M1(remote via ssh,
21.2s), both images rsync'd back and registered on the M3.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
