mindmap · hereandmore · 4D slice

Configuration space

Four projections of one discrete volume: Space (boards), Reach (the wire), Time (prefill window), Capability (what a mind can say across the ISA wall). The grid is model × harness. Filled cells were measured or documented. Empty cells are unknown — not zero, not “doesn’t work.” People are off this map.

projections · keep separate

People off

DeepSeek-V4-Flash × pd-bridge × 2 Spark + Studio × 10 GbE × 105K

Axonometric of discrete cells — not a metric rack scan. Measured boards are solid. Named rack ghosts are outlines. No line means no evidence of a link.

time window · T

Validated envelope on the bridge is ~20K–105K. 241K was run at the prefill pair’s 262K ceiling — measured, outside that envelope. Above that, cells go blank or blocked.

SPACE · boards as places

bench 2+1 · rack 7+5 named

Two Sparks and one Studio are the measured join. The other boards in the photo are named, not joined.

On a narrow screen, scroll the volume sideways to the Studios.

  • measured
  • documented (other lab / other mechanism)
  • recipe, unmeasured
  • posted, not RESULTS.md
  • goal
  • blocked
  • unknown (blank)

model × harness · one slice at T

Every pairing we will admit

Columns are models this bench could even name. Rows are harnesses that have touched this silicon family in public. Most intersections stay empty on purpose. Click a blank: the inspector will say unknown.

mind communication · capability at this T

What can cross the ISA wall

A mind here is tokens leaving metal after a prefix has been scheduled. Communication is a file the consumer can open — not a CUDA object, not a shared address space. The lists below are the Capability projection written as sentences.

instance · 2026-09-06 · RESULTS.md

Measured windows

DeepSeek-V4-Flash 284B / 13B active. Prefill: 2× DGX Spark, vLLM TP2, FP8, 149 GB. Decode: Mac Studio M3 Ultra, oMLX 0.6.4, MXFP4, 156 GB. Link: ordinary 10 GbE. Decode rate unchanged both ways (23–25 tok/s). Quality eval 5/5, same as native — five questions, not “equivalent.”

cold promptStudio aloneSparks → Studioratiostatus
~25K42.6 s28.2 s1.5×measured
~82K205.8 s72.9 s2.8×measured (old writer ceiling; fixed)
~105K245.6 s75.5 s3.3×measured · validated envelope top
~241K732.3 s200.3 s3.7×measured at 262K prefill ceiling
241K warm4.9 s / 19.3 sbypassmeasured · file already scheduled

Payload ~10 KB/token (0.80 GB for 81K, pulled in 1.08 s). Network is not the bottleneck; prefill is ~65% of wall. Warm turns never take the bridge. A silent native fallback can no longer enter a results table: every reply carries X-PD-Bridge (complete / partial B/T / declined).

Posted, not RESULTS.md

A 2026-09-07 lab post added longer rows and a rack photo (7 Sparks + 5 Studios). Those numbers are not the validated envelope. They sit in this map as posted, so they cannot be mistaken for RESULTS.md.

loadStudio alone2 Sparks → Studioratiostatus
100k5.4 min0.9 min5.8×posted
241k cold12 min3 min~4×posted (aligns ~732 s / ~200 s)
500k37.9 min6.3 min6.1×posted
1.25M2.43 h23.1 min6.3×posted
2.09M5.64 h52.3 min6.5×posted

Rack photo = goal inventory (896 GB CUDA + ~1.5 TB Metal). Bench = 2 Sparks + 1 Studio. Goal window 262K → 524K → 2M and a 1–2T 4-bit model on five Studios are named, not measured.

nine strata · the same volume, as matter

Substrate

The 4D slice sits on geology. Click a stratum if you want the mineral clock instead of a configuration cell.

  1. 01

    Mineral

    SiO2, Cu, Au, Li, water, rare earths in fans. Geology before software.

  2. 02

    Power

    Watts in. Decode is cheap. Prefill is the expensive clock.

  3. 03

    Silicon

    GB10 CUDA vs M3 Ultra Metal. Same element, two ISAs, two caches.

  4. 04

    Cooling

    Every watt that is not a token is heat. Thermal ceiling delays the first word.

  5. 05

    Metal

    The board you unplug. Hostname is a path. Die and serial are the place.

  6. 06

    File

    Weights, prefix, memory files. Do not ship an unreadable KV.

  7. 07

    Schedule

    Intent etched, not yet speaking. Existence of the prefix is not execute.

  8. 08

    Prompt

    Unlock. First word is the cost. Then 23–25 tok/s, all day.

  9. 09

    Mind

    Tokens leaving metal. Loop: mineral → die → file → schedule → prompt → token → new file.

CUDA prefill → Metal prefix store

1 / 6

Two halves. No shared cache format.