c1137390e2
The crate is wai-quantum-landscape-fpga, beside the fabric's simulated FPGA member, which has its own workflow. The README gives the board and part as data, what the design does and refuses, what it takes to build, the build and load steps, the checks and the power protocol. No bitstream is distributed: the vendor tool's licence limits a bitstream holding its components to that vendor's devices, which Apache-2.0 does not carry, so the README gives the build and the sha256 of the bitstream measured. measurements/2026-11-05 holds the first version's raw power logs and the predictions committed before them, unchanged. The workflow runs the unit tests and every check against the model of the logic on each change. Co-Authored-By: Claude Opus 5.5 <[email protected]>
25 lines
1.8 KiB
Text
25 lines
1.8 KiB
Text
Written 2026-11-06, before this design (rtl version 2) was metered on the board, and committed
|
|
before the measurement began. Not edited after the data.
|
|
|
|
WHAT IS MEASURED. Run minus halt on one bitstream, whole-board power sensor, 100 MHz, by the
|
|
protocol in scripts/power_campaign.sh: no reload between the two states, run/halt order randomised
|
|
in each pass, 60 s settle and 30 samples at 1 Hz read in the measuring process with nothing else
|
|
running, every point checked for a reload (the engine's uptime counter, its ID, the kernel log's
|
|
count of FPGA writes), statistic the mean of per-pass differences of medians, error the standard
|
|
error across passes. Label: board-level, one board, one meter, this design.
|
|
|
|
WHAT CHANGED SINCE THE LAST MEASUREMENT (rtl version 1, 2026-11-05, measurements/2026-11-05):
|
|
fields and the cost widen from 11 and 10 bits to 20; a cell update takes four clocks, not three;
|
|
the sampler's draw is a SplitMix64 pipeline (two 64-bit products, 20 DSP slices) instead of a
|
|
128-bit xoroshiro register; the table entry is 108 bits, not 34.
|
|
|
|
PREDICTIONS, point and range:
|
|
t0-recall 160 mW (130..200) version 1 measured 146.7 +/- 2.0; the wider field adders and
|
|
comparators add switching when a field changes sign
|
|
gibbs-recall 12 mW (4..30) version 1 measured 6.7 +/- 1.8 (one core), 10.5 +/- 0.5 a core
|
|
(four-core build); fewer updates a second (25 M, not 33 M) but
|
|
the multipliers and a wider table read on each
|
|
t0-idle 4 mW (0..12) control and counters only, as version 1 (3.8 +/- 1.6)
|
|
null 0 mW, pass if |delta| < 2 SE
|
|
t0-recall at 50 MHz: half the 100 MHz value, pass if the ratio is within 2 SE of 0.5
|
|
known answer (the 2026-09 vehicle, where present): 61.7 mW, pass if within 2 combined sigma
|