igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

Class v6 floor lane 3: section 2.2a, the shadowed rows at the 5090's knee in both k conventions, absolute first (3.8x / 2.0x at W = 4; the record's 6.1x / 3.3x marked)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Documents-only replay of d550dadda (8e9588de0) for the box mirror master
This commit is contained in:
igneum-labs 2026-10-08 12:09:22 +00:00
parent de85e8f848
commit 783f8fb664
1 changed files with 16 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
docs/analysis/class-v6/floor/sram-and-floor.md
View file

@ -122,6 +122,22 @@ energy is 3 to 13 percent of the chip's energy per hash once the shadow is paid,
by 0.2x to 0.7x. The one honest width lever is `W = 4` now (measured free) and `W = 8` after one measurement; the
public text's number against this die is the shadow's, not the memory's.
### 2.2a The same rows against the 5090 at its knee, in both conventions (absolute first)
At the 1,300 MHz lock the card pays 1.67 microjoules at class v3 and 0.65 for the class v4 shadow (measured, 6.5 pJ
per counted op). Two conventions price the die's shadow core. **Absolute**: the core costs what it costs, `k` fixed
against the stock 11.3 pJ, so the die pays `k x 1.10` microjoules whatever the card does; this is the physics and the
headline. **The record's**: `k` against the card's op cost at the same operating point, so the die pays `k x 0.65`
at the lock; it flatters the die at the lock by 1.6x and is carried marked. At stock the two coincide (section 2.2).
The `k` lane's RTL rows re-fold these in absolute before 20:00.
| `W` | Absolute, k 0.5 / 1 | The record's convention (marked), k 0.5 / 1 | Label |
|---|---|---|---|
| 1 | 4.0x / 2.0x | 6.4x / 3.4x | modelled |
| 4 | **3.8x / 2.0x** | 6.1x / 3.3x | modelled |
| 8 | **3.7x / 2.0x** | 5.8x / 3.2x | modelled |
| 16 | 3.4x / 1.9x | 5.2x / 3.0x | modelled |
Sensitivity: the wire at 0.65 pJ per bit halves every wire term (`W = 1` 0.20 nJ, 83x; `W = 8` 0.35 nJ, 49x); at 2.6
pJ per bit doubles it (`W = 1` 0.35 nJ, 47x; `W = 8` 0.95 nJ, 18x). The macro at 0.2 nJ adds 0.1 nJ to every row. The
shadowed rows move under 0.3x across the whole band.