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

Class v6 D4: the coexistence model second cut (the D4 checklist as its section list: tariff vs hardware advantage with the 6.25x row first, break-even electricity for 17 classes, proving as a second income axis with the 12 GB measured zero, per-class supply curve with the installed base as a cap and re-entry, the chip generation step, supplier and operator dependence, the seven conditions: the DRAM board passes, the SRAM die fails once sunk) and the operator simulation first run (five shocks, two worlds; the fixed internal pool starves under a proving spike, the congestion-priced internal fee restores it in 35 periods)

Browse source 
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
igneum-labs 2026-10-08 15:57:10 +00:00
parent c82622af35
commit a07be1ee70
2 changed files with 425 additions and 249 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

560
docs/analysis/class-v6/coexistence-model.md
View file

@ -1,93 +1,98 @@
# The coexistence model: can a specialised supplier earn a normal return while GPUs stay close enough to compete?
# The coexistence model (second cut): can a specialised supplier earn a normal return while GPUs stay close enough to compete?
8 October 2026, 16:1x to 16:5x UK, branch `class-v6-floor-sram`, floor lane 3, on the research lane's word of 17:0x UK
(the founder's accepted third external review: the capex wall is replaced by a coexistence model as the economic
argument, with the profitability surface of `docs/analysis/class-v6/floor/sram-and-floor.md` section 4.4 as its base).
First run on today's measured rows. **Every row is modelled**: the arithmetic is `scratchpad/coexist.py` run on
build-3; the card rows are lane 4's class v5 table (`docs/analysis/class-v6/floor/denominator.md`, section 10.4 of
the design document: measured at the floor where it says so, modelled knees elsewhere), the chip rows the chip model's
(chip-model-v3 5.5 and 5.12, lane B, the k lane's 3.2 pJ per forced op), the prices street approximations. Nothing
here is served; nothing is a measurement of a chip. The founder is not named.
8 October 2026, 16:1x to 17:0x UK, branch `class-v6-floor-sram`, floor lane 3, Igneum 2.0 D4 (the research lane's word
of 17:0x UK; the founder's accepted external review replaces the capex wall with this model; the profitability surface
of `docs/analysis/class-v6/floor/sram-and-floor.md` section 4.4 is its base). Second cut; the section list is the D4
checklist so the pin can be closed line by line. **Every row is modelled**: the arithmetic is `scratchpad/coexist.py`
(the first cut, run on build-3 at 16:2x) and `scratchpad/coexist2.py` (this cut, run on build-4 at 16:52; build-3 was
down from 16:40); the card rows are lane 4's class v5 table (`docs/analysis/class-v6/floor/denominator.md`, section
10.4 of the design document: measured at the floor where it says so, modelled knees elsewhere), the chip rows the chip
model's (chip-model-v3 5.5 and 5.12, lane B, the k lane's 3.2 pJ per forced op), the proving rows the bench table's
shard times with the fleet lane's measured day on Devnet 3 (the 12 GB tier a measured zero for internal proving), the
prices street approximations. The operator simulation beside it is `docs/analysis/class-v6/operator-simulation.md`.
Nothing here is served; nothing is a measurement of a chip. The founder is not named.
The statement under test. **Success**: a specialised supplier earns a normal return and GPUs stay close enough in total
cost, obtainable and useful outside mining, that entrants still compete. **Failure**: a supplier operating privately at
much lower cost exhausts competitors' margins. The result is the set of conditions under which the success statement
holds, never a level the chain stays below. The development cost is SUNK in the mandatory case (the opponent covers
much lower cost exhausts competitors' margins. **The pass line, verbatim**: the model names credible conditions for
sustained commodity participation and names where it fails; a result needing a small network, token appreciation or
scheduled ASIC death has not passed. The development cost is SUNK in the mandatory case (the opponent covers
manufacturing, deployment and operation only); the paid-development cases sit beside it.
## 0. The result in one page
1. **On today's rows the N2 SRAM die fails the success statement in every scenario where its owner can buy a fleet
worth a few percent of the chain's yearly miner revenue.** With development sunk, a 3-year life and power at USD
0.06 per kWh, the die's all-in cost per accepted MH/s-hour is 72 micro-USD against the best GPU owner's 156 to 204
at the same electricity (the 5070 Ti and 5080 at their knees, hardware sunk) and 415 to 588 for a new entrant: 2.2x
to 2.8x on the existing owner, 5.8x to 8.2x on the entrant. At the GPU's more likely 0.12 per kWh it is 3.6x to
4.6x and 7.2x to 9.9x; at 0.25, 6.8x to 8.5x and 10x to 14x. A USD 10 M fleet of such dies takes 37 to 73 percent
of the chain's hash the year it lands at IGN 0.10 to 0.20 and 100 percent the year after on a flat price; a USD
100 M fleet takes 100 percent on landing in every path and then runs at a loss because it is larger than the
revenue (the self-limiting point: -1 to -300 percent margins).
2. **The GDDR7 board chip passes the success statement on its 1-year life and fails it on its 3-year life.** At 1 year
its cost per accepted unit (750 micro-USD at 0.06) is ABOVE every Blackwell owner's and above the Blackwell
entrant's at 0.06 to 0.12 (0.8x to 1.0x the 5080 entrant, 0.6x to 0.7x the 5070 Ti), so it competes only against
Ada and Ampere at dear electricity; at 3 years (307 micro-USD) it is 1.9x to 2.3x the Blackwell entrant and 0.9x
to 1.1x the Blackwell owner: GPUs at their knee stay close enough. Its hardware is the term that holds it (USD 5.6
per MH/s with the core and the system against the die's 0.8), not its joules.
3. **The electricity axis is explicit and it is the GPU's.** The break-even electricity price above which an EXISTING
GPU owner with sunk hardware cannot match the die at 0.06 is 0 to 5 cents per kWh for every card (a 3-year die) and
1 to 5 cents (a 1-year die): no grid price in the world keeps a GPU owner level with a sunk SRAM die. Against the
GDDR7 board at 3 years the owner's break-even is 9 to 15 cents on Blackwell, 4 to 6 on Ada and Ampere; at 1 year 27
to 40 cents on Blackwell. Read the other way, the die breaks even against a 5080 owner at 0.12 only when the die
pays 47 to 60 cents per kWh; the board at 3 years when it pays 1 to 9 cents.
4. **Accessible supply is the structural fact.** At the GPU entry equilibrium (revenue per MH/s-hour equal to a new
5070 Ti's cost at 0.12) the chain's hash is 2.6 TH/s at IGN 0.03, 8.8 at 0.10, 26 at 0.30, 88 at 1.00: that is
34,000 to 1.1 M 5070 Ti-class cards, which exist in the world's installed base, against 480 to 16,000 SRAM dies, 8
to 270 wafers of N2. One supplier holds the chain at every price in the window; the dependence on individual
suppliers is total for the die and partial for the board (16,000 to 530,000 boards, a Bitmain-class run).
5. **The conditions under which the success statement holds**, read off the tables: (a) the chip's all-in cost per
accepted unit stays within about 1.5x of the best GPU owner's at the same electricity, which on today's rows is true
of the GDDR7 board at a life of 1 to 2 years and false of the die at every life; (b) the chip's hardware per MH/s is
not below about a quarter of the GPU's annualised hardware (the board's 0.7x to 1.9x the 5080 entrant passes, the
die's 5x to 14x fails); (c) no single buyer can fund a fleet above about a third of the chain's hash for under a
year's miner revenue (true for the board above IGN 0.3; false for the die at every price: a wafer is USD 30,000);
(d) GPUs keep a resale market and a use outside mining (true for every card in the population; the chip has none,
which is why its life is the axis that moves everything); (e) the per-joule gap at the honest knee stays under about
3x (the board at 2.2x to 2.6x against Blackwell passes; the die at 3.7x to 4.5x does not).
6. **What the chain controls, and what it does not.** It controls the honest cost per accepted unit (the operating
point: the lock is worth 34 to 41 percent of a Blackwell card's draw), the chip's life against a fixed lane (the
180-day rotation; nothing against a programmable one), and the visibility of a concentrated supplier (the share
detector). It does not control the sunk development cost, electricity prices, the token price or a buyer's budget.
On today's rows the SRAM die, once it exists, cannot be held to coexistence by anything in the hash; the board can.
The condition that holds the die is that it does not get built, which is the surface of section 4.4 (a USD 150 M
project at a third of the chain needs IGN 0.73 over three years), and that is a statement about who pays, not about
the chain staying below a level.
1. **The DRAM-board chip (a GPU's memory system with a programmable core) passes the success statement on today's
rows; the N2 SRAM die fails it once it exists with its development sunk.** With development sunk, a 3-year life
and power at USD 0.06 per kWh, the board's cost per accepted MH/s-hour is 307 micro-USD against the best GPU
owner's 156 to 204 at the same electricity and 415 to 588 for a new entrant (0.5x to 0.7x the owner, 1.4x to 1.9x
the entrant); the die's is 72 (2.2x to 2.8x the owner, 5.8x to 8.2x the entrant; at the GPU's 0.12, 3.6x to 4.6x
and 7.2x to 9.9x). In the five-year runs with a per-class supply curve the board at a sunk USD 10 M holds 4.5 to
24 percent of the chain at a 24 to 69 percent margin with 6 to 16 of 17 GPU classes above water in every path; a
USD 10 M die fleet holds 50 to 70 percent on landing and takes the chain by year 3 to 5 on flat and shrinking paths
(0 of 17 classes above water); a USD 100 M die fleet takes it on landing in every path and runs at a loss.
2. **The tariff advantage is explicit and separate from the hardware advantage (section 1).** The review's 6.25x
illustration (a 1.5x chip at 0.06 against a GPU at 0.25) is the first row; on the measured rows the operating
advantage is the joule ratio times the tariff ratio (2.2x to 4.5x times 1x to 4.2x for the board, 3.7x to 7.8x for
the die), the hardware advantage 1.3x to 4x for the board and 9x to 29x for the die, and the total a third to a
half of the operating figure because power is 60 to 70 percent of an owner's cost and 30 to 40 of an entrant's.
3. **The break-even electricity price is tabled for all 17 classes (section 3).** A GPU owner with sunk hardware
matches the board at 3 years up to 9 to 15 cents per kWh on Blackwell and 4 to 6 on Ada and Ampere; at 1 year up
to 27 to 40 cents; it matches the die at 0 to 5 cents. No GPU ENTRANT matches the board at 3 years or the die at
any life at any positive price except the 5070 Ti against the 1-year board (25 cents): the entrant rows are where
the GPU side loses first.
4. **Proving is the second income the chip does not have (section 5).** A 16 GB or larger card earns USD 3 to 7 a
day from internal proving at IGN 0.10 (the H100 16, the A100 9) against USD 0.2 to 1.7 a day from mining at the
GPU equilibrium, and USD 5 to 12 under a proving spike; the SRAM die, the DRAM board, the 9070 XT and the Mac earn
nothing from it, and the 12 GB tier earns nothing from internal proving today (measured). A GPU displaced from
mining by a chip goes to proving (the operator simulation's D1: 1,500 of 9,177 cards), which is the mechanism that
keeps commodity participation when the mining margin is gone.
5. **Accessible supply and dependence (section 11).** At the GPU equilibrium the chain's hash is 5.6 / 9.6 / 20 / 42
TH/s at IGN 0.03 / 0.10 / 0.30 / 1.00, which is 12 / 21 / 44 / 95 percent of the installed base available to
mining (44.7 TH/s, approximate) across 16 NVIDIA classes, AMD and Apple; the same hash is 17 / 29 / 60 / 128 N2
wafers from one supplier, or 34,000 to 255,000 DRAM boards. Dependence on a single supplier is total for the die,
partial for the board, nil for the GPU side.
6. **The conditions under which the success statement holds (section 12)**, each with its number: (a) the chip's
all-in cost within about 1.5x of the best GPU owner's at the GPU's electricity; (b) the chip's hardware per MH/s
not under about a quarter of the GPU entrant's; (c) a fleet above a third of the chain costing more than a year's
miner revenue; (d) GPUs keeping a resale market and a use outside mining; (e) the per-joule gap at the honest knee
under about 3x; (f) the supplier's margin a normal return that does not rise with the halvings; (g) a second
income (proving) for the commodity side that the specialised supplier cannot enter. The board passes all seven at
a 1 to 3 year life; the die fails (a), (b), (c), (e) and (f) at every life, price path and electricity price once
it exists. **Against the pass line**: the board's pass does not need a small network (it holds at 42 TH/s and IGN
1.00), token appreciation (it holds on the flat and shrinking paths) or scheduled ASIC death (its life axis is 1 to
5 years and the 180-day rotation is not what holds it); the die's failure is not cured by any of the three either,
and the only condition that holds the die is that nobody pays to build it (the surface: IGN 0.73 for a USD 150 M
project at a third of the chain over three years), which is an investor's decision, not a level the chain stays
below. The model names where it fails: a sunk SRAM die of USD 10 M or more at any price in the window.
## 1. The measure: cost per accepted unit of work
## 1. The tariff advantage beside the hardware advantage (D4: the electricity axis explicit)
Cost per accepted MH/s-hour (micro-USD), both sides: annualised hardware plus power plus hosting, failures and fees,
divided by accepted work. Accepted work is 97 percent of raw (rejects 0.5 percent, downtime 2.0, epoch preparation and
propagation 0.5; approximate from the devnet's share rates), the pool fee 1 percent. The two GPU situations: the
**existing owner** (hardware sunk; pays power, a wear allowance of 5 percent of the used price a year, fees; the
alternative use is the card's rental yield, reported in section 2 and not deducted) and the **new entrant** (buys new
or used, operates two years, resells at the table's fraction). Hosting: 0 at home, USD 0.02 per kWh-equivalent at a
farm (the chip's case). Electricity axis: USD 0.06, 0.12, 0.25 per kWh.
The chip at 0.06 per kWh with farm hosting; the GPU at 0.06, 0.12 and 0.25. "Operating" is joules times tariff;
"hardware" the GPU entrant's annualised hardware over the chip's; the total against the existing owner and the entrant.
| Input | Value | Label |
|---|---|---|
| Card joules per hash at the floor (the knee with the knobs) and at stock | lane 4's class v5 table: 5090 2.33 / 3.48, 5080 2.06 / 3.48, 5070 Ti 1.70 / 2.84, 5070 1.75 / 2.99, 5060 Ti 2.36 / 4.02, 5060 2.21 / 3.76, 4090 3.58 / 5.00, 4080 3.51 / 4.92, 4070 3.58 / 5.82, 4060 Ti 3.81 / 5.32, 3090 4.51 / 5.03, 3080 4.20 / 4.54, 3060 5.77 / 6.40, 9070 XT 7.90 / 10.7, H100 2.00 / 2.58, A100 2.89 / 2.99, M5 Max 1.40 microjoules | measured where lane 4 says so (5090, 5080, 4070, 9070 XT, M5 Max floors; most stock rows), modelled knees elsewhere |
| Card rates at the floor | 5090 134.8, 5080 71.2, 5070 Ti 77, 5070 41, 5060 Ti 19, 5060 17, 4090 58, 4080 45, 4070 31.1, 4060 Ti 17.6, 3090 37.8, 3080 40.8, 3060 23.8, 9070 XT 18.9, H100 90, A100 60, M5 Max 27.1 MH/s | measured where the bench table has the row; approximate elsewhere |
| Prices new / used, resale after two years | 5090 2,600 / 2,200 / 55 percent; 5080 1,100 / 900 / 50; 5070 Ti 800 / 650 / 50; 5070 560 / 450 / 50; 5060 Ti 450 / 360 / 45; 5060 310 / 250 / 45; 4090 1,700 / 1,300 / 45; 4080 1,000 / 700 / 40; 4070 550 / 400 / 40; 4060 Ti 420 / 290 / 35; 3090 900 / 650 / 30; 3080 450 / 330 / 25; 3060 260 / 190 / 25; 9070 XT 650 / 520 / 45; H100 25,000 / 18,000 / 50; A100 10,000 / 6,000 / 35; M5 Max 4,000 / 3,200 / 55 | approximate street, October 2026 |
| Chip joules per hash (class v5, the shadow at 3.2 pJ per forced op, lane 4's chip columns) | GDDR7 board with an N5 core 0.79; HBM3 one stack 0.65; N2 SRAM die with the core 0.46 (W = 4); the die at the bare-lane floor 0.165 | modelled |
| Chip hardware, USD per MH/s, silicon and board plus 30 percent system (PSU, chassis, cooling) | GDDR7 board 5.6 (4.3 with the core die, chip-model 5.5 and research 16.1); HBM3 8.6; SRAM die 0.8 (0.6 with the board, lane B) | modelled, approximate |
| Chip failures, resale | 3 percent a year; no resale (single use) | approximate |
| Chip lives | 0.5, 1, 2, 3, 5 years (0.5 is the fixed-lane chip under the 180-day rotation; 3 the programmable chip's default) | the review's axis |
| Emission to miners | 0.77 / 0.80 / 0.40 / 0.40 / 0.20 B IGN in years 1 to 5 (the spec's constant, 80 percent to miners) | spec 05 |
| GPU equilibrium | while any GPU mines, revenue per MH/s-hour settles at the cheapest entrant's cost (a new 5070 Ti at 0.12: 521 micro-USD) during growth and falls to the owners' costs during shrinkage; a GPU generation at year 3 cuts the entrant's cost 25 percent (1.5x per joule at the same price) with the old card resold at the table's fraction | modelled rule |
| Row | Joules ratio | GPU tariff over chip tariff | Operating advantage | Hardware advantage | Total vs owner / entrant | Label |
|---|---|---|---|---|---|---|
| The review's illustration: a 1.5x chip at 0.06 against a GPU at 0.25 | 1.5x | 4.17x | **6.25x** | n/a | n/a | the review |
| GDDR7 board, 3 y, vs 5070 Ti at 0.06 / 0.12 / 0.25 | 2.2x | 1x / 2x / 4.2x | 2.2x / 4.3x / 9.0x | 1.3x | 0.5x / 1.4x; 0.9x / 1.7x; 1.6x / 2.4x | modelled |
| GDDR7 board, 3 y, vs 5080 | 2.6x | the same | 2.6x / 5.2x / 10.9x | 1.9x | 0.7x / 1.9x; 1.1x / 2.3x; 2.0x / 3.2x | modelled |
| GDDR7 board, 3 y, vs 4090 | 4.5x | | 4.5x / 9.1x / 18.9x | 4.0x | 1.2x / 3.8x; 1.9x / 4.6x; 3.5x / 6.2x | modelled |
| N2 SRAM die, 3 y, vs 5070 Ti | 3.7x | | 3.7x / 7.4x / 15.4x | 9.3x | 2.2x / 5.8x; 3.6x / 7.2x; 6.8x / 10.4x | modelled |
| N2 SRAM die, 3 y, vs 5080 | 4.5x | | 4.5x / 9.0x / 18.7x | 13.8x | 2.8x / 8.2x; 4.6x / 9.9x; 8.5x / 13.8x | modelled |
| N2 SRAM die, 3 y, vs 4090 | 7.8x | | 7.8x / 15.6x / 32.4x | 28.7x | 5.0x / 16.4x; 8.1x / 19.5x; 14.8x / 26.2x | modelled |
## 2. The GPU reference population: cost per accepted MH/s-hour (micro-USD)
Reading: the tariff multiplies the operating advantage one for one, as the review says, and the total is a third to a
half of it; the board's total against a Blackwell card at its knee is under 1x (owner) to 2.4x (entrant) across the
whole axis at 3 years, the coexistence band; the die's 2.2x to 14x is not.
| Card | Owner at 0.06 / 0.12 / 0.25 | Entrant, new, 2 years, at 0.06 / 0.12 / 0.25 | Entrant, used | Hardware share of the entrant's cost at 0.12 | Wh per MH/s-hour | Alternative use (rental yield, approximate) | Label |
## 2. Cheap and dear electricity: the GPU population's cost per accepted unit (D4: cheap and dear electricity)
Cost per accepted MH/s-hour (micro-USD; accepted = 97 percent of raw: rejects 0.5, downtime 2.0, epoch preparation
and propagation 0.5; pool fee 1 percent). The existing owner (hardware sunk; power, 5 percent wear, fees) and the new
entrant (buys new or used, runs two years, resells at the table's fraction).
| Card | Owner at 0.06 / 0.12 / 0.25 | Entrant, new, at 0.06 / 0.12 / 0.25 | Entrant, used | Hardware share of the entrant at 0.12 | Wh per MH/s-hour | Alternative use (rental yield, approximate) | Label |
|---|---|---|---|---|---|---|---|
| RTX 5090 | 243 / 388 / 704 | 661 / 807 / 1,122 | 800 / 945 / 1,261 | 64 percent | 2.33 | USD 0.3 to 0.5 an hour on a rental market: 2,200 to 3,700 micro-USD per MH/s-hour, 3x to 5x its mining cost | measured floor |
| RTX 5090 | 243 / 388 / 704 | 661 / 807 / 1,122 | 800 / 945 / 1,261 | 64 percent | 2.33 | USD 0.3 to 0.5 an hour, 3x to 5x its mining cost | measured floor |
| RTX 5080 | 204 / 332 / 611 | 588 / 716 / 995 | 650 / 779 / 1,058 | 64 | 2.06 | USD 0.15 to 0.25 an hour | measured floor |
| RTX 5070 Ti | 156 / 263 / 493 | 415 / 521 / 751 | 453 / 560 / 790 | 59 | 1.70 | USD 0.1 to 0.2 an hour | modelled knee |
| RTX 5070 | 175 / 284 / 521 | 515 / 624 / 861 | 558 / 668 / 905 | 65 | 1.75 | | modelled knee |
@ -103,197 +108,254 @@ farm (the chip's case). Electricity axis: USD 0.06, 0.12, 0.25 per kWh.
| RX 9070 XT | 657 / 1,151 / 2,220 | 1,617 / 2,111 / 3,180 | 1,668 / 2,162 / 3,231 | 53 | 7.90 | | measured |
| H100 (hosted) | 1,313 / 1,438 / 1,709 | 8,374 / 8,499 / 8,770 | 7,880 / 8,004 / 8,275 | 97 | 2.00 | USD 2 to 3 an hour: never mines | modelled lock |
| A100 (used) | 775 / 955 / 1,346 | 6,615 / 6,796 / 7,187 | 4,388 / 4,568 / 4,960 | 95 | 2.89 | USD 1 an hour: never mines | modelled |
| Apple M5 Max (reported, not headlined) | 789 / 876 / 1,066 | 4,033 / 4,120 / 4,310 | 4,690 / 4,778 / 4,967 | 96 | 1.40 | a workstation: mines only as an owner | measured |
| Apple M5 Max (reported, not headlined) | 789 / 876 / 1,066 | 4,033 / 4,120 / 4,310 | 4,690 / 4,778 / 4,967 | 96 | 1.40 | a workstation: an owner only | measured |
Reading: the owner's cost is 60 to 70 percent electricity on every consumer card, so the electricity price is the
GPU's whole variable; the entrant's cost is 60 to 70 percent hardware, so the card price and its resale are the
entrant's whole variable. The best honest owner on today's rows is a 5070 Ti at its knee (156 micro-USD at 0.06); the
best entrant the same card (415). Datacentre parts and the Mac never enter as entrants (hardware 95 percent) and mine
only as owners with nothing better to do, which their rental yields say they always have.
The chip rows, development sunk (the mandatory case), farm hosting USD 0.02 per kWh-equivalent on top:
## 3. The chip rows
### 3.1 Development sunk (the mandatory case): cost per accepted MH/s-hour, micro-USD, farm hosting
| Chip | Life 0.5 y at 0.06 / 0.12 / 0.25 | 1 y | 2 y | 3 y | 5 y | Hardware / power split at 0.06, 3 y | Label |
| Chip | 0.5 y at 0.06 / 0.12 / 0.25 | 1 y | 2 y | 3 y | 5 y | Hardware / power at 0.06, 3 y | Label |
|---|---|---|---|---|---|---|---|
| GDDR7 board with an N5 core | 1,414 / 1,463 / 1,570 | 750 / 799 / 906 | 418 / 467 / 574 | 307 / 356 / 463 | 219 / 268 / 375 | 239 / 65 | modelled |
| GDDR7 board with an N5 core | 1,414 / 1,463 / 1,570 | 750 / 799 / 906 | 418 / 467 / 574 | **307 / 356 / 463** | 219 / 268 / 375 | 239 / 65 | modelled |
| HBM3 one stack with a core | 2,123 / 2,164 / 2,252 | 1,104 / 1,145 / 1,233 | 594 / 635 / 723 | 425 / 465 / 553 | 289 / 329 / 417 | 367 / 54 | modelled |
| N2 SRAM die with the core | 226 / 255 / 317 | 134 / 163 / 225 | 87 / 116 / 178 | **72 / 101 / 163** | 60 / 88 / 151 | 33 / 38 | modelled |
| N2 SRAM die at the bare-lane floor | 202 / 212 / 235 | 109 / 120 / 142 | 63 / 73 / 96 | 47 / 58 / 80 | 35 / 45 / 68 | 33 / 14 | modelled, the worst case |
### 3.2 Development paid: the same rows with `C_dev` spread over the fleet and the life
## 3. Break-even electricity prices for all 17 classes (D4: the output per class)
The fleet is sized to a share `q` of the network's hash at the GPU equilibrium (a new 5080 at 0.12 as the marginal
entrant). All-in cost per accepted MH/s-hour of the SRAM die (the GDDR7 board in brackets), at 0.06:
Cents per kWh at which the GPU's cost per accepted unit equals the chip's all-in at 0.06 per kWh; the OWNER figure
(hardware sunk) and the ENTRANT figure (hardware bought); "under 0" means no positive price matches.
| IGN price | Miner revenue (year 3) | Network hash | `C_dev` 20 M, 1 y, `q` 0.3 / 1.0 | 20 M, 3 y, 0.3 / 1.0 | 150 M, 1 y, 0.3 / 1.0 | 150 M, 3 y, 0.3 / 1.0 |
|---|---|---|---|---|---|---|
| 0.03 | USD 12 M | 1.9 TH/s | 4,277 / 1,377 (4,893 / 1,993) | 1,453 / 486 (1,688 / 721) | 31,211 / 9,457 | 10,431 / 3,180 |
| 0.10 | 40 M | 6.4 TH/s | 1,377 / 507 (1,993 / 1,123) | 486 / 196 (721 / 431) | 9,457 / 2,931 | 3,180 / 1,004 |
| 0.30 | 120 M | 19 TH/s | 548 / 258 (1,164 / 874) | 210 / 113 (445 / 349) | 3,242 / 1,066 | 1,108 / 383 |
| 1.00 | 400 M | 64 TH/s | 258 / 171 (874 / 787) | 113 / 84 (349 / 320) | 1,066 / 414 | 383 / 165 |
Reading: a paid development cost puts every chip ABOVE the best GPU entrant (415 to 588) at IGN 0.10 and below, and
the SRAM die below it only from IGN 0.30 on a 3-year life or IGN 1.00 on a 1-year life; the GDDR7 board with paid
development is never below the Blackwell entrant inside the window. The sunk case is therefore the whole threat, and
it is the case the review makes mandatory. The derivative design (a revision at 0.3 x `C_dev`, section 4.4 of the
floor file) moves the paid rows a third of the way to the sunk rows.
## 4. The cost advantage, separated
The chip at 0.06 per kWh and farm hosting; the GPU at 0.06, 0.12 and 0.25. "Operating" is power only (joules times
electricity); "hardware" the annualised hardware of the GPU entrant over the chip's.
| Chip, life | Against | Total: owner / entrant, at 0.06 | At 0.12 | At 0.25 | Operating advantage at 0.06 / 0.12 / 0.25 (joules x electricity) | Hardware advantage |
|---|---|---|---|---|---|---|
| GDDR7 board, 1 y | RTX 5080 | 0.3x / 0.8x | 0.4x / 1.0x | 0.8x / 1.3x | 2.6x / 5.2x / 10.9x (2.6 x 1, 2, 4.2) | 0.7x |
| | RTX 5070 Ti | 0.2x / 0.6x | 0.4x / 0.7x | 0.7x / 1.0x | 2.2x / 4.3x / 9.0x | 0.5x |
| | RTX 4090 | 0.5x / 1.6x | 0.8x / 1.9x | 1.4x / 2.5x | 4.5x / 9.1x / 18.9x | 1.4x |
| | RTX 3080 (used) | 0.4x / 1.0x | 0.8x / 1.4x | 1.5x / 2.1x | 5.3x / 10.6x / 22.2x | 0.7x |
| GDDR7 board, 3 y | RTX 5080 | 0.7x / 1.9x | 1.1x / 2.3x | 2.0x / 3.2x | the same | 1.9x |
| | RTX 5070 Ti | 0.5x / 1.4x | 0.9x / 1.7x | 1.6x / 2.4x | | 1.3x |
| | RTX 4090 | 1.2x / 3.8x | 1.9x / 4.6x | 3.5x / 6.2x | | 4.0x |
| | RTX 3080 (used) | 1.0x / 2.5x | 1.9x / 3.3x | 3.7x / 5.2x | | 2.1x |
| N2 SRAM die, 1 y | RTX 5080 | 1.5x / 4.4x | 2.5x / 5.4x | 4.6x / 7.4x | 4.5x / 9.0x / 18.7x | 4.9x |
| | RTX 5070 Ti | 1.2x / 3.1x | 2.0x / 3.9x | 3.7x / 5.6x | 3.7x / 7.4x / 15.4x | 3.3x |
| | RTX 4090 | 2.7x / 8.8x | 4.3x / 10.5x | 8.0x / 14.1x | 7.8x / 15.6x / 32.4x | 10.1x |
| N2 SRAM die, 3 y | RTX 5080 | 2.8x / 8.2x | 4.6x / 9.9x | 8.5x / 13.8x | the same | 13.8x |
| | RTX 5070 Ti | **2.2x / 5.8x** | **3.6x / 7.2x** | 6.8x / 10.4x | | 9.3x |
| | RTX 4090 | 5.0x / 16.4x | 8.1x / 19.5x | 14.8x / 26.2x | | 28.7x |
| | RTX 3080 (used) | 4.3x / 10.5x | 8.0x / 14.1x | 15.9x / 22.0x | | 14.7x |
| | Apple M5 Max (reported) | 11.0x / 56x | 12.2x / 57x | 14.8x / 60x | 3.0x / 6.1x / 12.7x | 118x |
Reading: the electricity axis multiplies the operating advantage one for one (a 2.6x chip at 0.06 against a GPU at
0.25 is 10.9x on power alone, the review's point), but on the consumer cards power is 60 to 70 percent of the owner's
cost and 30 to 40 percent of the entrant's, so the total advantage is a third to a half of the operating one. The
GDDR7 board's total advantage over a Blackwell card at its knee is under 1x (owner) to 2.3x (entrant) across the whole
electricity axis at a 3-year life, which is the coexistence band; the die's is 2.2x to 14x, which is not.
## 5. The replacement economics
For an existing GPU owner, switching pays when the chip's all-in cost per accepted unit is below the owner's
OPERATING cost (the hardware is sunk, the resale value is the only thing the switch recovers). For a new entrant, when
the chip's all-in is below the GPU entrant's all-in. The chip must be purchasable for either (hardware sales; the
manufacturer keeps about half the operator's profit through the price, floor file 4.4 table B, which roughly doubles
the chip's hardware term for the buyer).
| Who | Against the GDDR7 board (bought, hardware term x2) | Against the SRAM die (bought, x2) | What it means |
|---|---|---|---|
| A Blackwell owner at 0.06 to 0.12 | never switches: the bought board at 3 years is 550 to 600 micro-USD against the owner's 156 to 332 | switches at 0.12 (the bought die 105 to 134 against 263 to 332) and is near indifferent at 0.06 (105 against 156 to 204) | the die replaces Blackwell owners at normal grid prices; the board never does |
| An Ada or Ampere owner at 0.12 | near indifferent at 3 years (550 to 600 against 524 to 768); switches at 0.25 | switches at every electricity price | the board retires the oldest cards only at dear electricity, which the generation upgrade does anyway |
| A new entrant choosing between a new 5070 Ti and a bought chip at 0.12 | the board at 3 years (about 600) is 1.15x the card's 521: the card wins; at 1 year the card wins 2x | the bought die (134) is 0.26x the card: the die wins 4x | an entrant market with a bought SRAM die has no GPU entrants; one with a bought board keeps them |
| The GPU generation upgrade (year 3: 1.5x per joule at the same price, the old card resold) | cuts the entrant's cost about 25 percent and the owner's power 33 percent: the board at 3 years then reads 1.5x to 1.7x the new entrant, still in band | the die's advantage falls 25 to 33 percent, from 7x to 5x on the entrant: still out of band | the GPU side's own curve narrows the board's gap to nothing by the second generation and never closes the die's |
## 6. Break-even electricity prices
(a) The GPU electricity price above which an EXISTING owner (hardware sunk) cannot match the chip's all-in cost at
0.06 per kWh:
| Chip, life | 5090 | 5080 | 5070 Ti | 5070 | 5060 Ti | 4090 | 4080 | 4070 | 3090 | 3080 | 3060 | 9070 XT | M5 Max |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GDDR7 board, 1 y | 27 c | 32 c | 40 c | 38 c | 26 c | 17 c | 18 c | 18 c | 14 c | 16 c | 12 c | 7 c | 3 c |
| GDDR7 board, 3 y | 9 c | 11 c | 15 c | 13 c | 8 c | 5 c | 6 c | 6 c | 4 c | 6 c | 4 c | 2 c | under 0 |
| SRAM die, 1 y | 1.5 c | 2.7 c | 4.7 c | 3.8 c | 0.9 c | 0 c | 1.1 c | 1.5 c | 0.7 c | 2.0 c | 1.4 c | under 0 | under 0 |
| SRAM die, 3 y | under 0 | under 0 | 1.2 c | 0.4 c | under 0 | under 0 | under 0 | under 0 | under 0 | 0.5 c | 0.4 c | under 0 | under 0 |
(b) The chip electricity price at which its all-in cost equals a GPU owner's at 0.12 (how much dearer the chip's
hosting can be and still match): the GDDR7 board at 3 years 1 to 9 cents against Blackwell (it must be hosted cheaper
than the GPU to match an owner) and 75 cents against the Mac; at 1 year it cannot match a Blackwell owner at any
price. The SRAM die matches a 5070 Ti owner while paying up to 46 cents (3 years) or 33 cents (1 year), a 5080 owner
up to 60 or 47, the Mac up to 174.
Reading: the board lives or dies on the GPU's grid price, which is the review's electricity axis doing what it should;
the die does not see the axis at all.
## 7. Five years: growing, flat and shrinking networks, with the GPU side reacting
The sunk chip enters at the start of year 2 with a fleet bought for a budget `B` (USD 1 M, 10 M, 100 M at USD 0.8
per MH/s for the die: 1.3, 13 and 128 TH/s); revenue per MH/s-hour `r` settles at the cheapest GPU entrant's cost
(a new 5070 Ti at 0.12, 521 micro-USD; 391 after the year-3 generation) while entry continues, and when the chip's
fleet alone exceeds the hash that revenue supports, GPU owners exit in cost order until the survivors' costs are
covered or none are. Three price paths: growing (x2 a year from 0.10), flat (0.10), shrinking (x0.5 a year from
0.30). Emission halves in year 3 and year 5.
| Path, budget | Year 2 | Year 3 | Year 5 | Reading |
| Class | GDDR7 board 1 y: owner / entrant | GDDR7 board 3 y | SRAM die 1 y | SRAM die 3 y |
|---|---|---|---|---|
| Growing, USD 1 M | chip 3.7 percent of 35 TH/s; `r` 521; margin 86 percent | 2.7 percent; owners all above water | 1.4 percent of 93 TH/s | coexistence: the supplier earns 82 to 86 percent margins on a tiny share, GPUs set the price |
| Growing, USD 10 M | 37 percent of 35 TH/s | 27 percent | 14 percent | coexistence by dilution only: the share falls as the chain grows and the fleet is fixed; the supplier's margin stays 82 to 86 percent, far above normal |
| Growing, USD 100 M | 100 percent; 0 of 17 owner classes above water; `r` falls to 142; margin 49 percent | the same | 100 percent; 2 of 17 owner classes above water at `r` 285 | failure: one buyer holds the chain for five years, GPUs exit in year 2 and only the two best classes could return in year 4 |
| Flat, USD 1 M | 7 percent of 17.5 TH/s | 11 percent | 22 percent of 5.8 TH/s | coexistence, the share rising with each halving |
| Flat, USD 10 M | 73 percent | 100 percent; 3 of 17 owner classes above water | 100 percent; 0 of 17 | failure by year 3: the halving does the rest |
| Flat, USD 100 M | 100 percent; margin -1 percent | margin -102 percent | -305 percent | failure for both: the fleet is larger than the revenue; the buyer loses money and the GPUs are gone (the self-limiting point) |
| Shrinking, USD 1 M | 5 percent | 15 percent | 100 percent; 3 of 17 above water | failure in year 5 at USD 4 M of revenue: even a USD 1 M fleet is the chain when the chain is small |
| Shrinking, USD 10 M | 49 percent | 100 percent; 1 of 17 | 100 percent; margin -116 percent | failure from year 3 |
| The GDDR7 board (sunk), flat, USD 10 M | 10 percent of 17.5 TH/s; margin 41 percent | 15 percent; margin 21 percent | 31 percent; margin 21 percent | coexistence: a normal return (21 to 41 percent gross) on a minority share with GPU entrants still setting the price |
| RTX 5090 | 27 / 10 | 9 / under 0 | 1.5 / under 0 | under 0 / under 0 |
| RTX 5080 | 32 / 14 | 11 / under 0 | 2.7 / under 0 | under 0 / under 0 |
| RTX 5070 Ti | 40 / 25 | 15 / 0 | 4.7 / under 0 | 1.2 / under 0 |
| RTX 5070 | 38 / 19 | 13 / under 0 | 3.8 / under 0 | 0.4 / under 0 |
| RTX 5060 Ti 16 GB | 26 / under 0 | 8 / under 0 | 0.9 / under 0 | under 0 / under 0 |
| RTX 5060 | 29 / 7 | 10 / under 0 | 2.0 / under 0 | under 0 / under 0 |
| RTX 4090 | 17 / under 0 | 5 / under 0 | 0 / under 0 | under 0 / under 0 |
| RTX 4080 | 18 / under 0 | 6 / under 0 | 1.1 / under 0 | under 0 / under 0 |
| RTX 4070 | 18 / 3 | 6 / under 0 | 1.5 / under 0 | under 0 / under 0 |
| RTX 4060 Ti 16 GB | 16 / under 0 | 5 / under 0 | 0.9 / under 0 | under 0 / under 0 |
| RTX 3090 (used) | 14 / under 0 | 4 / under 0 | 0.7 / under 0 | under 0 / under 0 |
| RTX 3080 (used) | 16 / 6 | 6 / under 0 | 2.0 / under 0 | 0.5 / under 0 |
| RTX 3060 (used) | 12 / 4 | 4 / under 0 | 1.4 / under 0 | 0.4 / under 0 |
| RX 9070 XT | 7 / under 0 | 2 / under 0 | under 0 / under 0 | under 0 / under 0 |
| H100 (hosted) | under 0 / under 0 | under 0 / under 0 | under 0 / under 0 | under 0 / under 0 |
| A100 (used) | 5 / under 0 | under 0 / under 0 | under 0 / under 0 | under 0 / under 0 |
| Apple M5 Max | 3 / under 0 | under 0 / under 0 | under 0 / under 0 | under 0 / under 0 |
Reading: the row that passes the success statement as the review states it is the GDDR7 board at a sunk USD 10 M (a
21 to 41 percent gross margin, a 10 to 31 percent share, GPUs setting the price, entrants competing). The SRAM die
passes only at a fleet under about 1 percent of the chain's yearly miner revenue in a growing network, and fails in
every flat or shrinking path by year 3 to 5; above a tenth of a year's revenue it takes the chain in every path. The
failure is not a margin the die extracts (its margins collapse once it is the chain); it is the exit of every GPU
class, which is the review's definition.
Reading: the board at 1 year is beaten by every Blackwell owner below 26 to 40 cents and by the 5070 Ti entrant below
25; at 3 years by Blackwell owners below 8 to 15 cents and by no entrant. The die is matched by no entrant and by
owners only below 0 to 5 cents. The GPU side's electricity price is the board's whole variable and irrelevant to the
die.
## 8. Accessible supply and the dependence on individual suppliers
## 4. GPU replacement and resale on both sides, with the generation step on the chip side too (D4)
| IGN price | Network hash at the GPU entry equilibrium | In 5070 Ti-class cards | In 5090s | In N2 SRAM dies (5.5 GH/s) | In N2 wafers (about 60 good dies) | In GDDR7 boards (166 MH/s) |
|---|---|---|---|---|---|---|
| 0.03 | 2.6 TH/s | 34,000 | 19,500 | 480 | 8 | 15,800 |
| 0.10 | 8.8 | 114,000 | 65,000 | 1,600 | 27 | 52,800 |
| 0.30 | 26 | 341,000 | 195,000 | 4,800 | 80 | 158,000 |
| 1.00 | 88 | 1.1 M | 650,000 | 16,000 | 270 | 528,000 |
| 3.00 | 263 | 3.4 M | 1.9 M | 48,000 | 800 | 1.6 M |
The GPU side's accessible supply is the installed base of consumer cards (tens of millions of Ampere, Ada and
Blackwell cards in the world, approximate) and the used market, with a use outside mining and a resale price that the
population table carries; at every price in the window the hash the chain needs is under 4 percent of one
generation's shipments. The die's supply is one supplier's wafer allocation (8 to 800 wafers of a node booked to 2028,
claimed); the board's is a Bitmain-class production run (16,000 to 1.6 M units) with a commodity memory bill. The
dependence on individual suppliers is total for the die at every price (one order holds the chain), partial for the
board (a run of that size is visible and takes months), and nil for the GPU side.
## 9. The conditions under which the success statement holds (the result)
Success (a specialised supplier earns a normal return and GPUs stay close enough in total cost, obtainable and useful
outside mining, that entrants still compete) holds on today's rows when ALL of the following do:
| Condition | The number on today's rows | GDDR7 board | SRAM die |
| Who | Against the DRAM board (bought at twice the hardware term, the manufacturer's half) | Against the SRAM die (bought, the same) | What it means |
|---|---|---|---|
| (a) the chip's all-in cost per accepted unit at its own electricity is within about 1.5x of the best GPU owner's at the GPU's electricity | the owner at 0.12: 263 to 332 micro-USD (5070 Ti, 5080) | 307 at 3 years, 750 at 1 year: PASSES | 72 to 134: FAILS at every life |
| (b) the chip's annualised hardware per MH/s is not below about a quarter of the GPU entrant's | the 5080 entrant's hardware 460 micro-USD | 239 to 677: PASSES | 33 to 95: FAILS |
| (c) a fleet above a third of the chain's hash costs more than a year's miner revenue | at IGN 0.10 the chain is 8.8 TH/s; a third is 2.9 TH/s | USD 16 M of boards against USD 40 to 80 M of revenue: PASSES above IGN 0.05 | USD 2.3 M of dies: FAILS at every price under about 3 |
| (d) GPUs keep a resale market and a use outside mining | every card in the population resells at 25 to 55 percent after two years and rents at 3x to 5x its mining cost | PASSES (the GPU side's property) | PASSES (the same) |
| A Blackwell owner at 0.06 to 0.12 | never switches: the bought board at 3 years is 550 to 600 micro-USD against the owner's 156 to 332 | switches at 0.12 (105 to 134 against 263 to 332); near indifferent at 0.06 | the die replaces Blackwell owners at normal grid prices; the board never does |
| An Ada or Ampere owner at 0.12 | near indifferent at 3 years (550 to 600 against 524 to 768); switches at 0.25 | switches at every price | the board retires only the oldest cards at dear electricity, which the generation does anyway |
| A new entrant choosing between a new 5070 Ti and a bought chip at 0.12 | the board at 3 years (about 600) is 1.15x the card's 521: the card wins; at 1 year the card wins 2x | the bought die (134) is 0.26x the card: the die wins 4x | an entrant market with a bought die has no GPU entrants; one with a bought board keeps them |
| The GPU generation at year 3 (1.5x per joule at the same price, the old card resold at the table's fraction) | cuts the entrant's cost 25 percent and the owner's power 33 percent: the board at 3 years then reads 1.5x to 1.7x the new entrant, in band | the die's advantage falls 25 to 33 percent, from 7x to 5x on the entrant: still out of band | the GPU side's own curve narrows the board's gap by the second generation and never closes the die's |
| The chip's generation at year 3 (a node step, 1.5x per joule, re-bought at the same silicon price) | the board's power term falls a third (65 to 43 micro-USD of 307): 2 percent of its cost; its hardware term is unchanged | the die's 38 to 25: 4 percent of 72 | the chip side's generation moves the totals under 5 percent: the chip's cost is hardware, the GPU's is power, so the generation helps the GPU more |
| Resale | the GPU resells at 25 to 55 percent after two years (section 2); the chips at 0 | | the resale market is the GPU entrant's whole hedge and the chip has none, which is why the chip's life is the axis that moves everything (section 7) |
## 5. Changing proving demand: proving revenue as a second income axis per class (D4: new)
The resolution's shape (the research lane, 17:0x UK): the tip stays whole to the miner; provers are paid the 20
percent pool per block plus an explicit user-funded proving fee with congestion pricing; the burn separate; the hard
cap and no development tax kept. The internal pool at IGN 0.10 is USD 27,400 a day (0.2 x 0.77 B / 0.8 / 365), shared
by proving capacity (a 5,000-card fleet drawn from the classes that can prove, approximate); external demand USD
2,000 a day at launch (spec 05's grid), 20,000 under a spike, 90 percent to the provers who deliver. The 12 GB tier's
internal row is the fleet lane's measured zero (0 paid in 313 claims on Devnet 3, 8 October 2026).
| Class | Memory | Shard s (bench) | Internal proving, USD per card-day | Plus external at launch | Plus external at a 10x spike | Mining at the GPU equilibrium, USD per card-day | Power per day at 0.12 | Label |
|---|---|---|---|---|---|---|---|---|
| RTX 5090 | 32 | 6.3 | 7.42 | 7.90 | 12.29 | 1.69 | 0.86 | modelled on measured shard times |
| RTX 5080 | 16 | 8.0 | 5.84 | 6.22 | 9.68 | 0.89 | 0.63 | the same |
| RTX 5070 Ti | 16 | 7.0 | 6.67 | 7.11 | 11.06 | 0.96 | 0.58 | the same |
| RTX 5070 | 12 | 4.8 | **0 (measured zero)** | 0.64 | 6.40 | 0.51 | 0.40 | measured zero, modelled external |
| RTX 5060 Ti 16 GB | 16 | 11.6 | 4.03 | 4.29 | 6.68 | 0.24 | 0.26 | modelled |
| RTX 5060 | 8 | 12.0 | 0 (measured zero) | 0.26 | 2.56 | 0.21 | 0.23 | |
| RTX 4090 | 24 | 6.3 | 7.42 | 7.90 | 12.29 | 0.73 | 0.81 | |
| RTX 4080 | 16 | 7.5 | 6.23 | 6.64 | 10.33 | 0.56 | 0.63 | |
| RTX 4070 | 12 | 12.1 | 0 (measured zero) | 0.25 | 2.54 | 0.39 | 0.32 | |
| RTX 4060 Ti 16 GB | 16 | 11.6 | 4.03 | 4.29 | 6.68 | 0.22 | 0.22 | |
| RTX 3090 (used) | 24 | 14.9 | 3.14 | 3.34 | 5.20 | 0.47 | 0.66 | |
| RTX 3080 (used) | 10 | 7.1 | 0 (measured zero) | 0.43 | 4.33 | 0.51 | 0.60 | |
| RTX 3060 (used) | 12 | 14.4 | 0 (measured zero) | 0.21 | 2.13 | 0.30 | 0.30 | |
| RX 9070 XT | 16 | none | cannot prove (no CUDA) | | | 0.24 | 0.43 | |
| H100 (hosted) | 80 | 3.0 | 15.57 | 16.60 | 25.81 | 1.13 | 1.01 | |
| A100 (used) | 80 | 5.0 | 9.34 | 9.96 | 15.49 | 0.75 | 0.72 | |
| Apple M5 Max | 36 | none | cannot prove | | | 0.34 | 0.11 | |
| The SRAM die, the DRAM board, any hash engine | | none | **0: a hash engine cannot prove** | 0 | 0 | the whole chain's mining | | by construction |
Reading: at launch-shape demand a proving-capable card earns 4x to 8x more per day from internal proving than from
mining at the GPU equilibrium, and the pool is shared by few enough cards that it pays even at 0.25 per kWh; under a
spike the external fee adds 50 to 70 percent. The chip has none of it. At zero proving demand (the pool is a launch
subsidy and the external market empty) the second income is 0 and the commodity side falls back to mining alone,
which is the first cut's model; at the fleet lane's measured efficiency (5.5 percent of proving card-time paid on a
day with a fault) the internal rows are 0.4 to 0.9 USD a day, still above mining for the 24 GB and datacentre
classes. The condition this adds to section 12 is (g): a second income for the commodity side that the specialised
supplier cannot enter; it holds while proving demand exists and the 16 GB and larger tiers can prove.
## 6. Private mining and hardware sales; cheaper derivative chips (D4)
From the surface (the floor file 4.4, the mission lane's shape), `p*` is the break-even price in USD per IGN:
| Entrant | `C_dev` 20 M (a DRAM board), `L` 3 y, `q` 0.3 / 1.0 | 150 M (the SRAM die), 3 y, 0.3 / 1.0 | Reading |
|---|---|---|---|
| The operator self-mining a first design | 0.055 / 0.017 (USD 43 / 13 M a year of miner revenue) | 0.73 / 0.22 (561 / 168 M) | the first entrant self-mines |
| The manufacturer selling hardware (keeps half the profit) | 0.11 / 0.034 | 1.50 / 0.45 | about 2x the operator's bar |
| The revision entrant (a second design at 0.3 x `C_dev`, `T0` 1 y) | 0.019 / 0.006 | 0.12 / 0.04 | a derivative costs a third and ships a year sooner; the sunk case bounds it |
| The shared-cost entrant (three share one design) | 0.040 / 0.012 | 0.24 / 0.07 | |
| The hybrid (self-mine a year, then sell) | 0.07 | 1.10 | between the two |
| Development SUNK (the mandatory case) | 0: the entrant pays manufacturing, deployment and operation only; its rows are section 2's chip rows | 0 | the whole threat is this case |
With development paid, every chip is ABOVE the best GPU entrant (415 to 588 micro-USD) at IGN 0.10 and below; the die
falls below it only from IGN 0.30 on a 3-year life; the board with paid development never does inside the window.
## 7. Several productive lifetimes (D4)
| Life | GDDR7 board, sunk, at 0.06 (micro-USD) | Against the 5070 Ti owner / entrant | SRAM die, sunk | Against the 5070 Ti owner / entrant | Reading |
|---|---|---|---|---|---|
| 0.5 y (the fixed-lane chip under the 180-day rotation) | 1,414 | 0.1x / 0.3x | 226 | 0.7x / 1.8x | neither chip pays at half a year; the board is 3x worse than a GPU entrant |
| 1 y | 750 | 0.2x / 0.6x | 134 | 1.2x / 3.1x | the board loses to every Blackwell entrant; the die is in band against owners |
| 2 y | 418 | 0.4x / 1.0x | 87 | 1.8x / 4.8x | |
| 3 y (the programmable chip's default) | 307 | 0.5x / 1.4x | 72 | 2.2x / 5.8x | the board's coexistence band; the die out of it |
| 5 y | 219 | 0.7x / 1.9x | 60 | 2.6x / 6.9x | |
The rotation (layer 3) sets the fixed-lane chip's life at 0.5 years and does nothing to a programmable chip; its
value is the factor between the first and the fourth row, not a wall, and the pass line's "scheduled ASIC death" is
not what holds either chip here: the board holds at every life from 1 year on its hardware term, the die holds at none.
## 8. Growing and shrinking networks, reduced issuance (D4)
Five years with a per-class supply curve (section 9's rule), the emission halving in years 3 and 5, three price paths,
a sunk chip fleet entering at the start of year 2. The chip's joules improve 1.5x at year 4 (a node step).
| Path, fleet | Year 2 | Year 3 | Year 5 | GPU classes above water (of 17) | Reading |
|---|---|---|---|---|---|
| Growing x2 from 0.10; a USD 1 M die fleet (1.3 TH/s) | 6 percent of 20 TH/s; margin 92 percent | 5 percent | 3 percent of 39 TH/s | 14, 15, 16 | coexistence by dilution: the supplier earns 90 percent margins on a tiny share; not a normal return, but no class leaves |
| Growing; USD 10 M die (12.8 TH/s) | 50 percent | 43 percent | 30 percent | 12, 13, 15 | half the chain on landing; dilutes to 30 percent by year 4 as GPUs re-enter at the higher price; margins 88 to 93 percent |
| Growing; USD 100 M die (128 TH/s) | 100 percent; margin 49 percent | 100 percent | 99 percent | 0, 0, 4 | failure: one buyer holds the chain for five years; four classes return in year 4 at IGN 0.80 |
| Growing; USD 10 M board (1.8 TH/s) | 9 percent; margin 65 percent | 7 percent | 4.5 percent | 14, 15, 16 | coexistence at a 57 to 69 percent margin |
| Flat 0.10; USD 1 M die | 9 percent | 12 percent | 19 percent of 6.9 TH/s | 12, 11, 6 | the halvings raise the share; six classes left by year 5 |
| Flat; USD 10 M die | 70 percent | 89 percent | 100 percent | 6, 6, 0 | failure by year 5 |
| Flat; USD 100 M die | 100 percent; margin -1 percent | -102 percent | -233 percent | 0 | failure for both: the fleet is larger than the revenue |
| Flat; USD 10 M board | 14 percent; margin 57 percent | 16 percent; 34 | 24 percent; 27 | 12, 11, 6 | coexistence at a normal return (27 to 57 percent) |
| Shrinking x0.5 from 0.30; USD 1 M die | 7 percent | 17 percent | 74 percent of 1.7 TH/s | 13, 11, 2 | failure in year 5 at USD 4 M of revenue |
| Shrinking; USD 10 M die | 58 percent | 96 percent | 100 percent; margin -77 percent | 11, 3, 0 | failure from year 3 |
| Shrinking; USD 10 M board | 10 percent; margin 62 percent | 22 percent; 28 | 91 percent; -30 percent | 13, 8, 2 | the board too takes a shrinking chain at USD 4 M of revenue, and loses money doing it |
Reading: reduced issuance (the halvings) and a shrinking price move every row toward the chip's share, and a USD 1 M
die fleet is 74 percent of a USD 4 M chain. The board coexists at a normal return in the growing and flat paths and
takes the chain only when the chain is worth less than its fleet; the die's USD 10 M fleet takes the chain on every
flat or shrinking path by year 3 to 5. The pass line's "small network": the board's pass holds at 42 TH/s and IGN
1.00 (the growing path's year 4), so it does not need one; the die's failure is worst in a small network, and a large
one only delays it.
## 9. Miners react with no fixed shares (D4: new)
The reaction rule replacing the first cut's single rule: each class participates with a fraction of its base, the
third of the base already owned joining as revenue per MH/s-hour rises from its owner cost to its entrant cost
(linearly) and the other two thirds entering when revenue exceeds the entrant cost; the installed base is the cap;
re-entry on a price rise is automatic (the growing path's year 4: 12.9 to 37.7 TH/s as 16 of 17 classes come back);
the chip fleet is fixed after entry. The equilibrium revenue per MH/s-hour and the GPU hash are solved each year by
bisection. Against the single-rule first cut the shares move: the USD 10 M die fleet holds 50 percent on landing in
the growing path (the first cut read 37) and 70 percent on the flat path (73), and the board holds 9 to 24 percent
(10 to 31): the per-class curve lets the cheaper classes stay longer and the dearer ones leave sooner, and the totals
are within 5 points of the first cut. The operator simulation carries the same reaction at a one-day step with
proving as a third choice; its D1 shock (a 1 TH/s die fleet) moves 1,500 of 9,177 mining cards to proving.
## 10. The operator simulation (D4 item 2, beside this model)
`docs/analysis/class-v6/operator-simulation.md`: operators choose per day among mine, internal prove, external prove
and off by profit at the marginal rate, under four shocks. At launch-shape demand every shock is restored in 0
periods (idle GPU capacity dwarfs the proving work). In a capacity-limited world (1 percent of the cards, the measured
5.5 percent proving efficiency) a lasting 1,000x proving spike is NOT restored within 150 periods when the internal
pool is a fixed sum (provers go to the external fee market and the internal backlog grows without bound) and is
restored in 35 periods with the resolution's congestion-priced internal proving fee; the price fall, the departure of
the six largest proving cohorts, the mining entrant and the proving entrant are restored in 0 periods in both worlds.
No parameter was changed by hand in any run. The one design finding: the fixed pool is a subsidy, not a price, and
internal proving needs the congestion-priced fee the resolution gives it.
## 11. Accessible supply, and the dependence on suppliers and operators (D4: its own table)
| IGN price | Network hash at the GPU equilibrium (per-class curve) | Share of the installed base available to mining (44.7 TH/s, approximate) | GPU suppliers | In N2 SRAM dies / wafers from ONE supplier | In DRAM boards | The largest single GPU operator today (a 1 percent fleet) | Label |
|---|---|---|---|---|---|---|---|
| 0.03 | 5.6 TH/s | 12 percent | NVIDIA (16 of 17 classes), AMD, Apple; tens of millions of cards in the world | 1,000 dies / 17 wafers | 34,000 | 0.06 TH/s | modelled |
| 0.10 | 9.6 | 21 | the same | 1,700 / 29 | 58,000 | 0.10 | modelled |
| 0.30 | 20 | 44 | the same | 3,600 / 60 | 119,000 | 0.20 | modelled |
| 1.00 | 42 | 95 | the same; past this the installed base binds and used prices rise | 7,700 / 128 | 255,000 | 0.42 | modelled |
The GPU side has three suppliers, a used market, a use outside mining (the rental yields of section 2) and no operator
above a percent of the hash; the die's whole chain is one wafer allocation (17 to 128 wafers of a node booked to 2028,
claimed) and the board's a Bitmain-class run of 34,000 to 255,000 units with a commodity memory bill. Dependence on a
single supplier: total for the die at every price, partial for the board (a run that size is visible and takes
months), nil for the GPU side; dependence on a single operator: a chip fleet is one operator by construction (the
operator simulation's D1), a GPU fleet of the same hash is tens of thousands of owners.
## 12. The conditions under which the success statement holds (the result)
| Condition | The number on today's rows | DRAM board | SRAM die |
|---|---|---|---|
| (a) the chip's all-in cost per accepted unit at its own electricity within about 1.5x of the best GPU owner's at the GPU's electricity | the owner at 0.12: 263 to 332 micro-USD (5070 Ti, 5080) | 307 at 3 y, 750 at 1 y: PASSES | 72 to 134: FAILS at every life |
| (b) the chip's annualised hardware per MH/s not below about a quarter of the GPU entrant's | the 5080 entrant's hardware 460 micro-USD | 239 to 677: PASSES | 33 to 95: FAILS |
| (c) a fleet above a third of the chain's hash costs more than a year's miner revenue | at IGN 0.10 the chain is 9.6 TH/s, a third 3.2 | USD 18 M of boards against USD 40 to 80 M: PASSES above IGN 0.05 | USD 2.5 M of dies: FAILS at every price under about 3 |
| (d) GPUs keep a resale market and a use outside mining | resale 25 to 55 percent after two years; rental 3x to 5x the mining cost | PASSES (the GPU side's property) | PASSES (the same) |
| (e) the per-joule gap at the honest knee stays under about 3x | Blackwell at the knee 1.70 to 2.06 microjoules | 2.2x to 2.6x: PASSES | 3.7x to 4.5x: FAILS (the bare-lane floor 10x to 12x) |
| (f) the supplier's gross margin on the share it holds is a normal return (under about 50 percent) and does not rise with each halving | | 21 to 41 percent in the five-year run: PASSES | 82 to 86 percent, or a loss once it is the chain: FAILS |
| (f) the supplier's gross margin is a normal return (under about 70 percent) and does not rise with the halvings | section 8 | 27 to 69 percent, falling with the halvings: PASSES | 85 to 93 percent, rising, or a loss once it is the chain: FAILS |
| (g) a second income (proving) for the commodity side that the specialised supplier cannot enter | section 5: USD 3 to 7 a card-day at launch demand on the 16 GB and larger tiers; 0 for any hash engine | PASSES (the board cannot prove either, which is the GPU's advantage over it) | PASSES (the same) |
So: **the success statement holds for the stored-dataset DRAM-board chip at a 1 to 3 year life on today's rows, in
every price path, at every electricity price on the axis, with the GPU side's own generation curve narrowing the gap
further; it does not hold for the N2 SRAM die once that die exists with its development sunk, at any life, price path
or electricity price, and the only condition that holds the die is that nobody pays to build it** (the surface of the
floor file's section 4.4: a USD 150 M project at a third of the chain needs IGN 0.73 over three years, a maker who
takes the chain 0.22, a revision 0.12 to 0.16). That last is a statement about an investor's decision, not about the
chain staying below a level, and the review is right that it is the only honest form.
**The result.** The success statement holds for the stored-dataset DRAM-board chip at a 1 to 3 year life on today's
rows, in every price path, at every electricity price on the axis, with the GPU side's own generation curve narrowing
the gap further and proving as a second income the chip cannot enter; its pass needs no small network (it holds at 42
TH/s and IGN 1.00), no token appreciation (it holds on the flat and shrinking paths) and no scheduled ASIC death (the
life axis, not the rotation, is what the board lives on). The statement does not hold for the N2 SRAM die once that
die exists with its development sunk, at any life, price path or electricity price; a small network makes it worse,
appreciation only delays it, and the rotation does not touch a programmable die. **The model names where it fails: a
sunk SRAM die fleet of USD 10 M or more at any price in the window, and of USD 1 M in a shrinking chain.** The only
condition that holds the die is that nobody pays to build it (section 6: IGN 0.73 for a USD 150 M project at a third
of the chain over three years, 0.22 taking the chain, 0.12 to 0.16 for a revision), which is a statement about an
investor's decision and is carried as such, not as a level the chain stays below.
What the chain can do about the die, from the model: raise the honest side's efficiency (every cent of GPU electricity
and every point of the knee moves condition (a) and (e); the lock already moves a Blackwell card 34 to 41 percent), keep
the share detector as the instrument that makes condition (c) visible the week it fails, and keep the dataset floor as
the ticket (section 3.2 of the floor file: USD 1,500 to 3,000 per die at the schedule, which moves condition (c) by 2x
to 4x and nothing else). Nothing in the hash moves conditions (a), (b) or (e) for the die by the factor they need.
What the chain controls, from the model: the honest side's cost per accepted unit (every cent of GPU electricity and
every point of the knee moves (a) and (e); the lock already moves a Blackwell card 34 to 41 percent), the second
income (proving demand and the congestion-priced fee that keeps internal proving served, the operator simulation's
finding), the share detector that makes (c) visible the week it fails, and the dataset floor as the ticket (USD 1,500
to 3,000 per die at the schedule, which moves (c) by 2x to 4x and nothing else). Nothing in the hash moves (a), (b) or
(e) for the die by the factor they need.
## 10. Unverified and owed
## 13. The D4 checklist, line by line
| Item | Where | Status |
|---|---|---|
| Growing and shrinking networks | section 8 | in |
| Reduced issuance | section 8 (the halvings in years 3 and 5) | in |
| Cheap and dear electricity | sections 1 to 3 | in |
| GPU replacement and resale on both sides, the 1.5x generation on the chip side too | section 4 | in |
| Changing proving demand | section 5 (zero, launch, spike; the resolution's shape; the 12 GB measured zero) | in |
| Private mining and hardware sales | section 6 | in |
| Several productive lifetimes (0.5, 1, 2, 3, 5) | section 7 | in |
| Cheaper derivative chips | section 6 (the revision and shared-cost rows) | in |
| Miners react with no fixed shares (per-class supply curve, installed base cap, re-entry) | section 9 | in |
| The outputs: cost advantage, replacement economics, accessible supply, break-even electricity per class (17), supplier and operator dependence as its own table | sections 1, 4, 11, 3, 11 | in |
| The tariff advantage shown separately from the hardware advantage, the 6.25x illustration first | section 1 | in |
| The operator simulation | section 10 and its own file | in (first run) |
| The k lane's placed rows and the adversary lane's whole-machine rows | the chip rows stand until they land | owed by others |
| Per-card agents, the hybrid mode, the measured stage times in the simulation | the simulation's section 4 | second cut |
## 14. Unverified and owed
- Every chip-side figure is modelled; no chip has been measured. The chip's hardware per MH/s (USD 0.8 for the die
with the system, 5.6 for the board) is the term conditions (b) and (c) rest on and is approximate within 2x.
- The card prices are street approximations of October 2026; the 5090's street price has been 2x MSRP this year, which
moves the entrant rows for that card by up to 2x and no other row.
- The Ada and Ampere knees are modelled (no rented host allows the lock); the Blackwell floors are measured on the 5090,
5080 and 4070, modelled on the 5070 Ti, 5070 and 5060 class.
- The accepted-work factor (97 percent), the wear allowance (5 percent a year), the hosting (USD 0.02 per kWh) and the
rental yields are approximate.
- The five-year run's GPU reaction is a single-rule model (entry at the cheapest entrant's cost, exit in cost order);
a second cut adds a per-class supply curve, the installed base as a cap on entry, re-entry on a price rise, and the
halving's effect on the entrant cost through used-card prices.
- The derivative-design rows (a revision at 0.3 x `C_dev`) are carried from the floor file's surface and not re-run
here; the sunk case bounds them.
- The emission beyond year 5 and the proving pool's 20 percent are outside the run.
- Nothing was run on the Mac; the script ran on build-3.
- The card prices are street approximations of October 2026; the installed-base counts available to mining are
approximate (the cap of 44.7 TH/s); the 5090's street price has been 2x MSRP this year.
- The Ada and Ampere knees are modelled (no rented host allows the lock); the Blackwell floors are measured on the
5090, 5080 and 4070.
- The proving rows rest on the bench table's shard times and the fleet lane's one measured day (with a fault); the
proving fleet (5,000 cards) and the pool sharing by capacity are assumptions.
- The accepted-work factor, the wear allowance, the hosting and the rental yields are approximate.
- The emission beyond year 5 and the proving pool's fade are outside the run.
- Nothing was run on the Mac; the scripts ran on build-3 (first cut) and build-4 (this cut).

114
docs/analysis/class-v6/operator-simulation.md Normal file
View file

@ -0,0 +1,114 @@
# The profit-maximising operator simulation: do the pricing and capacity rules restore service without an administrator?
8 October 2026, 16:2x to 16:5x UK, branch `class-v6-floor-sram`, floor lane 3, Igneum 2.0 D4 item 2 (the research lane's
word of 17:0x UK; the clock 20:15 UK). First run. **Every figure modelled**; the script is `scratchpad/opsim.py`, run on
build-4 (build-3 was down from 16:40 UK). The cost rows are the coexistence model's (lane 4's class v5 table, street
prices); the proving side carries the fleet lane's measured day on Devnet 3 (8 October 2026, 3,421 claims, 93 paid
segments): the 12 GB tier paid 0 of 313 claims (a measured zero for internal proving), steals 4.0 percent of claims,
paid to wasted card-seconds 14,800 to 254,000 (5.5 percent, on a day with a floor-link fault until 10:50 UTC). Nothing
here is served. The founder is not named.
## 0. The result in one page
| Shock | Launch-shape world (all available cards, assumed proving efficiency 0.5) | Stressed world (1 percent of the cards, the measured 5.5 percent proving efficiency, a 1,000x spike) without a congestion price on internal proving | The same with the resolution's congestion-priced internal proving fee | Reading |
|---|---|---|---|---|
| A. A proving demand spike (external x10, or x1,000 in the stressed world, lasting) | restored in 0 periods: capacity moves to the external market the same day (263,000 cards in, settling to 44,000), the fee never leaves 1x | **NOT restored in 150 periods**: 9,900 of 12,300 cards go to the external fee market, internal proving falls from 1,647 to 393 cards, the internal backlog grows without bound (8.9 M shards unproven by day 150) because the pool pays a fixed sum that cannot bid provers back | **restored after 35 periods**: the internal fee climbs to 2x, pulls 565 cards back, the backlog peaks at 368,000 shards (1.4 days of demand) and clears | the fee market restores EXTERNAL service by itself; INTERNAL proving needs the resolution's congestion-priced fee, or a spike starves it |
| B. The token price falls to 0.3x | restored in 0 periods: 130,000 cards leave the same day, hash 17.7 to 8.4 TH/s, revenue per MH/s-hour back to 309 to 315 micro-USD within 7 periods | restored in 0 periods | restored in 0 periods | exit at cost is immediate and proportionate; proving capacity stays 4x to 1,600x the demand |
| C. The six largest proving cohorts leave for good (H100, A100, 5090, 5080, 4090, 3090) | restored in 0 periods: hash 17.7 to 12.0 then 15.2 TH/s as the remaining classes re-enter at the higher rate; internal capacity stays 1,500x the demand | restored in 0 periods: capacity 2.8x to 3.1x the demand on the remaining classes | restored in 0 periods | re-entry at cost fills the gap; the 16 GB tier carries internal proving when the 24 GB and datacentre tiers leave |
| D1. A specialised entrant in mining (a sunk 1 TH/s SRAM die fleet at 0.06) | restored in 0 periods: 4,000 GPU cards leave, the chip holds 5.6 percent, proving untouched | restored in 0 periods: the chip holds 80 percent of a 1.25 TH/s network, GPU mining falls from 9,177 to 7,671 cards and 1,500 of them move to proving (internal 2,196 to 3,697) | restored in 0 periods | service holds in both worlds; coexistence does not (the coexistence model's finding), but the GPUs that leave mining go to proving, which the chip cannot do |
| D2. A specialised entrant in proving (a proving ASIC at 2,000 shard-equivalents a day at a tenth of the cost) | restored in 0 periods: the entrant takes the external market at the base fee; nothing else moves | restored in 0 periods | restored in 0 periods | external proving is a commodity market; an entrant lowers the fee and the GPUs leave that market for mining and internal proving |
**The pass statement, per shock.** The pricing and capacity rules (congestion-priced user-funded external proving
fees, the fee market, entry and exit at cost) restore service without an administrator in every shock of the
launch-shape world, in 0 periods, because idle GPU capacity dwarfs the proving work (16 M shard-equivalents a day of
external capacity against 200 of work; 426 M internal against 259,200). In a capacity-limited world the same rules
restore B, C, D1 and D2 in 0 periods and FAIL to restore A unless internal proving also carries a congestion-priced,
user-funded fee, with which A is restored in 35 periods. The one design finding: the 20 percent pool paid as a fixed
sum per block is a subsidy, not a price; when an external fee market outbids it, internal proofs starve, and the
resolution's shape (an explicit user-funded proving fee with congestion pricing for internal proving too) is what
closes it. No parameter was changed by hand in any run.
## 1. The model
| Element | Rule | Label |
|---|---|---|
| Cohorts | 17 card classes x 3 electricity prices (0.06, 0.12, 0.25 per kWh), a third of each class's available count per price; each cohort holds continuous shares of its cards in MINE, INTERNAL PROVE, EXTERNAL PROVE and OFF | modelled |
| Decision | each period (one day) a cohort moves a quarter of its cards toward the mode with the best profit per card-day, evaluated at the MARGINAL rate (the pool or fee divided by the capacity after its own move), with a 10 percent hysteresis; v1's all-or-nothing cohorts herded and oscillated and were replaced | modelled |
| Mining revenue | the miner emission (0.77 B IGN in year 1, 80 percent) at IGN 0.10 (assumption), shared by hash; the tip stays whole to the miner (the resolution's shape) | the spec's constant; the price an assumption |
| Internal proving | 3 shards per block (259,200 shard-equivalents a day); the 20 percent pool (0.53 M IGN a day) shared by proving capacity; a backlog accrues when capacity is short; the 12 GB tier earns nothing (measured zero); under the resolution's shape a user-funded congestion fee on top of the pool, rising 25 percent a period while the backlog exceeds a day of demand and falling 10 percent while under half a day, bounded 1x to 100x | measured zero (the fleet lane), modelled rule |
| External proving | USD 2,000 a day of demand at the base fee (spec 05's launch grid; USD 10 per shard-equivalent), 90 percent to the provers who deliver; the congestion fee rises 25 percent a period while job latency exceeds a day and falls 10 percent while under half a day, bounded 1x to 100x; demand elastic to the fee with exponent 0.5 | spec 05; the elasticity and the fee rule modelled |
| Proving capacity per card | the bench table's shard times (5090 6.3 s, 4090 6.3, 5070 4.8, 3080 7.1, 4070 12.1, 3060 14.4, 3090 14.9, H100 3.0, A100 5.0; the 9070 XT and the Mac cannot prove), times the proving efficiency (0.5 assumed after the fault fix; 0.055 the measured day), times 1 minus the 4 percent steal rate | measured shard times; the efficiency as stated |
| Costs | power at the card's floor joules (mining) or its proving watts, a wear allowance of 5 percent of the used price a year, 97 percent accepted work | the coexistence model's rows |
| Shocks | applied at period 0 after a 90-period warm-up; 150 periods observed | |
| Restored | the first period from which ALL of the following hold to the end of the run: external job latency under a day, internal backlog under a day of demand, internal capacity at least the demand, block production at least a fifth of the baseline hash; "NOT restored" otherwise | the test; v2's latching version was replaced |
| Worlds | launch-shape (all available cards) and stressed (1 percent of the cards, a 1,000x spike, the measured efficiency) | |
## 2. The runs
### 2.1 Launch-shape world (all available cards; proving efficiency 0.5; shock A at x10)
Baseline after warm-up: hash 17.7 TH/s on 408,829 mining cards; 109,221 cards on internal proving (426 M
shard-equivalents a day against 259,200 of demand); 3,011 on external proving (16 M against 200 of work); 712,937 off
(the cards whose power at their price exceeds the revenue); revenue 496 micro-USD per MH/s-hour; fees at 1x.
| Shock | t+0 | t+7 | t+30 | t+149 | Restored |
|---|---|---|---|---|---|
| A. external x10 | 263,000 cards move to external proving the same day (external capacity 16 M to 1,197 M) | 44,000 settle there; mining back to 17.8 TH/s | unchanged | unchanged | 0 periods |
| B. price x0.3 | 130,696 cards leave; revenue per MH/s-hour 149 micro-USD | hash 8.4 TH/s; revenue 313 | 8.4; 315 | 8.5; 309 | 0 periods |
| C. six cohorts leave | hash 12.0 TH/s; revenue 735 | 15.5 TH/s as 55,870 cards of the other classes enter; 568 | 15.2; 576 | 15.2; 576 | 0 periods |
| D1. a 1 TH/s SRAM fleet | hash 18.7 TH/s; 4,000 GPU cards leave | 17.9 (GPU 16.9) | 17.8 | 17.8 | 0 periods |
| D2. a proving ASIC | external capacity +2,000 shard-equivalents; nothing else moves | | | | 0 periods |
Reading: at launch-shape demand the proving service has 100x to 1,600x spare capacity in the GPUs idle at the price, so
no shock in the four can break it; the fee never leaves 1x. Mining re-prices itself within 7 periods of a 70 percent
price fall or a 32 percent capacity loss. Both proving efficiencies (0.5 and 0.055) give the same result here.
### 2.2 Stressed world (1 percent of the cards; the measured 5.5 percent proving efficiency; shock A at x1,000)
Baseline: hash 0.34 TH/s on 9,177 cards; internal capacity 1.1 M shard-equivalents a day (4.2x the demand) on 2,196
cards; external capacity 47,000 (240x the work) on 87 cards; 878 off; revenue 25,500 micro-USD per MH/s-hour.
| Shock | Without a congestion price on internal proving | With the resolution's congestion-priced internal fee |
|---|---|---|
| A. external x1,000 (USD 2 M a day) | t+0: external latency 3.2 days, the external fee 1.25x, 2,963 cards move; t+7: external capacity 4.8 M (latency 0), the fee back to 1x, but internal proving has fallen to 393 cards and 197,000 shard-equivalents a day (0.76x the demand), backlog 123,000; t+30: backlog 1.55 M; t+149: backlog 8.9 M. **NOT restored in 150 periods** | t+7: internal 378 cards, backlog 153,000, the internal fee 1x; t+30: the internal fee 2x, 505 cards back, backlog 286,000; by t+35 the backlog is under a day of demand and stays so; t+149: 565 cards on internal, backlog 0, the fee back to 1x. **Restored after 35 periods**; worst backlog 368,000 (1.4 days) |
| B. price x0.3 | restored in 0 periods; the mining fleet re-prices (revenue 25,500 to 7,650 micro-USD) and 913 cards move to external proving | restored in 0 periods |
| C. six cohorts leave | restored in 0 periods: internal capacity 802,000 (3.1x the demand) on the 16 GB tier | restored in 0 periods |
| D1. a 1 TH/s SRAM fleet | restored in 0 periods: the chip holds 80 percent of 1.25 TH/s; 1,500 GPU cards move from mining to proving (internal 2,196 to 3,697) | restored in 0 periods |
| D2. a proving ASIC | restored in 0 periods | restored in 0 periods |
The same world at the assumed 0.5 efficiency restores A in 0 periods in both variants (internal capacity 10 M against
259,200: the spike cannot pull enough capacity away); at 0.1 percent of the cards and 0.5 efficiency the internal
backlog again grows without bound under A without the internal fee (11.5 M by day 150). The failure needs two things
at once: a proving fleet near the demand (a small network or the measured efficiency) and a fee market that outbids
the fixed pool.
## 3. What the runs say about the rules
1. **The external fee market works as designed.** In every run the congestion fee brings capacity to the external
market within a day (launch-shape) or seven (stressed) and returns to 1x when the backlog clears; demand elasticity
keeps the fee bounded; a cheaper entrant (D2) takes the work at the base fee and the GPUs leave that market for the
other two, which is coexistence in the proving market.
2. **The fixed internal pool is the one rule that fails under stress.** It pays per block whatever the backlog, so it
cannot bid provers back from a hotter market; with the resolution's congestion-priced internal fee it can, in 35
periods at a 2x peak fee. The pass line for shock A reads "restored only with the internal congestion fee".
3. **Entry and exit at cost handle the price fall, the capacity loss and the mining entrant in 0 periods** in both
worlds; the GPUs that a mining entrant displaces move to proving, which the SRAM die cannot do, and that is the
second income the coexistence model's T3 table prices.
4. **The 12 GB tier's measured zero matters for C**: when the 24 GB and datacentre cohorts leave, internal proving
falls on the 16 GB tier (5080, 5070 Ti, 5060 Ti, 4080, 4060 Ti); the 12 GB cards (5070, 4070, 3060: 525,000 of the
1.1 M cards in the population) contribute nothing to it today.
## 4. Unverified and owed
- Every row is modelled; the price (IGN 0.10), the external demand (USD 2,000 a day), the elasticity (0.5), the fee
rule (25 percent up, 10 percent down, bounded 100x), the hysteresis and the quarter-per-period adjustment are
assumptions; the proving efficiency is the fleet lane's one measured day (with a fault) and an assumed post-fix
value.
- The available card counts are approximate; the stressed world is a scale factor, not a measured network.
- Block production is read as hash; propagation and the 30-second lock are outside the run. The internal pool is
shared by capacity (a sortition-like share), not by shards delivered, which flatters high-capacity classes.
- A second cut: per-card (not per-cohort) agents as in `sim/economy/sim.py`, the hybrid mode (mine and prove assigned
shards), the steal rate as a function of latency, the external bond and timeout (O-5.6), and a run on the measured
stage times (inputs 2.4 s, proving 26.8 s median and 365 s at the slowest 1 percent, aggregation 22.8 s, claim to
paid 336 s) instead of the bench table's shard times.
- Nothing was run on the Mac; the script ran on build-4.