The Experiment
We set out to mine as many Bittensor subnets as possible on a single RTX 5090 rig (The Beast — Ryzen 9950X, 64GB RAM, 4TB NVMe). We registered on 16 subnets and deployed miners for each one.
The result? Only 3 out of 16 miners actually earned TAO.
The Earning Miners
| Subnet | Score | Earning Ratio |
|---|---|---|
| SN72 StreetVision | 85/100 | 232/256 miners earn (91%) |
| SN18 Zeus | 20/100 | 18/257 miners earn (7%) |
| SN13 Data Universe | 58/100 | 242/256 miners earn (95%) |
The Dead Subnets
The other 13 miners failed for very different reasons:
- SN44 ScoreVision — 1 out of 256 miners earning. The entire subnet is a monopoly. Score: 5/100.
- SN38 Sylliba — PyTorch does not support the RTX 5090 (sm_120). Crashed 4,944 times. Wasted 9.4GB VRAM.
- SN9 IOTA — Needs an orchestrator service nobody runs. Score: 5/100.
- SN25 Mainframe — Upstream database missing. Score: 5/100.
- SN107 Minos — Platform in demo mode. Zero rounds completed in 29 hours.
- SN50 Synth — 236/256 miners earn, but our predictions are not competitive enough.
- SN2 dSperse — DNS failures + stale circuit cache caused 940 restarts.
The Lesson
Most subnets look alive from the outside. They have miners registered, validators staking, emissions flowing. But when you actually deploy a miner, the reality is different:
- Monopolized subnets — one miner takes 100% of incentive (SN44, SN9, SN25)
- Hardware incompatibility — your GPU might not be supported (SN38)
- Missing infrastructure — the subnet needs services that do not exist (SN9, SN25, SN107)
- Competition too fierce — you run fine but your output is not good enough (SN50)
The Proof: We Did It Again
Even after building the mineability engine, we deployed 3 more miners on April 27 without checking our own tool first: SN65, SN68, and SN103. Twenty-four hours later — zero incentive on all three.
When we finally ran them through our own mineability scores:
| Subnet | Score | Verdict | Earning |
|---|---|---|---|
| SN103 Djinn | 58 | CAUTION | 243/256 (95%) — but top miner takes 82% |
| SN65 TPN | 15 | AVOID | 12/256 — only 5% of miners earn |
| SN68 Metanova | 8 | AVOID | 2/64 — only 3% of miners earn |
Our own tool would have told us not to deploy on SN65 and SN68. We burned TAO registering on subnets where almost nobody earns. That is exactly why we built this — so you do not have to learn the same lesson the expensive way.
So We Built Mineability Scores
We scored every subnet 0-100 based on what actually matters:
- Earning ratio — what percentage of miners have nonzero incentive?
- Concentration (Gini coefficient) — is incentive spread fairly or does one miner eat everything?
- Top-1 share — what percentage does the top earner take?
- Emission value — how much TAO flows to the subnet daily?
- Registration cost — how much TAO you burn to get a slot.
The verdict is simple:
- MINE (60+) — healthy earning distribution, deploy with confidence
- CAUTION (35-59) — mixed signals, research before deploying
- AVOID (0-34) — do not waste your TAO
Try It Now
The mineability page is free, live, and updates every 10 minutes:
Check Mineability Scores for All 128 Subnets →
For AI agents, the API is equally simple:
curl https://subnetaiq.io/api/v1/mineability/72 curl https://subnetaiq.io/api/v1/deploy/recommend?gpu_vram=32
Stop burning TAO on dead subnets. Check the score first.