Skip to content

ROADMAP.md

Latest commit

 

History

History
157 lines (115 loc) · 8.26 KB

ROADMAP.md

File metadata and controls

157 lines (115 loc) · 8.26 KB
Raw

Roadmap & research-backed feature plan

This document is the product plan for TrueCollider: what exists, what competitors do, and what we should build next — ordered by impact for real users (puzzles, vanity, multi-coin, beginners).

Landscape (what “serious” key search looks like in 2025–2026)

Approach Best for Notes
Sequential / random scan + bloom/fuse Address lists, vanity, puzzles with no pubkey Classic Keyhunt / VanitySearch path. TrueCollider already has this + endomorphism + AVX2/512.
BSGS Known pubkey, mid-size ranges Memory–time tradeoff. TrueCollider has CPU BSGS.
Pollard's Kangaroo / Lambda Known pubkey, narrow–medium ranges (Bitcoin puzzles) State of the art on GPU (RCKangaroo / PSCKangaroo / JeanLucPons lineage). Better memory scaling than full BSGS for large ranges.
GLV endomorphism secp256k1 address/rmd160/vanity ~√3–3× effective coverage. TrueCollider: CPU -e.
SIMD hash160 Bottleneck after EC SSE 4 / AVX2 8 / AVX-512 16. Done.
GPU EC + host filter Address hunt on NVIDIA TrueCollider CUDA path (batch still conservative).
GPU hash only (OpenCL) AMD/Intel/NVIDIA hash offload Done for compress address/rmd160.
ed25519 grind (Solana) SOL vanity / target lists CPU path done; GPU grinders elsewhere hit tens of M keys/s with comb + fe10 + base58 prefilter.
CreateAccountWithSeed (Solana) Fast “vanity” that is not a standalone keypair SHA256 only — different product; optional later mode.
Hybrid patterns Exploration spiral/chaos/gravity already in TrueCollider (ordering heuristics, not crypto breakthroughs).

Sources of practice in the wild: albertobsd/keyhunt, JeanLucPons Kangaroo/VanitySearch, RCKangaroo / PSCKangaroo forks, Solana CUDA vanity grinders (e.g. suffix GPU tools).

TrueCollider today (honest matrix)

Capability Status
Multi-coin address (+ SOL ed25519, troot) Done
Vanity / mnemonic / poetry / brainwallet / minikeys Done (CPU)
BSGS CPU Done
AVX2 + AVX-512 hash160 Done
CUDA secp EC (BTC-family + ETH/ETC + troot) Done (host encode/bloom; -M VRAM batch auto)
OpenCL hash160 Done (EC on CPU)
Kangaroo CPU Done (tiny sequential / larger DP)
Kangaroo CUDA Done (batch scan ≤2²⁴ + multi-walker DP; not RCKangaroo SOTA yet)
Memory budget -M Done (CUDA VRAM + BSGS blooms)
Dry-run / FOUND_*.txt / fixtures 66/125 Done
Online balance -N Done (wired on writekey hit path)
GPU BSGS / SOL GPU BSGS GRP + SOL ed25519 Done (tune next)
Beginner docs Done
One-click Windows installer Not yet

GPU for every currency + GPU BSGS (user goal)

Important architecture fact

Most TrueCollider coins share one curve:

Group Coins GPU work
A — secp256k1 BTC, LTC, DOGE, XRP, BCH, BTG, ETH, ETC, troot, all One GPU EC engine + coin-specific address encode (hash160 / keccak / x-only)
B — ed25519 SOL Separate GPU ed25519 scalar mult + base58
C — wordlist modes mnemonic / poetry / brainwallet / minikeys GPU helps little until after entropy→key; usually stay CPU or hybrid

So “all cryptocurrencies on GPU” does not mean 11 separate GPU apps — it means finish Group A completely, then add Group B, then hybrid C.

Phased build (recommended)

Phase Deliverable Why forums/tools do this
G0 Fix CUDA hash160 OR keep host hash but raise batch safely (streams, pinned mem, TDR-aware) Current batch=1 wastes GPU
G1 GPU address path for all Group A coins (ETH keccak + troot on device or host after GPU EC) VanitySearch / BitcoinAddressFinder / KeyHunt-Cuda pattern
G2 GPU vanity for Group A Highest user demand after puzzles
G3 GPU BSGS (giant-step on GPU, baby table in VRAM/RAM) iceland2k14 KeyHunt-Cuda, Etayson BSGS-cuda; Bitcointalk puzzles thread
G4 GPU Kangaroo (SOTA) RCKangaroo / PSCKangaroo — better than BSGS for large puzzle ranges
G5 GPU Solana ed25519 grind (+ mod-58^K prefilter) Modern SOL vanity CUDA tools
G6 OpenCL twin of CUDA EC/BSGS for AMD BitcoinAddressFinder multi-GPU OpenCL lesson

BSGS vs Kangaroo (Bitcointalk consensus)

  • BSGS: insane “keys/s” with huge RAM (-k); great for mid ranges / when you can afford tables.
  • Kangaroo: ~O(√n) ops, less RAM; wins for large puzzles (community: pure BSGS is hopeless at puzzle-135 scale vs kangaroo).
  • TrueCollider should ship both: GPU BSGS for “I have RAM”, GPU Kangaroo for “I have GPUs + pubkey + big range”.

AVX stack (status)

Level Hash160 Status
SSE 4-wide Yes Done
AVX2 8-wide Yes Done
AVX-512 16-wide Yes Done
AVX1-only Falls to SSE Correct (no integer YMM)
Future AVX-512 IFMA / Cyclone-style EC Optional CPU EC speedups (Keyhunt-Cyclone class)

P0 — users who “don’t have a clue”

  1. Getting Started + Commands (this folder) linked from README TL;DR.
  2. Example target files for every mode (tests/…) with comments.
  3. --dry-run / config dump: print resolved mode, coin, range, vector level, GPU backend before search.
  4. Windows “double-click” helpers: .bat launchers (run_puzzle66.bat, run_sol_sample.bat) that call keyhunt.exe with safe defaults.
  5. Clearer hit files: one format per coin (FOUND_BTC.txt, FOUND_SOL.txt) plus keep legacy KEYFOUNDKEYFOUND.txt.

P1 — speed & puzzles (secp256k1)

  1. Pollard's Kangaroo mode (-m kangaroo): pubkey + bit/range; CPU first, then CUDA/Vulkan.
  2. Safer CUDA batches: fix device hash160 or keep host hash but raise batch without TDR (stream sync, pinned memory).
  3. OpenCL secp256k1: parity with CUDA EC for AMD.
  4. BSGS UX: auto -k suggestion from free RAM; progress % for random/dance; documented bloom save/load.
  5. Checkpoint / resume for long random searches (save last key + RNG state).

P2 — Solana & vanity

  1. -m vanity -c sol: prefix/suffix on base58 ed25519 addresses (CPU), then CUDA ed25519 grind.
  2. Base58 mod-58^K prefilter (as in modern SOL GPU grinders) before full encode.
  3. Optional CreateAccountWithSeed grind mode (document that it is not a normal wallet keypair).

P3 — product polish

  1. HTML/Markdown report after run (keys tried, rate, mode, hardware).
  2. JSONL stats for scripts / dashboards.
  3. Unified binary CPU+CUDA+OpenCL with runtime -U.
  4. Signed releases + GitHub Release assets (keyhunt-windows-x64.zip, keyhunt-linux-x64.tar.gz).
  5. Large pages / huge bloom optional allocator for BSGS.

Explicit non-goals (for now)

  • Claiming “full 256-bit BTC search is practical.”
  • Shipping closed-source “AI finds keys” nonsense.
  • Mixing Solana into -c all (different curve).

Implementation sketch: Kangaroo (next big crypto feature)

  1. Input: compressed/uncompressed pubkey + -b / -r.
  2. CPU prototype: distinguished points, tame/wild herds, endomorphism-aware walks (constants verified vs bitcoin-core).
  3. Persist DP table; resume.
  4. Port hot loop to CUDA (learn from RCKangaroo design; re-verify endo constants).
  5. Document: “use Kangaroo when you have a pubkey and a range; use address when you only have an address.”

Doc / GitHub shape (vs Alberto)

Alberto’s Keyhunt README is a long mode tutorial. TrueCollider keeps:

Alberto style TrueCollider style
One giant README Hub README + docs/GETTING_STARTED + docs/COMMANDS + docs/ROADMAP
Linux-first / WSL Windows + Linux first-class
BTC/ETH focus Multi-coin + SOL + troot, with honesty about GPU
Puzzle TL;DR Same idea, plus “which mode do I pick?” table

Success metrics

  • New user finds a hit on tests/sol_sample.txt or a tiny BTC range without asking for help.
  • Puzzle runners know when to use address vs bsgs vs (future) kangaroo.
  • README never claims GPU modes that are CPU-only.