ts3-sec-cuda-rs

TeamSpeak 3 Security Level Tool - GPU-accelerated Rust implementation for improving TS3 identity security levels.

Features

• 🔓 Decode identity files or strings
• ⚡ GPU acceleration via CUDA (NVIDIA GPUs)
• 🔧 Tunable kernel parameters for optimal performance
• 📊 Benchmarking suite included

Security levels are based on SHA-1 trailing zero bits, where level N proves ~2^N hashes were computed.

Quick Start

# Build
make build

# Increase security level to 30 (GPU)
cargo run --release -- increase --file identity.ini --target 30 --method cuda

# Decode identity
cargo run --release -- decode --file identity.ini

Installation

Requirements:

• Rust toolchain
• NVIDIA GPU + CUDA 13.0+ (optional, for GPU acceleration)

make build
# Or: cargo build --release

Usage

Increase Security Level (GPU - Recommended)

# From file (saves progress)
cargo run --release -- increase --file identity.ini --target 30 --method cuda

# From string (stdout only)
cargo run --release -- increase --string '14VHjz+...' --target 25 --method cuda

# Custom parameters
cargo run --release -- increase --file identity.ini --target 25 --method cuda \
  --cuda-threads 64 --batch-size 2000000

Decode Identity

cargo run --release -- decode --file identity.ini
# Or: cargo run --release -- decode --string '14VHjz+...'

Performance

GPU (CUDA)

Hardware: NVIDIA GeForce RTX 4080 (Ada Lovelace) Performance: 2.15 billion hashes/sec

Level Range	Time (GPU)	Time (CPU)
8 → 20	~1 second	~15 minutes
8 → 25	~5 seconds	~8 hours
8 → 30	~2 minutes	~11 days

Optimized defaults:

• Threads per block: 128
• Batch size: 4,000,000
• Shared memory: 16,384 bytes (auto)
• Stack usage: 80 bytes

CPU Fallback

Hardware: AMD Ryzen 9 5900X (Zen 3) Performance: 15 million hashes/sec

Options

--method <METHOD>              cpu (default) or cuda
--batch-size <SIZE>            CPU: 10k, CUDA: 4M (default)
--cuda-threads <THREADS>       32-1024 (default: 128)
--cuda-shared-mem <BYTES>      Auto-calculated by default

Run cargo run -- increase --help for all options.

Benchmarking

Find optimal parameters for your GPU:

make bench

Tests thread counts (32/64/128/256) and batch sizes (50k-16M) for both short and long messages.

Development

make build        # Build release binary
make test         # Run all tests (20 tests: CPU + CUDA)
make bench        # Run kernel parameter benchmarks
make nvcc-check   # Check CUDA stack usage (80 bytes)

Technical Details

CUDA Optimizations:

• Dual-path kernel (fast <56 bytes, slow for TS3 keys ~109-113 bytes)
• Shared memory L1 caching (128 bytes static)
• Ring buffer (w[16] instead of w[80])
• Stack optimization (80 bytes, down from 112)
• Dynamic shared memory (128 bytes/thread for multi-block)

Benchmark Results (RTX 4080):

Message Type	Threads	Batch Size	Throughput
Short (~16B)	32	16M	3.88 GH/s
Long (TS3)	128	4M	2.15 GH/s ✓

---

Identity Format: COUNTER || 'V' || base64(obfuscate(base64(KEYPAIR_ASN1))) Security Level: Trailing zero bits in SHA-1(public_key || counter) Key Type: P-256 ECC (secp256r1)

License

MIT License - see LICENSE file for details.