Device Management¶

The KrakenHashes agent provides comprehensive device detection, management, and optimization capabilities for password cracking workloads. This guide covers everything you need to know about configuring and optimizing hardware resources.

Overview¶

The agent uses hashcat's built-in device detection capabilities to identify and manage compute devices. This approach ensures compatibility with hashcat's device handling and provides accurate performance characteristics for each device.

Key Features¶

Automatic device detection using hashcat's -I flag
Multi-GPU support with intelligent device allocation
Cross-platform compatibility (Windows, Linux, macOS)
Backend optimization with priority-based device selection
Real-time monitoring during job execution
Alias filtering to prevent duplicate device entries

Supported Hardware¶

NVIDIA GPUs¶

Supported backends: - CUDA (Primary) - OpenCL (Fallback)

Requirements: - NVIDIA driver 450.80.02 or newer - CUDA toolkit 11.0 or newer (for CUDA backend) - OpenCL 1.2 or newer (for OpenCL backend)

Installation:

# Ubuntu/Debian
sudo apt-get install nvidia-driver nvidia-cuda-toolkit

# Verify installation
nvidia-smi
nvidia-settings --version

Optimal configuration: - Use CUDA backend when available (higher performance) - Enable GPU boost for maximum clock speeds - Ensure adequate power supply (750W+ for high-end cards) - Monitor temperatures (keep below 83°C for optimal performance)

AMD GPUs¶

Supported backends: - HIP (Primary for modern cards) - OpenCL (Universal)

Requirements: - AMD Radeon Software 22.7.1 or newer - ROCm 5.2 or newer (for HIP backend) - OpenCL 2.0 or newer

Installation:

# Ubuntu/Debian - ROCm installation
wget https://repo.radeon.com/amdgpu-install/5.4/ubuntu/jammy/amdgpu-install_5.4.50400-1_all.deb
sudo dpkg -i amdgpu-install_5.4.50400-1_all.deb
sudo amdgpu-install --usecase=rocm

# Verify installation
rocm-smi
clinfo | grep AMD

Optimal configuration: - Use HIP backend for RX 6000/7000 series and newer - Use OpenCL for older cards (RX 500/Vega series) - Enable GPU memory overclocking for hash-heavy algorithms - Monitor junction temperatures (keep below 110°C)

Intel GPUs¶

Supported backends: - OpenCL - Level Zero (Arc series)

Requirements: - Intel Graphics Driver 30.0.101.1404 or newer - Intel GPU tools for monitoring - OpenCL runtime

Installation:

# Ubuntu/Debian
sudo apt-get install intel-gpu-tools intel-opencl-icd

# Verify installation
intel_gpu_top
clinfo | grep Intel

Notes: - Intel Arc GPUs provide competitive hash rates for certain algorithms - Integrated graphics can be used for light workloads - Limited hashcat optimization compared to NVIDIA/AMD

CPU Processing¶

Supported: - All x86-64 processors - ARM processors (limited algorithm support)

Requirements: - Modern multi-core processor - Sufficient system memory (8GB+ recommended)

Optimal configuration: - Enable all CPU cores for maximum throughput - Ensure adequate cooling for sustained workloads - Consider CPU-only for specific algorithms (bcrypt, scrypt)

Device Detection¶

Automatic Detection¶

The agent automatically detects devices on startup using hashcat's device enumeration:

./agent -debug  # Enable debug output to see detection process

Detection process: 1. Locates latest hashcat binary in data directory 2. Executes hashcat -I command 3. Parses device information and capabilities 4. Filters aliases and invalid devices 5. Stores device configuration for job allocation

Device Properties¶

Each detected device includes:

{
  "device_id": 1,
  "device_name": "NVIDIA GeForce RTX 4090",
  "device_type": "GPU",
  "enabled": true,
  "processors": 128,
  "clock": 2520,
  "memory_total": 24576,
  "memory_free": 23552,
  "pci_address": "01:00.0",
  "backend": "CUDA",
  "is_alias": false
}

Backend Priority¶

When multiple backends are available for the same device, the agent uses this priority order:

HIP (AMD optimized)
CUDA (NVIDIA optimized)
OpenCL (Universal fallback)

This ensures optimal performance by selecting the most efficient backend for each device.

Multi-GPU Configuration¶

Automatic Load Balancing¶

The agent automatically distributes workload across available GPUs:

Equal distribution for identical GPU models
Proportional allocation based on compute capability
Dynamic adjustment based on real-time performance

Manual Device Selection¶

You can manually enable/disable specific devices:

# Through the web interface:
# Agent Details → Device Management → Toggle device status

# Or via API:
curl -X PUT http://localhost:8080/api/agents/{agent_id}/devices/{device_id} \
  -H "Content-Type: application/json" \
  -d '{"enabled": false}'

Optimal Multi-GPU Setups¶

Recommended configurations:

Identical GPUs: Best performance and load balancing

4x RTX 4090 → ~400 GH/s MD5
8x RTX 3080 → ~640 GH/s MD5

Mixed GPUs: Group similar performance tiers

2x RTX 4090 + 2x RTX 4080 → Separate job allocation

CPU + GPU hybrid: Use CPU for specific algorithms

GPU: Fast hashes (MD5, SHA1, NTLM)
CPU: Slow hashes (bcrypt, scrypt, Argon2)

Performance Optimization¶

GPU Optimization¶

Memory optimization:

# Enable optimized kernels (uses more VRAM but faster)
-O

# Set workload tuning
-w 1  # Low power usage
-w 2  # Default
-w 3  # High performance (recommended)
-w 4  # Insane performance (may cause instability)

Hashcat performance flags:

# Example optimization for RTX 4090
--gpu-loops 1024 --gpu-accel 128 --gpu-threads 1024

Thermal Management¶

Temperature monitoring: - NVIDIA: Use nvidia-smi or built-in monitoring - AMD: Use rocm-smi or amdgpu-pro - Intel: Use intel_gpu_top

Thermal limits: - NVIDIA: 83°C (optimal), 91°C (maximum) - AMD: 110°C (junction), 95°C (edge) - Intel: 100°C (throttling)

Cooling recommendations: - Ensure adequate case ventilation - Monitor ambient temperature (keep below 25°C) - Consider undervolting for 24/7 operations - Use custom fan curves for sustained workloads

Power Management¶

Power considerations:

# Check GPU power limits
nvidia-smi -q -d POWER     # NVIDIA
rocm-smi --showpower       # AMD

Power optimization: - Ensure adequate PSU capacity (add 20% headroom) - Enable power limit increases where possible - Monitor power consumption during long jobs - Consider efficiency curves for different algorithms

Device Allocation Strategies¶

Job-Based Allocation¶

Strategy selection: 1. Round-robin: Distribute tasks evenly across devices 2. Performance-based: Allocate based on device capability 3. Memory-based: Consider VRAM requirements 4. Thermal-aware: Avoid overheated devices

Workload Distribution¶

Hash type considerations: - Fast hashes (MD5, SHA1): Use all available GPUs - Medium hashes (SHA256, SHA512): Balance GPU count vs. memory - Slow hashes (bcrypt, scrypt): May benefit from CPU processing - Memory-hard (Argon2): Requires high VRAM devices

Example allocations:

# Large wordlist + fast hash = all devices
hashcat -m 0 -a 0 hashes.txt wordlist.txt -d 1,2,3,4

# Complex rules + medium hash = subset of devices
hashcat -m 1000 -a 0 hashes.txt wordlist.txt -r rules.txt -d 1,2

# Brute force + slow hash = single high-end device
hashcat -m 3200 -a 3 hashes.txt ?a?a?a?a?a?a -d 1

Benchmarking and Capabilities¶

Hashcat Benchmarking¶

Run benchmarks:

# Full benchmark suite
hashcat -b

# Specific algorithm benchmark
hashcat -b -m 1000  # NTLM benchmark

# Device-specific benchmark
hashcat -b -d 1     # Benchmark only device 1

Interpreting results:

Speed.#1.........:   123.4 GH/s (95.2ms) @ Accel:512 Loops:1024 Thr:64 Vec:1

- 123.4 GH/s: Hash rate (Giga-hashes per second) - 95.2ms: Kernel execution time - Accel:512: Acceleration factor - Loops:1024: Iteration loops - Thr:64: Thread count - Vec:1: Vector width

Performance Baselines¶

Expected performance (RTX 4090): - MD5: ~100 GH/s - SHA1: ~35 GH/s - NTLM: ~180 GH/s - SHA256: ~15 GH/s - bcrypt: ~150 KH/s

Expected performance (RX 7900 XTX): - MD5: ~75 GH/s - SHA1: ~25 GH/s - NTLM: ~130 GH/s - SHA256: ~12 GH/s - bcrypt: ~120 KH/s

Hardware Requirements¶

Minimum Requirements¶

System specifications: - CPU: 4 cores, 2.5GHz - RAM: 8GB - Storage: 100GB available space - GPU: Any OpenCL 1.2 compatible device - Network: 100 Mbps for file synchronization

Recommended Requirements¶

High-performance setup: - CPU: 16+ cores, 3.0GHz+ - RAM: 32GB+ (64GB for large hashlists) - Storage: 1TB+ NVMe SSD - GPU: RTX 4080/4090 or RX 7900 XT/XTX - Network: 1 Gbps for large file transfers - PSU: 1000W+ 80+ Gold rated

Enterprise Requirements¶

Large-scale deployment: - CPU: 32+ cores server processor - RAM: 128GB+ ECC memory - Storage: 10TB+ enterprise SSD array - GPU: Multiple high-end cards with NVLink/Infinity Cache - Network: 10 Gbps with redundancy - Power: Redundant 1600W+ PSUs - Cooling: Dedicated server room cooling

Driver Requirements¶

NVIDIA Drivers¶

Recommended versions: - Production: Latest stable driver (535.x+) - Development: Latest beta driver for new features - Enterprise: Long-term support versions (470.x LTS)

Installation verification:

nvidia-smi
nvcc --version  # CUDA compiler
nvidia-settings --version

AMD Drivers¶

Recommended versions: - ROCm: 5.4+ for HIP support - AMDGPU-PRO: 23.20+ for OpenCL - Mesa: 23.0+ for open-source stack

Installation verification:

rocm-smi
rocminfo | grep "Agent"
clinfo | grep AMD

Intel Drivers¶

Recommended versions: - Graphics Driver: 30.0.101.1404+ - OpenCL Runtime: 22.43+ - Level Zero: 1.8+ (Arc series)

Installation verification:

intel_gpu_top
clinfo | grep Intel
vainfo | grep "Driver version"

Troubleshooting Device Issues¶

Common Issues¶

Device not detected: 1. Verify driver installation 2. Check hardware compatibility 3. Ensure proper PCI Express connection 4. Verify power supply adequacy 5. Test with hashcat directly: hashcat -I

Poor performance: 1. Check thermal throttling 2. Verify power limits 3. Update drivers 4. Check for conflicting processes 5. Validate hashcat parameters

System instability: 1. Reduce workload tuning (-w 2 instead of -w 3) 2. Lower GPU clocks and memory speeds 3. Improve cooling and power delivery 4. Check for hardware defects 5. Verify system memory integrity

Debug Commands¶

Hardware diagnostics:

# System information
lspci | grep -i vga
lshw -c display

# GPU status
nvidia-smi -l 1        # NVIDIA monitoring
rocm-smi -l            # AMD monitoring
intel_gpu_top          # Intel monitoring

# Temperature monitoring
sensors                # System sensors
nvidia-smi dmon        # NVIDIA detailed monitoring

Hashcat diagnostics:

# Device information
hashcat -I

# Test device functionality
hashcat -t

# Benchmark specific device
hashcat -b -d 1

# Debug mode
hashcat --debug-mode=1 -m 1000 hash.txt wordlist.txt

Performance Troubleshooting¶

If hash rates are lower than expected:

Check thermal throttling:

nvidia-smi dmon -s pucvmet -c 60  # Monitor for 60 seconds

Verify power limits:
```
nvidia-smi -q -d POWER
```

Test with different parameters:

# Conservative settings
hashcat -w 2 -O -m 1000 hash.txt wordlist.txt

# Aggressive settings (may cause instability)
hashcat -w 4 -O -m 1000 hash.txt wordlist.txt

Check for competing processes:

ps aux | grep -E "(hashcat|john|nvidia|rocm)"
nvidia-smi pmon  # Process monitoring

Best Practices¶

Hardware Selection¶

Match workload to hardware:
Fast hashes: High core count GPUs (RTX 4090, RX 7900 XTX)
Slow hashes: High memory bandwidth (RTX 3090, RX 6900 XT)
Mixed workloads: Balanced systems with CPU + GPU
Consider total cost of ownership:
Power consumption over lifetime
Cooling requirements and costs
Maintenance and replacement cycles
Performance per dollar ratios

Configuration Management¶

Document hardware configurations:
GPU models, VRAM, clock speeds
Driver versions and update schedules
Optimal hashcat parameters for each device
Thermal and power limit settings
Monitor performance trends:
Track hash rates over time
Monitor for degradation or throttling
Log hardware errors and failures
Schedule preventive maintenance
Implement redundancy:
Deploy multiple agents for high availability
Use mixed hardware to avoid single points of failure
Maintain spare hardware for critical operations
Implement proper backup and recovery procedures

Security Considerations¶

Physical security:
Secure hardware from unauthorized access
Monitor for tampering or theft
Implement proper access controls
Use hardware-based attestation where possible
Driver security:
Keep drivers updated for security patches
Verify driver signatures and authenticity
Monitor for driver-level exploits
Use enterprise driver branches when available
Performance isolation:
Isolate cracking workloads from other processes
Use dedicated hardware for sensitive operations
Monitor for unauthorized resource usage
Implement resource quotas and limits

By following this comprehensive guide, you'll be able to effectively configure, optimize, and manage hardware resources for maximum password cracking performance while maintaining system stability and security.