Lesson 08 — GPU Acceleration & Performance

Lesson 08 · Slide 1

GPU Acceleration & Performance

Modern GPUs transform password cracking from hours to minutes. This lesson covers how to harness GPU parallel processing in Hashcat, tune performance, manage thermals, and troubleshoot common issues.

Learning Objectives

Understand why GPUs outperform CPUs for hash cracking
Prepare your system with the correct drivers and compute frameworks
Use device selection and workload profiles to control performance
Configure temperature limits and runtime constraints
Troubleshoot common GPU issues in Hashcat

Lesson 08 · Slide 2

Why GPU?

CPUs are designed for complex sequential tasks with a few powerful cores. GPUs contain thousands of smaller cores optimized for parallel computation — perfect for hashing.

GPU Advantages

Parallel computation: Thousands of cores process hashes simultaneously
High throughput: 10–100x faster than CPU for most hash types
Efficient for large volumes: Millions of hash candidates tested per second

Example: An NVIDIA RTX 4090 can compute ~160 billion MD5 hashes per second. A high-end CPU manages ~5 billion. That is a 32x difference on a single GPU.

Lesson 08 · Slide 3

Preparation

Before using GPU acceleration, your system must have the correct drivers and compute frameworks installed.

NVIDIA GPUs

Install the latest NVIDIA driver from nvidia.com
Install CUDA Toolkit (includes OpenCL support)
Verify with nvidia-smi

AMD GPUs

Install AMD Adrenalin drivers or ROCm
Ensure OpenCL runtime is available
Verify with clinfo

Note: Hashcat supports OpenCL, CUDA, and HIP backends. CUDA is preferred for NVIDIA; OpenCL works for both vendors.

Lesson 08 · Slide 4

Checking Available Devices

Use hashcat -I to list all available compute devices on your system.

# List all OpenCL/CUDA devices
hashcat -I

Sample output:

OpenCL Info:
============
Platform ID #1
  Name.....: NVIDIA CUDA
  Vendor...: NVIDIA Corporation
  Device ID #1
    Name...........: NVIDIA GeForce RTX 4090
    Type...........: GPU
    Memory.........: 24564 MB
    OpenCL Version.: OpenCL 3.0 CUDA

Platform ID #2
  Name.....: Intel(R) OpenCL
  Device ID #2
    Name...........: Intel(R) Core(TM) i9-13900K
    Type...........: CPU

Device IDs matter. Use these IDs with -d to target specific devices.

Lesson 08 · Slide 5

Using GPU in Hashcat

The -D flag selects the device type. Use -D 2 to target GPUs exclusively.

# Dictionary attack on GPU
hashcat -m 0 -a 0 -D 2 -o results.txt hashes.txt rockyou.txt

Flag	Value	Description
`-D 1`	CPU	Use CPU only
`-D 2`	GPU	Use GPU only
`-D 1,2`	CPU + GPU	Use both (not always optimal)

Selecting Specific GPUs

# Use only GPU devices 1 and 2
hashcat -m 0 -a 0 -D 2 -d 1,2 hashes.txt rockyou.txt

Lesson 08 · Slide 6

Workload Profiles

The -w flag controls how aggressively Hashcat uses the GPU. Higher profiles yield faster speeds but can make the system unresponsive.

Profile	Name	Description	Use Case
`-w 1`	Low	Minimal GPU usage	Desktop remains usable
`-w 2`	Medium	Default balance	Background cracking
`-w 3`	High	Aggressive GPU usage	Dedicated cracking rig
`-w 4`	Maximum	Full GPU saturation	Headless server, no display

# High workload profile
hashcat -m 0 -a 3 -D 2 -w 3 hashes.txt ?a?a?a?a?a?a

Warning: Profile -w 4 can cause display freezes and system instability on machines with a single GPU that also drives the display.

Lesson 08 · Slide 7

Temperature Control

Hashcat provides built-in thermal management to protect your hardware.

# Abort if GPU reaches 85C, throttle to stay under 75C
hashcat -m 0 -a 0 -D 2 \
  --gpu-temp-abort=85 \
  --gpu-temp-retain=75 \
  hashes.txt rockyou.txt

Temperature Flags

--gpu-temp-abort=N — Immediately stop if any GPU reaches N degrees Celsius
--gpu-temp-retain=N — Dynamically throttle speed to keep the GPU at or below N degrees

Note: Monitor temperatures in real-time with nvidia-smi -l 1 (NVIDIA) or watch -n1 sensors (Linux).

Lesson 08 · Slide 8

Performance Tuning

Advanced flags for fine-tuning GPU performance:

Flag	Description	Notes
`-n`	Kernel accel (threads per block)	NVIDIA tuning; higher = more parallelism
`-u`	Kernel loops (work blocks)	AMD tuning; controls work granularity
`-T`	Kernel threads (vector width)	Number of vector elements per thread

Runtime Limits

# Run for at most 1 hour, show status every 60 seconds
hashcat -m 0 -a 3 -D 2 \
  --runtime=3600 \
  --status-timer=60 \
  hashes.txt ?a?a?a?a?a?a

Tip: Let Hashcat auto-tune first (-w 3 without -n/-u/-T). Only manually tune if you understand your GPU architecture and need to squeeze out the last few percent.

Lesson 08 · Slide 9

Troubleshooting

Common GPU issues and solutions:

Low Hash Speed

Check workload profile (-w 3 or -w 4)
Ensure GPU drivers are up to date
Verify you are using -D 2 (not CPU)

GPU Overheating

Set --gpu-temp-retain=75
Improve case airflow / cooling
Reduce workload profile

Multi-GPU Issues

Use -d to select specific devices
Check for PCIe bandwidth bottlenecks
Ensure matching driver versions

Driver / RAM Problems

Reinstall CUDA or OpenCL runtime
Check GPU memory with nvidia-smi
Close other GPU-intensive applications

Lesson 08 · Slide 10

Lab & Homework

Lab Exercises

Device check — run hashcat -I and system tools
GPU dictionary attack — crack hashes with -D 2
GPU mask attack — brute-force with GPU
Workload comparison — test -w 1 through -w 4
Temperature limits — set thermal safeguards
CPU vs GPU — compare performance

Homework Tasks

Benchmark all hash modes on your GPU
Optimize workload for a specific hash type
Document multi-GPU or CPU-vs-GPU results