Lab Guide — GPU Acceleration & Performance

Detect compute devices with Hashcat and system tools

Verify that your GPU is properly detected before running any attacks.

1. Run hashcat -I and note all detected devices, their IDs, types, and memory
2. For NVIDIA: run nvidia-smi and record the GPU name, driver version, CUDA version, and current temperature
3. For AMD or general OpenCL: run clinfo and note the platform and device details
4. Run hashcat -b -m 0 to benchmark MD5 on your system and record the speed

# Step 1: Hashcat device info
hashcat -I

# Step 2: NVIDIA system info
nvidia-smi

# Step 3: OpenCL info (all vendors)
clinfo

# Step 4: Quick MD5 benchmark
hashcat -b -m 0

Checkpoint: You can identify your GPU by name and device ID. The benchmark completes without errors.

Run a dictionary attack using the GPU

Crack the test hashes using rockyou.txt on the GPU.

1. Run the attack with GPU device type:

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

2. Observe the output: note the speed (H/s), progress, and time to completion
3. Check the results file:

# View cracked passwords
cat cracked_gpu.txt

Checkpoint: All test hashes are cracked. The speed shows millions or billions of H/s (depending on your GPU).

Run a brute-force mask attack on the GPU

Use a mask to brute-force short passwords on the GPU.

1. Create a hash of a short known password (e.g., "test1"):

# Create a test hash for mask attack
echo -n "test1" | md5sum | cut -d' ' -f1 > mask_test.txt

2. Run a mask attack targeting 5-character lowercase+digit passwords:

# Mask attack: 4 lowercase letters + 1 digit
hashcat -m 0 -a 3 -D 2 -o cracked_mask.txt mask_test.txt ?l?l?l?l?d

3. Note the speed, keyspace size, and estimated time to exhaustion
4. Verify the result:

cat cracked_mask.txt

Checkpoint: The mask attack finds "test1". You can explain the keyspace size (26^4 * 10 = 4,569,760 candidates).

Compare speeds across all four workload profiles

Test each workload profile and record the hash rate.

1. Run the benchmark for each profile and record the speed:

# Test workload profile 1 (Low)
hashcat -b -m 0 -D 2 -w 1

# Test workload profile 2 (Medium)
hashcat -b -m 0 -D 2 -w 2

# Test workload profile 3 (High)
hashcat -b -m 0 -D 2 -w 3

# Test workload profile 4 (Maximum)
hashcat -b -m 0 -D 2 -w 4

2. Record the results in a comparison table:

Profile | Speed (H/s)    | System Responsiveness
--------|----------------|----------------------
-w 1    | ______________ | ______________________
-w 2    | ______________ | ______________________
-w 3    | ______________ | ______________________
-w 4    | ______________ | ______________________

3. Note the speed increase percentage between profiles

Checkpoint: You have a completed comparison table. Speed increases with each profile level.

Set thermal safeguards and observe throttling behaviour

Configure temperature limits and monitor GPU thermals during an attack.

1. Open a second terminal and start monitoring GPU temperature:

# NVIDIA: refresh every 1 second
nvidia-smi -l 1

# Linux alternative (AMD/NVIDIA)
watch -n1 sensors

2. In the first terminal, run an attack with thermal limits:

# Attack with temperature controls
hashcat -m 0 -a 3 -D 2 -w 3 \
  --gpu-temp-abort=85 \
  --gpu-temp-retain=75 \
  hashes.txt ?a?a?a?a?a?a

3. Observe the Hashcat status output — look for temperature readings and any throttling messages
4. Press s during the attack to see the current status including temperature
5. Press q to quit when you have observed enough data

Checkpoint: Temperature stays at or below the retain value. You can see Hashcat adjusting speed to manage thermals.

Compare CPU and GPU cracking speeds

Run the same attack on CPU and GPU to quantify the performance difference.

1. Benchmark MD5 on CPU only:

# CPU benchmark
hashcat -b -m 0 -D 1

2. Benchmark MD5 on GPU only:

# GPU benchmark
hashcat -b -m 0 -D 2

3. Run the same dictionary attack on both and compare wall-clock time:

# CPU dictionary attack (time it)
time hashcat -m 0 -a 0 -D 1 -o cracked_cpu.txt hashes.txt rockyou.txt

# GPU dictionary attack (time it)
time hashcat -m 0 -a 0 -D 2 -o cracked_gpu2.txt hashes.txt rockyou.txt

4. Calculate the speedup factor: GPU speed / CPU speed
5. Repeat the comparison for a slower hash type like SHA-256 (-m 1400) or bcrypt (-m 3200)

# Benchmark SHA-256 on CPU vs GPU
hashcat -b -m 1400 -D 1
hashcat -b -m 1400 -D 2

Checkpoint: You can state the GPU speedup factor for at least two hash types. GPU is significantly faster for both.