GPU AI Calculator

Calculate VRAM requirements and costs for LLM training and inference

Model Configuration

Model Size (Billion Parameters)

Quantization (Precision)

Lower precision reduces memory but may affect accuracy

Context Window (Sequence Length)

Common values: 4096, 8192, 32768, 131072 (128k)

Select GPU

VRAM: 80 GB

TFLOPS (FP16): 1979

Bandwidth: 3350 GB/s

Price: $4.00/hr

Number of GPUs

Total VRAM: 80.00 GB

Required VRAM

17.88 GB

Model: 8B params × 2 bytes × 1.2 (overhead)

Fit Status

Fits comfortably

Available: 80.00 GB

VRAM Usage22.4%

Used: 17.88 GBTotal: 80.00 GB

Hourly Cost

$4.00

1 × NVIDIA H100 SXM @ $4.00/hr

Monthly Cost (24/7)

$2,920.00

730 hours/month

Note: Prices are estimated hourly rates. Actual costs may vary by cloud provider and region. Consider spot instances for training workloads.