GPU Calculator - clockwork

Number of GPUs

AI job types

Training

Inferencing

Fine-tuning

Model type

LLMs

Bidirectional Contextual Models

Diffusion-based Generative Models

Vision Transformers (ViT)

Mixture of Experts (MoE)

Multimodal Models

Other

Job time completion in weeks

Number of concurrent AI jobs per cluster

GPU owned or rented

AI job crash frequency per week

Wasted time (crash, restart / recover from previous checkpoint) in hours

FTE cost per incident in dollars

Checkpoint interval

Average GPU cost per hour

GPU owned models

NVIDIA A100

NVIDIA H100

NVIDIA H200

NVIDIA Blackwell

AMD MX300 Series

Google TPU

Amazon Trainium

Other

GPU service providers

Estimate Impact of your GPU Downtime