The right compute for the job.
What matters, what to ask providers and what to avoid, for the most common GPU workloads.
Training large models
Multi-GPU and multi-node runs that take days or weeks.
Read the guideServing models (inference)
Low-latency, always-on serving for real users.
Read the guideFine-tuning
Adapting an existing model with far less compute.
Read the guideImage and video generation
Diffusion and generative media workloads.
Read the guideScientific and HPC workloads
Simulation, modelling and research computing.
Read the guide