Best Cloud GPUs by Use Case

GPU requirements are set by what you are running, not by which card tops a spec sheet. Each page below covers what a workload demands of a GPU, which models suit it, and what providers currently charge for them.

Best GPUs for LLM Training

Multi-GPU nodes with HBM and high-bandwidth interconnects

7 GPUs80 GB+ VRAMfrom $0.580/hr

Best GPUs for AI Inference

Throughput per dollar, with enough VRAM to hold the weights

8 GPUs24 GB+ VRAMfrom $0.160/hr

Best GPUs for Fine-Tuning

LoRA and QLoRA on one GPU, full fine-tunes on a node

7 GPUs24 GB+ VRAMfrom $0.160/hr

Best GPUs for AI Image Generation

Diffusion models fit in 24 GB; batch size buys throughput

7 GPUs16 GB+ VRAMfrom $0.090/hr

Best GPUs for Scientific Computing

FP64 throughput, ECC memory, and node-level interconnect

7 GPUs32 GB+ VRAMfrom $0.085/hr

Best GPUs for Video Rendering and Transcoding

Hardware encoders, VRAM for scene data, and stream density

7 GPUs16 GB+ VRAMfrom $0.160/hr

Best GPUs for RAG Pipelines

An embedding model and a generator, sized separately

7 GPUs16 GB+ VRAMfrom $0.086/hr

Best Budget GPUs for AI

The lowest cost per unit of work, not the lowest hourly rate

7 GPUs16 GB+ VRAMfrom $0.086/hr