Rent database-priced GPUs for LLM inference, fine-tuning, multimodal workloads, computer vision and research with clear specifications and model-aware sizing.
Browse hardware the calculator understands before running a workload-specific recommendation. Catalogue entries are separate from live capacity and approved pricing.
Best for high-memory LLM inference, quantized 70B+ workloads and batch serving with generous VRAM headroom.
Best for production LLM serving, 70B-class quantized inference and high-bandwidth multi-GPU training.
Best for powerful single-node inference, smaller fine-tuning jobs and cost-efficient Hopper access.
Best for proven CUDA workloads, LoRA fine-tuning, research and stable production inference.
Best for image generation, computer vision, embeddings and mid-size LLM inference.
Best for small-model inference, embeddings, reranking and lightweight production workloads.
Use the catalogue for product selection, then run the recommender when model size, context length or budget matters.
Best for high-memory LLM inference, quantized 70B+ workloads and batch serving with generous VRAM headroom.
The calculator estimates model weights, KV cache where applicable, runtime workspace, workload overhead and production headroom before filtering inventory.
Inference and training are sized separately. Training estimates include optimiser state, gradients or adapter state, activations, workspace and topology constraints.
Lower precision can reduce model weight memory, but quantisation metadata, runtime workspace and technical limits still apply.
Quote requests are locked to the model snapshot, calculator version, selected configuration and pricing version so later catalogue changes do not rewrite the quote.
The estimator combines model weights, KV cache, runtime workspace, batching, workload overhead and safety headroom.
Yes. Fine-tuning and training include gradients, optimiser state, activations and topology constraints.
Yes. You can enter parameter count and architecture details; lower-confidence estimates are labelled.
No. Estimated results and measured benchmark data are labelled separately.
No. Unavailable hardware can remain in the catalogue, but it is not offered as immediately purchasable inventory.
Approved prices are shown where cost and market inputs have been reviewed. Otherwise the offer remains request-quote.
