Design

Architecture Center

Reference architectures for common GPU platforms, plus an interactive designer that turns a GPU choice and topology into a bill of materials, capacity and power estimates, and a generated diagram.

Architecture Designer

Pick a GPU and topology → get a live diagram, BOM, and capacity, power & cost estimates.

Heterogeneous cluster designer

Compose a cluster from multiple GPU pools — mix types or vendors (NVIDIA + AMD).

Reference architectures
Bare metalSingle cluster

LLM Training Cluster

Rail-optimized bare-metal fabric for large distributed training.

LLM trainingHPC
Open reference
Public cloudMulti-cluster

Real-Time Inference Platform

Low-latency, autoscaling model serving behind a gateway.

LLM inferenceMultimodal
Open reference
HybridSingle cluster

RAG Platform

Retrieval-augmented generation with a vector store and reranking.

RAGLLM inference
Open reference
Air-gappedSingle cluster

Secure Regulated-Industry AI Platform

Air-gapped, compliance-first platform with strict tenant isolation.

LLM inferenceFine-tuningRAG
Open reference
HybridMulti-cluster

Shared Enterprise GPU-as-a-Service

Multi-tenant GPU pool with quotas, queues, and showback.

LLM trainingLLM inferenceFine-tuningDev sandbox
Open reference
Public cloudSingle cluster

Batch Inference Platform

Throughput-first offline scoring and embedding generation.

LLM inferenceVisionMultimodal
Open reference
EdgeMulti-region

Edge Inference Cluster

Power-constrained inference near the data, with central sync.

LLM inferenceVisionSpeech
Open reference
HybridSingle cluster

GPU Development Sandbox

Fast, isolated GPU access for experimentation before production.

Dev sandboxFine-tuningLLM inference
Open reference
Saved designs

No saved designs yet.

Open the designer →