From GPU kernels to production systems
AI Infra Glossary
Understand GPU execution, memory, CUDA, networking, training and serving through 110 terms, nine illustrated cases and linked study paths.
Paid Premium or approved access
The guide library
Understand the systems.
Build your way up.
Illustrated guides to AI infrastructure, from your first GPU to running a fleet. Browse the covers, read the samples and find your next step.
The accelerator PDF is free with an account. All other guides, including the Proxmox GPU Server Guide, have public previews; their PDFs require paid Premium or approved complimentary access. Referral Premium excludes paid guide downloads.
All guides
15 illustrated PDFs
Your first complete view of AI infrastructure
The Beginner's Guide to AI Infrastructure
Follow one request from a prompt to the GPU. Learn model memory, cables, training, serving, security and cost through original illustrations, worked examples and a laptop companion.
Paid Premium or approved access
From one action to a trained agent
AgentGym Engineering Handbook
Build interactive environments, inspect trajectories and GRPO updates, size rollout capacity and evaluate production candidates. Covers AgentGym, AgentGym-RL, ScalingInter-RL and AgentGym2, with a tested CPU companion.
Paid Premium or approved access
Run, measure and repair LLM serving
LLM Inference in Production
Defend vLLM, SGLang and TensorRT-LLM launch commands flag by flag. Budget memory and the KV cache, benchmark with goodput, alert and autoscale correctly, and follow runbooks for OOMs, p99 regressions, failing GPUs and NCCL hangs.
Paid Premium or approved access
From the connector to the cluster
GPU Networking for AI Engineers
Recognize cables, adapters and rear panels. Design NVLink and RDMA networks, count a 1,024-GPU fabric, place firewalls and validate the complete path.
Paid Premium or approved access
Build and secure a GPU host
Proxmox GPU Server Guide
Install Proxmox, plan GPU ownership and firewall boundaries, and rehearse recovery. Includes real demo screenshots, B200 fabric requirements and platform comparisons.
Paid Premium or approved access
Start with the foundations
AI Infrastructure Field Guide
Understand how GPUs, memory, models and networks work together. Start with the basic terms, then build up to inference, distributed training and running systems in production.
Paid Premium or approved access
From one GPU to a fleet
The LLM Inference Systems Design Handbook
Follow a request through an LLM serving system. Work through cache capacity, batching, latency and scaling calculations, then practice the design decisions in mock interview rounds.
Paid Premium or approved access
Run inference across a cluster
Distributed Inference on Kubernetes
Learn how a fleet routes requests, moves cached attention state and scales its workers. Compare NVIDIA Dynamo and llm-d through deployment examples, failure scenarios and capacity calculations.
Paid Premium or approved access
Operate the model you choose
Self-Hosting Frontier Open-Weight Models
Plan and operate an open-weight model service: per-rank memory, compressed attention, precision, expert traffic, engine acceptance, long context, recovery and cost. Includes worked design rounds and a Python companion.
Paid Premium or approved access
Build the reinforcement learning loop
RL Post-Training Infrastructure
Build the reinforcement learning loop from rollout to policy update. Work through model memory, weight transfer, stale trajectories, code sandboxes, numerical mismatch, MoE routing and recovery, with a runnable Python companion.
Paid Premium or approved access
Plan the cost of useful AI
The Cost-per-Token Playbook
Plan an AI service budget from useful outcomes to tokens, GPUs and peak demand. Worked scenarios, seven interactive calculators and an offline Python companion explain API bills, caching, capacity, break-even, ownership and human review.
Paid Premium or approved access
Defend a complete system design
AI Infrastructure System Design
Work through complete designs for serving, training and shared GPU platforms. Derive capacity, trace state through failures, challenge an unsafe architecture, and rehearse design rounds with an executable calculation companion.
Paid Premium or approved access
Choose and compare hardware
AI Accelerator Field Guide
Get familiar with the GPUs and TPUs used for AI. Compare memory and connectivity, estimate whether a model fits, and plan what to measure before choosing a machine.
Free with an account
Understand the model you serve
Open Model Field Guide
Follow open model releases and the architecture changes behind them. Connect each design to its memory needs, deployment options and the checks that matter for your workload.
Paid Premium or approved access