Deep dives on Kubernetes, AI infrastructure, GitOps, and the cloud-native stack, written by practitioners.
GPU sharing in Kubernetes explained: time-slicing vs MPS vs MIG, every nvidia-smi command to enable and disable MIG on one GPU or eight, GPU Operator automation, pitfalls, and DCGM monitoring.
A beginner-friendly guide to local LLM inference, with the same Qwen model tested through Ollama, llama.cpp, Docker Model Runner, vLLM, SGLang, and TensorRT-LLM on NVIDIA DGX Spark.
Real Bonsai 27B benchmarks on an RTX PRO 6000 and a DGX Spark, including the supported llama.cpp setup, ternary vs 1-bit results, and speculative decoding.
7 parts
A week-long journey from your first docker run to production-ready containers.
by Saloni Narang5 parts
Running serious local LLMs hands-on on NVIDIA DGX Spark, from unboxing to 120B-parameter models.
by Saiyam Pathak2 parts
Put your AI agents behind a gateway: lock down their tools, then see what they cost.
by Shubham KataraDeep dives on the world's most-deployed orchestrator.
Container internals, image building, and developer workflows.
Running AI/ML workloads on Kubernetes and modern infrastructure.
CI/CD, GitOps, IaC, and the platform-engineering playbook.
Hardening containers, Kubernetes, and the supply chain.
The OS underneath every container, cluster, and cloud.
Newsletter
Subscribe to get our latest Kubernetes, AI infra, and cloud-native articles delivered to your inbox.
Subscribe on SubstackPowered by Substack · Unsubscribe anytime
191 posts · page 1 of 13
GPU sharing in Kubernetes explained: time-slicing vs MPS vs MIG, every nvidia-smi command to enable and disable MIG on one GPU or eight, GPU Operator automation, pitfalls, and DCGM monitoring.
A beginner-friendly guide to local LLM inference, with the same Qwen model tested through Ollama, llama.cpp, Docker Model Runner, vLLM, SGLang, and TensorRT-LLM on NVIDIA DGX Spark.
Real Bonsai 27B benchmarks on an RTX PRO 6000 and a DGX Spark, including the supported llama.cpp setup, ternary vs 1-bit results, and speculative decoding.
kiac runs local Kubernetes on macOS where every node is its own lightweight VM via apple/container: kubeadm or k3s flavors, Cilium on a custom kernel, built-in LoadBalancer, Grafana, Gateway API, and clusters that survive reboots.
Part 2: scrape agentgateway with Prometheus, build a Grafana dashboard of token cost and per-tool usage, see blocked tool calls, and alert on spend.
Run AI agents behind agentgateway on Kubernetes: route their LLM and MCP tool calls through one proxy, keep secrets out of the agent, and block tools by policy.
A practical, beginner-friendly guide to BF16, FP8, NVFP4, MXFP4, INT4, and GGUF Q4_K_M on NVIDIA DGX Spark. Bytes per parameter, quality vs size, and which format to pick when.
A practical teardown of NVIDIA DGX Spark's GB10 Grace Blackwell Superchip, unified memory, sm_121, NVFP4 tensor cores, memory reporting, and decode limits.
A practical Wandler deep dive with a local M1 Max WebGPU demo, real latency numbers, architecture diagrams, and getting-started commands.
Day-one deep dive into mlxcel v0.1.0, a Rust-native MLX inference engine. Real M1 Max benchmarks vs mlx-lm and Ollama on Llama 3.2 3B and Qwen 2.5 7B, with architecture diagrams and an honest take on TurboQuant.
A beginner-friendly walkthrough of tokenization, prefill, KV cache, decode, batching, TTFT, and why memory bandwidth shapes local LLM performance on NVIDIA DGX Spark.
How kube-proxy turns Kubernetes Services into kernel rules. iptables, IPVS, nftables packet paths and which to pick in 2026. Verified against k/k 1.36 source.
Why local LLMs are becoming practical in 2026, what changed across open weights, hardware, and inference software, and why DGX Spark makes the desk feel like a small AI lab.
A Service has no pod IPs in it. We covered that in the last post. So somewhere, something is keeping a list of every pod IP that matches the Service's…
NVIDIA just open-sourced the full NVCF platform under Apache 2.0. Not a thin SDK, not a client library. The actual control plane, invocation plane,…