About StackCostGrid

Independent Cloud Cost & Hardware Sizing for Developers

StackCostGrid is an open capacity planning calculator designed to help developers estimate real infrastructure demands and monthly hosting bills for AI microservices, self-hosted automation, and full-stack web applications.

Built by a Solo Engineer

StackCostGrid is created and maintained by Md. Ashraful Haque. It was built out of frustration with vague cloud marketing claims and surprise overage bills. When developers evaluate self-hosting vs. SaaS APIs, they need transparent math: what does the operating system consume? What happens when a build spikes? Which VPS tier actually fits?

Sizing Methodology & Principles

750MB OS Kernel & Daemon Baseline

A bare Ubuntu 24.04 LTS server with systemd, journald, SSH, and the Docker engine allocates ~750MB of RAM before user containers start. Ignoring this baseline leads directly to Out-Of-Memory (OOM) crashes on 1GB and 2GB instances.

Steady-State vs. Build Spike Separation

Compilers, bundlers, and package managers (like next build) consume transient memory spikes up to 1.5GB above runtime. We separate sustained runtime from compilation spikes and recommend NVMe swapfiles only to absorb safe transient bursts.

No Swap for LLM Inferences

Linux swapfiles are dangerous for LLMs and databases. Swapping model weights or active database caches across disk causes catastrophic latency spikes and token stalls. We enforce hard warnings whenever sustained runtime exceeds physical VPS RAM.

Shared vCPU Transparency

Budget VPS tiers frequently use shared vCPUs. While excellent for background automation and web servers, sustained CPU-bound LLM token generation may trigger hypervisor throttling. We document these trade-offs clearly.

Pricing & Independence

All cloud provider specs and monthly pricing (Contabo, Vultr, DigitalOcean, Railway) are checked directly against official provider storefronts. Outbound links are standard direct links. If referral tracking is introduced in the future, it will be explicitly disclosed, and referral relationships will never alter our technical sizing formulas or hardware recommendations.