עב
עב

Stack Sizer

Size a self-hosted stack from official sources.

Apps

Pick one or more apps. Only published profiles can be sized.

Load and options

LocalAI

v1.0.0
Load

Model up to 8B params at Q4_K_M (≤5.1 GB file, e.g. gallery qwen3-8b), 8,192-token context, 1 request at a time, one model loaded.

Options

Recommended resources

App requirements + system baseline + safety headroom. Nothing else is added, and nothing is rounded to a machine size.

vCPU 5.52, RAM 10.5 GiB, Disk 16.153 GiB

vCPU5.52
RAM10.5GiB
Disk16.153GiB
Breakdown
LineFromvCPURAM (GiB)Disk (GiB)Sources
LocalAILocalAI · Medium load4.57.55.94612345678910
System baselineOperating system, Container runtime (Docker Engine, containerd), System logs (journald), plus logs for 1 containers0.11.258.1111213141516
Safety headroomvCPU 20%, RAM 20%, disk 15% of apps + baseline0.921.752.1071712181519
Recommended total5.5210.516.153
Sources
  1. MaziyarPanahi/Qwen3-8B-GGUF: Qwen3-8B.Q4_K_M.gguf, 5,027,783,904 B (LocalAI, read 14 Oct 2026)
  2. Qwen/Qwen3-8B config.json: 36 layers, 8 KV heads, head_dim 128 (8.2B parameters) (LocalAI, read 14 Oct 2026)
  3. LocalAI v4.11.0 gallery/qwen3.yaml: backend llama-cpp, context_size 8192, mmap true (LocalAI, read 14 Oct 2026)
  4. llama.cpp server README: KV cache type -ctk/-ctv default f16; --cache-ram default 8192 MiB (LocalAI, read 14 Oct 2026)
  5. LocalAI docs, Text generation (llama.cpp): one parallel slot unless set; host prompt cache cache_ram 8192 MiB with idle-slot saving on by default since 4.3 (LocalAI, read 14 Oct 2026)
  6. LocalAI v4.11.0 pkg/vram/estimate.go: memory estimate = GGUF size + KV cache + overhead (0.02 × 2 × GGUF size + 0.15 GB) (LocalAI, read 14 Oct 2026)
  7. LocalAI docs, Model configuration: threads default = processor count (example threads: 8), cache_type_k/v map to llama.cpp -ctk/-ctv (LocalAI, read 14 Oct 2026)
  8. Docker Hub manifest of localai/localai:v4.11.0 (linux/amd64): 18 layers, 305,525,063 B compressed; layers decompressed and counted: 878,025,216 B (LocalAI, read 14 Oct 2026)
  9. Docker Hub manifest of localai/localai-backends:v4.11.0-cpu-llama-cpp (linux/amd64, same digest as latest-cpu-llama-cpp): 1 layer, 50,037,362 B compressed; decompressed: 172,602,368 B (LocalAI, read 14 Oct 2026)
  10. Docker: containerd image store is the default on fresh Engine 29+ installs and keeps images compressed and uncompressed (LocalAI, read 14 Oct 2026)
  11. Ubuntu Server: System requirements (Sizing policy, read 10 Oct 2026)
  12. Google Kubernetes Engine: About GKE node sizing (Sizing policy, read 10 Oct 2026)
  13. ANAROQ measurement: dockerd and containerd at rest (Ubuntu 24.04, Docker 29.8.2) (Sizing policy, read 10 Oct 2026)
  14. Docker Community Forums: Docker resource usage & overhead (Sizing policy, read 10 Oct 2026)
  15. journald.conf(5), Ubuntu Manpages (Sizing policy, read 10 Oct 2026)
  16. Docker Docs: Local file logging driver (Sizing policy, read 10 Oct 2026)
  17. AWS Compute Optimizer: Rightsizing recommendation preferences (Sizing policy, read 10 Oct 2026)
  18. Kubernetes: Node-pressure Eviction (Sizing policy, read 10 Oct 2026)
  19. mke2fs(8), Ubuntu Manpages (Sizing policy, read 10 Oct 2026)
Basis
Catalog release
2026.10.10
Sizing policy
2026.10.0
Profile LocalAI
1.0.0

How the numbers are built

Deployment Spec

Turn this result into a link you can keep or share. Everything is in the link; nothing is sent to us.

A short lowercase image id, for example ubuntu-24.04.