> BROMANDER_LABS

CAN YOURUN IT?

Honest, math-based estimates for running local LLMs on your hardware. VRAM fit, KV cache, and tokens-per-second — calculated the way llama.cpp actually runs them. Stack up to eight cards and see what layer split really buys you versus tensor parallelism.

All math runs in your browser. We never see your specs.

01 — Hardware

GPU

How many

System RAM

02 — Model

LLM

The "default" local model. Sweet spot for 8–16 GB VRAM.

System report

Verdict

S-TIER

38.4

tokens / sec

Fits comfortably in VRAM. You're running Llama 3.1 8B at full GPU speed.

Best fit: Q8_0auto-selected

Memory budget

Memory required9.98 GB / 11.0 GB VRAM usable

╳ 11.0 GB

Weights8.53 GB

KV cache0.54 GB

Overhead0.91 GB

VRAM

12 GB

Bandwidth

504 GB/s

Context

→ Digital wellness

Running models is easy. Running yourself is harder.

Shinery tracks how AI fits into your day so it stays a tool, not a crutch.

Open Shinery

─── Your Report Card ───

This is the actual image that shows on X, LinkedIn, and Facebook when you share the link.