Observed arrival · 2026-11-18
LLM Roofline
A browser-based calculator for estimating open-weight LLM memory use, inference speed, and hardware-versus-cloud costs.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- People sizing hardware to run open-weight language models
- Worth noticing
- The interface includes Apple unified-memory machines, discrete GPUs, multi-GPU setups, and cloud accelerators in the same hardware selector.
Field notes
The model selector includes dense models as well as MoE entries showing total and active parameter counts. Hardware profiles range from Mac minis and consumer GPUs to multi-GPU rigs and cloud accelerators, with capacity and bandwidth shown in the selector. The visible controls also let users vary context length, prompt size, concurrency, and cache format; the page presents a sample memory breakdown with weights, KV cache, overhead, and headroom.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue