March 13, 2026
Missing GPUs, Mac flexes, and speed wars
Can I Run AI locally?
Did your computer make the cut? Missing graphics cards, Mac memory drama, and speed flexes
TLDR: A new tool estimates which AI models your computer can run, but the comments pounced on missing hardware, Mac memory limits, and “too pessimistic” speed numbers. Fans want benchmarks and a model-first view, turning this into a buyer’s guide debate that matters for anyone eyeing local AI.
A new “Can I run AI locally?” page is lighting up the comments, promising simple answers about which AI models your computer can handle — using browser estimates of memory and speed. But the community is here for the drama. One user kicked the door in with, “RTX Pro 6000 is a glaring omission,” setting off the classic “my graphics card is special” energy. Apple fans rolled up right after: one M3 Ultra owner pointed out the site caps out at 192GB RAM even though “M3 Ultra supports up to 512GB.” Translation: the Mac crowd wants their memory flex recognized.
Then came the speed wars. A Radeon owner called the whole thing “pessimistic,” bragging that a mid-range AMD card runs a hefty model at snappy speeds at home. That clash sparked the bigger question: are these estimates too cautious, or are people cherry-picking best-case scenarios? Meanwhile, a practical chorus asked for real benchmarks and the ability to pick a model first and see which machines run it — a feature that would turn the page into a buyer’s cheat sheet.
Amid the spice, there’s genuine love: “Oh how cool. Always wanted a tool like this.” It’s the perfect internet combo — useful tool, messy specs, and a comments section that doubles as consumer council. If you’ve ever wondered “can my machine run one of these AI brains?” the crowd is saying: yes, but show us the receipts — and don’t forget our hardware.
Key Points
- •The tool estimates local AI model feasibility using WebGPU and browser APIs, with a disclaimer that actual specs may vary.
- •Model entries list parameters, memory footprint, context window, release date, and architecture (Dense or MoE), plus use-case tags (chat/vision/code/reasoning).
- •Small models (0.8B–1B) require about 0.5 GB memory and target on-device or edge use; mid-range (8B–32B) range from ~4.1 GB to 16.4 GB.
- •Large models include Dense (e.g., 70B to 123B) and MoE systems with active parameter counts (e.g., 109B–1T total), requiring 35.9 GB to over 500 GB.
- •Context windows span from 16K to 256K tokens across listings, covering diverse capabilities and sizes.