BLK · COMPATIBILITY PROBE
Will it run?
Identify your hardware. We compute which open-weight AI models you can run locally, how fast, with what tradeoffs.
> Backed by real benchmarks when available, computed from memory bandwidth + model architecture otherwise. Confidence tier always shown next to every number.
MODULE 01 · QUICK PRESET
Pick a common build
windows
RTX 5090 + Ryzen 9 9950X + 64GB DDR5
macos
MacBook Pro M4 Max 64GB
windows
RTX 5080 + Ryzen 9 9950X + 32GB DDR5
macos
MacBook Pro M4 Max 128GB
windows
RTX 4090 + Ryzen 9 7950X + 64GB DDR5
linux
RTX 3090 + Ryzen 9 5950X + 64GB DDR4 (used market)
windows
RTX 5070 Ti + Ryzen 7 9700X + 32GB DDR5
windows
RTX 3060 12GB + Ryzen 5 5600 + 32GB DDR4 (cheapest path)
windows
RTX 5060 Ti 16GB + Ryzen 5 9600X + 32GB DDR5
windows
Lenovo Legion 5 Pro Gen 7 (RTX 3080 16GB)
windows
RTX 4060 Ti 16GB + Ryzen 5 7600 + 32GB DDR5
windows
RTX 4080 Super + i7-14700K + 32GB DDR5
macos
Mac Studio M3 Ultra 256GB
windows
RTX 4070 Ti Super + Ryzen 7 7700X + 32GB DDR5
linux
RX 7900 XTX + Ryzen 9 7950X + 64GB DDR5 (Linux/ROCm)
linux
RX 9070 XT + Ryzen 7 9700X + 32GB DDR5 (Linux/ROCm)
windows
Razer Blade 16 (2025, RTX 5090 Mobile 24GB)
linux
RTX PRO 6000 Blackwell + Threadripper PRO + 128GB
MODULE 02 · CLUSTER TOPOLOGYMulti-GPU or cluster setup?
Dual GPUs, quad workstation, mixed cards, Apple unified, Exo cluster — total VRAM ≠ pooled VRAM, and we explain why.▼
MODULE 02 · CLUSTER TOPOLOGY
Multi-GPU or cluster setup?
Dual GPUs, quad workstation, mixed cards, Apple unified, Exo cluster — total VRAM ≠ pooled VRAM, and we explain why.
vLLM tensor-parallel 4× H100 80GB workstation
expert300/320 GB·single node multi gpu
4× Mac Mini M4 Pro Exo cluster (256 GB total)
expert180/256 GB·apple cluster
Mac Studio M3 Ultra 192GB
beginner140/192 GB·apple unified·POOLED
Quad RTX 3090 (24 GB × 4)
advanced88/96 GB·single node multi gpu
Ray Serve multi-node distributed inference (4 nodes × 2× RTX 4090)
expert80/192 GB·distributed
Dual RTX 3090 (24 GB × 2)
intermediate46/48 GB·single node multi gpu
Dual RTX 4090 (24 GB × 2)
intermediate45/48 GB·single node multi gpu
RTX 4090 + RTX 3090 (asymmetric 24+24 GB)
advanced42/48 GB·mixed gpu
MODULE 03 · CUSTOM BUILDER
Build your own rig
Pick any GPUs (single/dual/mixed/multi-node), set CPU/RAM/OS, choose runtime + use case. We compute effective VRAM honestly, recommend a runtime, and tell you which models fit comfortably.
OR · BASIC SINGLE-GPU FORM
MODULE 04 · MANUAL INPUT
Your hardware
> all four fields required
LEGEND
MEASUREDBenchmarked on our hardware. Reproducible.
EXTRAPOLATEDEstimated from same-bandwidth hardware benchmark.
ESTIMATEDComputed from memory bandwidth + model architecture.