Local AI Hardware
Mini PCs, unified-memory systems and DIY builds for running language models locally.
14 items · human-reviewed
See the ranked shortlist →Hardware
Mac Studio (M3 Ultra)
Apple
9.8Score—View Details →
Mac Studio M2 Ultra
Apple
9.8ScoreGPU76-core GPUEnthusiastMemoryUp to 192GB Unified
Nvidia GeForce RTX 5090 (32GB)
Nvidia
9.7ScoreMemory32GB GDDR7—Cuda Cores21760
Mac Mini M4 Pro (64GB)
Apple
9.5ScoreUnified memory64 GBSweet spotThroughput35-50 tok/s
NVIDIA GeForce RTX 4090
NVIDIA
9.5ScoreTdp450WEnthusiastVram24GB GDDR6X
Nvidia GeForce RTX 5070 Ti (16GB)
Nvidia
9.3ScoreMemory16GB GDDR7—Cuda Cores8960
ClawBox
ClawBox
9.2Score—View Details →
Minisforum MS-S1 Max
Minisforum
9.2Score—View Details →
Threadripper Pro + Dual RTX 4090 Build
9.2ScoreTotal VRAM48ProfessionalRAM256 GB
Nvidia GeForce RTX 5080 (16GB)
Nvidia
9.1ScoreMemory16GB GDDR7—Cuda Cores10752- MinisforumImage unavailable
Minisforum MS-01
Minisforum
8.5ScoreForm factormini-pcSweet spotExpandableYes 
Nvidia GeForce RTX 3090 (Usada)
Nvidia
8.5Score—View Details →
GMKtec EVO-X2 (128GB)
GMKtec
8.0ScoreUnified memory128 GBEnthusiastThroughput15-25 tok/s
Ryzen 5 + RTX 4060 Ti Build
7.5ScoreVRAM16 GBEntryRAM64 GB
Further reading
- guideHow to build a local AI server with used GPUs: the RTX 3090 & Tesla P40 guide
- guideRunning and fine-tuning open models on Windows (WSL2, DirectML, CUDA)
- guideOptimising local LLM inference: where the bottleneck actually is
- guideBuilding a Tesla P40 rig for local AI: what you're really signing up for