DeepSeek Hits 300 tok/s on Old ThinkPad T480
dee_hw · x · 2026-08-24
A user deployed a local AI environment on a salvaged ThinkPad T480. By combining Omarchy (Linux), OpenCode, and Local AI Grid, they successfully ran the DeepSeek V4F model at approximately 300 tokens/s, demonstrating the viability of running local AI on older hardware.
More from Infra
- Xiaomi's New Chip Matches Apple Single-Core Performance, Features SME2 Matrix Acceleration — lemire · 2026-08-24
- llama.cpp docs get a fresh redesign with speculative decoding and agents coming soon — mervenoyann · 2026-08-24
- Data center vacancy stays at 1% with tenants booking 2028 capacity — Beth_Kindig · 2026-08-24
- Australia Plans Legislation for AI and Data Center Rules Amid Energy Boom — nordicinst · 2026-08-24
- Cut RAG Token Costs by 67% with Qdrant's Native Features — qdrant_engine · 2026-08-24
- MCP reveals roadmap: HTTP support and long-running workloads on the horizon — _philschmid · 2026-08-24