AMD 780M iGPU Runs 35B LLMs: Sub-€1000 Local Inference Setup
MaximusSenior · reddit · 2026-08-09
A Reddit user shares a budget-friendly local LLM inference setup using AMD Ryzen 7 260 (or similar) CPUs with 780M iGPU and 64GB DDR5 RAM, running llama.cpp with Vulkan on Ubuntu.
- Hardware cost: barebone mini PC €300-400, used 64GB DDR5 €500, total well under €1000.
- Kernel params like amdgpu.gttsize=49152 map system RAM as VRAM, yielding 48GB "VRAM".
- Performance (Q8 quant): Qwen 3.6 35B-A3B generates 21 t/s; Gemma 4 31B 2.5 t/s, but with MTP reaches 5.76 t/s.
- Bonus: adding a small discrete GPU (e.g., RTX 5060 8GB) can boost MoE models via partial expert offloading.
More from Infra
- MLCommons launches MLPerf Endpoints, a benchmark that measures inference endpoints the way buyers actually purchase them — TheKanter · 2026-10-16
- Neocloud M&A decoded: CoreWeave, Nebius, Crusoe, Nscale all bet margins vanish beyond GPU scarcity — AccBalanced · 2026-10-16
- AI-Driven Timing-Aware Power Optimization Uses Learned Ranking With Baseline-Preserving Repair — Amit Sharma · 2026-10-16
- Nscale AI datacenter plagued by coolant leaks, transformer fire, 10 safety violations — saranormous · 2026-10-16
- WSL 3.0 ships with native containers and Linux 6.18 kernel, syscall overhead down ~16% — solyarisoftware · 2026-10-16
- Are 22 GB modded RTX 2080 Ti cards a viable budget local LLM option? — Dave86ch · 2026-10-16