rtx-3070

run a 27b local ai on your gaming gpu, rtx 3060 and up, radeon rx 6600 and up in beta, 8gb or more, linux and windows, one line to install

KytyPS5 Demon's Souls PC with NVIDIA DLSS Frame Generation for RTX 30 series (Ampere) — Native 1440p 60 FPS verified on RTX 3070

FP32-accumulate tensor-core unlock for RTX 3070 LHR (10de:2488) on Windows via a pre-OS UEFI app. Measured mma ratio 0.507 -> 1.000. Guide, patch, reproducible build, verification.

Deploy Bonsai-2-27B (ternary 27B, 5.95GB) on a single 8GB consumer GPU via an AI agent. 用 AI 部署 Bonsai-2-27B：一条提示词让 AI 装驱动、编 CUDA、下模型并接入 Claude Code。实测 51 tok/s @4k / 39 tok/s @100k，含按机器调参指南与完整测评报告。

FP16/BF16 tensor speed-unlock for the RTX 3070 Laptop (10de:249d, GA104M) on Windows 10 — MUXless Optimus adaptation of Zagitta/impacifist gpu-unlock. ~2x FP16/BF16/TF32, verified with PyTorch + raw mma.sync.

Full-rate tensor cores (FP16/BF16/TF32) on GeForce RTX 3070 under Windows. Based on impacifist/Windows-GPU-UNLOCK and Zagitta/gpu-unlock.

2B 级多模态模型在 8GB 显存上的可行性核对表：权重 + KV Cache + 激活 + 框架开销