rtx-5070

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

Fork of NullMoth's NVIDIA Metal driver for macOS, kept for one RTX 5070 Laptop (10DE:2D18) in a Legion 5 15IAX10 on Tahoe: 115 W TGP mode, WindowServer shader cache fix, A8 render targets, Dock badge digits, scaled desktop modes (upstream PR 273 ahead of its merge), nvgfx-smi. PolyForm Noncommercial 1.0.0, credit NullMoth Systems.

Q3_PLE quantization and learned MTP research for Qwen3.8-Flash-Next on an RTX 5070 12GB.

OpenCore 1.0.8 EFI, ACPI tables and config for macOS Tahoe 26.7.2 and Sequoia on the Lenovo Legion 5 15IAX10 (Core Ultra 9 275HX Arrow Lake-HX, RTX 5070 Laptop, MT7925): shutdown fix, MADT rewrite, DSDT patches, USB port fixes, kext list. Identity replaced by placeholders.

Reproduce Strata Flash-Next on Windows RTX5070+V10032GB, with V100-only Qwen27B RANA comparison and measured results.

Windows RTX 5070 + Tesla V100 dual-GPU llama.cpp kit with driver coexistence notes, GPU monitor, portable launcher and sanitized evidence

CUDA stress tool that checks whether your PSU can handle NVIDIA GPU power transients. It runs sustained loads, 1-200 Hz square waves and bursts from idle (optionally with a full CPU load), logs power, clocks and temperature through NVML to CSV, and plots the results. Windows and Linux.

Pre-built onnxruntime-gpu 1.24.1 with Blackwell sm_120 CUDA kernels (RTX 5090/5080/5070)