rtx-5060-ti

RTX 5060 Ti performance work for NInfer, targeting Qwen3.8-27B on one or two 16 GB cards.

Researching local AI image editing on affordable consumer hardware: models, ComfyUI workflows, benchmarks, and real outputs on an RTX 5060 Ti 16GB.

NInfer fork: two-GPU tensor parallelism (--tp 2) for Qwen3.8-27B on 5060 Ti 32GB(16GB x2ea) without P2P. Developed by Claude Opus 5.5 (medium effort) in Claude Code.

Practical local LLM recipes & benchmarks for dual RTX 5060 Ti setups (tensor-split, MTP, llama.cpp)

Reproducibility files, benchmarks and deployment guide for the Nex-N2.5-mini Mixed Q2/Q3 128K GGUF release.

Strata tuned for an RTX 3090 + RTX 5060 Ti: Qwen3.8-Flash-Next 125B MoE with unsloth UD-IQ4_XS at 93 tok/s (3090 alone) / 116 tok/s (both cards) and UD-Q4_K_XL at 71 / 85 tok/s, on a Ryzen 7 5700X with 128 GB DDR4 (Linux). eddoursul's fork + fixes + Niko1221 engine ports (elastic K/V), with measurements and configs.

RTX 5060 Ti 16GB on macOS Tahoe: Ryzen Hackintosh, verified Metal compute, 1440p 180Hz output and owner-tested RTX-only operation.

Fish Audio S2 Pro as a real-time local TTS voice on a 16 GB GPU (RTX 5060 Ti, Windows) next to a 12B LLM: audio.cpp GGUF streaming build notes, benchmarks, samples, adapter, Ollama prompt-cache trick. EN + PT-BR.

Run DeepSeek V4 Flash locally with bounded MoE streaming on dual 16 GB NVIDIA GPUs

本地推理生产线①：NInfer 单卡长上下文线（引擎内建 KVMem 分页 KV 池）