P104-100 (GP104) PCIe Gen2 x4 unlock tool for Windows. Gen1 x4 -> Gen2 x4: 0.78 -> 1.55 GB/s measured.

windows pascal benchmark gpu reverse-engineering nvidia pcie hardware-hacking vbios cryptomining p104 mining-gpu pcie-gen2 gp104

• Updated Nov 1, 2026
• C#

CMP 40HX 解锁状态恢复手册（本机实测：算力解锁 + PCIe Gen2 x16）

windows nvidia turing cmp pcie vbios gpu-unlock mining-gpu cmp-40hx cmp40hx

• Updated Oct 16, 2026
• PowerShell

支持 NVIDIA P106-100 / P106-090（10DE:1C06 / 1C07 / 1C09）的 PCIe Gen2 解锁工具，包含源码、驱动、自动破解与开机自启。Unlock NVIDIA P106 PCIe Gen2 with source, drivers and autostart.

windows nvidia autostart pcie p106 gpu-unlock mining-gpu pcie-gen2 p106-100 p106-090

• Updated Oct 29, 2026
• C#

CMP 90HX full unlock: SM compute (0.72 -> 18.2 TFLOPS) + PCIe Gen2 (minimal 2-mask), NVIDIA open 610.43.03

cuda nvidia unlock cmp90hx mining-gpu pcie-gen2

• Updated Sep 15, 2026
• Python

Unlock NVIDIA CMP 100-100 mining card from 6 GB to full 12 GB HBM2 on Linux. ~$25 GP100 card for local LLMs (Qwen3.8-27B ~32 tok/s on two cards).

linux pascal gpu nvidia gp100 llama-cpp local-llm llm-inference hbm2 mining-gpu cmp-100-100 cmp100-100 vram-unlock

• Updated Nov 4, 2026
• C

将 NVIDIA P106-90 (10DE:1C09) 的 PCIe 链路宽度从 x4 解锁到 x16 的 UEFI EFI 注入方案。本项目只针对 x16 宽度，不独立承诺 Gen2/Gen3 速率。Unlock P106-90 PCIe width from x4 to x16 via UEFI EFI.

pascal nvidia uefi efi pcie p106 gpu-unlock mining-gpu p106-90 pcie-x16

• Updated Oct 30, 2026
• PowerShell

64 GB of HBM2e on a dead mining card: running Qwen3-Next-80B on an NVIDIA CMP 170HX with vLLM. Configs, gotchas, and measured benchmarks.

benchmark gpu nvidia homelab a100 llm-serving vllm local-llm llm-inference qwen cmp-170hx mining-gpu

• Updated Nov 1, 2026
• Python

Layer split vs tensor parallelism in llama.cpp - measured on two platforms 16x apart in interconnect bandwidth

pascal benchmark gpu cuda inference self-hosted quantization multi-gpu model-parallelism nccl v100 pipeline-parallelism tensor-parallelism llm llama-cpp qwen speculative-decoding mining-gpu p104-100

• Updated Nov 12, 2026
• Python

What it costs to split an image generation model across GPUs - layer vs row split vs worker parallel, measured on P104-100 x4 (PCIe Gen1 x4) and V100 x4 (PCIe Gen3 x16)

flux pascal benchmark gpu cuda self-hosted image-generation quantization multi-gpu model-parallelism v100 diffusion-models tensor-parallelism stable-diffusion-cpp mining-gpu p104-100

• Updated Sep 16, 2026
• Shell