1. 安装 AMD GPU 驱动和 ROCm
wget https://repo.radeon.com/amdgpu-install/7.1.1/ubuntu/noble/amdgpu-install_7.1.1.70101-1_all.deb
sudo apt install ./amdgpu-install_7.1.1.70101-1_all.deb
sudo apt update
sudo apt install "linux-headers-$(uname -r)" "linux-modules-extra-$(uname -r)"
sudo apt install amdgpu-dkms
sudo apt install python3-setuptools python3-wheel
sudo usermod -a -G render,video $LOGNAME
sudo apt install rocm
2. 通过 GTT 解锁 96G 显存
- 进入 BIOS -> Setup -> Advanced -> NBIO common options -> GFX configuration -> UMA Frame buffer size 修改显存大小。
- 修改 GRUB 配置。
vim /etc/default/grub
# 设置 GRUB_CMDLINE_LINUX_DEFAULT="quiet splash amd_iommu=off ttm.pages_limit=27648000 ttm.page_pool_size=27648000 amdttm.pages_limit=27648000 amdttm.page_pool_size=27648000 apparmor=0"
update-grub
重启生效。
3. 安装 Vulkan
sudo apt install vulkan-tools
4. 下载 llama.cpp release
wget https://github.com/ggml-org/llama.cpp/releases/download/b7503/llama-b7503-bin-ubuntu-vulkan-x64.tar.gz
# 解压后进入目录
# 从 HuggingFace Mirror 获取模型，例如：https://hf-mirror.com/ggml-org/collections
# 启动服务，模型将自动下载到 ~/.cache/llama.cpp/ 目录
./llama-server -hf ggml-org/gpt-oss-120b-GGUF -c 0 --jinja
5. 安装 amdgpu_top
sudo dpkg -i amdgpu-top_0.11.0-1_amd64.deb
amdgpu_top
