System Specifications

Nvidia Driver: NVIDIA-Linux-aarch64-570.86.16.run
Host: Ampere Altra + ALTRAD8UD
Host OS: Ubuntu 22.04 with HWE kernel (6.8)
Guest OS: Ubuntu 22.04 (ubuntu-22.04-live-server-arm64.iso)
GPU: Nvidia RTX 4080 16GB

Assumptions

1. You are familiar with Ubuntu and its basic commands.
2. You have experience using Virt-manager.
3. All commands are executed as the root user.

If anything is unclear, you can refer to external resources for additional guidance.

Host Configuration

Enable IOMMU

To enable IOMMU, you need to enable the SR-IOV option in the BIOS and verify whether the Linux kernel has IOMMU enabled by default.

You can check if IOMMU is enabled by running:

$ dmesg | grep -i iommu

Example output:

[    0.000000] Kernel command line: BOOT_IMAGE=/boot/vmlinuz-6.8.0-52-generic root=UUID=6b78fa89-a575-432d-a445-1497c3467214 ro iommu=on
[    0.000000] Unknown kernel command line parameters "BOOT_IMAGE=/boot/vmlinuz-6.8.0-52-generic iommu=on", will be passed to user space.
[   11.561684] iommu: Default domain type: Translated
[   11.566470] iommu: DMA domain TLB invalidation policy: strict mode

If IOMMU is not enabled, add iommu=on to the Linux kernel boot parameters:

$ vim /etc/default/grub 

Modify the line:

GRUB_CMDLINE_LINUX_DEFAULT="iommu=on"

Then update GRUB and reboot:

$ update-grub2  
$ reboot  

Additionally, enable SR-IOV in the BIOS. The exact location of this setting varies depending on the BIOS, but it is typically found under the PCIe subsystem or related options.

Upgrade Host to HWE Kernel

I recommend using the Hardware Enablement (HWE) kernel on the host. While I’m unsure if the regular kernel works, the HWE kernel has been reliable in my experience. Install it with:

sudo apt install linux-generic-hwe-22.04

Configure VFIO on Host

The VM relies on the VFIO driver for GPU passthrough. To configure VFIO, you need to pass the PCIe device information to the VFIO driver.

First, identify the GPU’s PCIe device IDs:

$ lspci -nn

Example output:

0005:01:00.0 VGA compatible controller [0300]: NVIDIA Corporation Device [10de:2704] (rev a1)
0005:01:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:22bb] (rev a1)

Here, 10de:2704 is the GPU’s PCIe device ID, and 10de:22bb is the audio device ID. At a minimum, you need to pass through the GPU device.

Next, edit the VFIO configuration file to include these IDs:

$ vim /etc/modprobe.d/vfio.conf

Add the following line:

options vfio-pci ids=10de:2704,10de:22bb

Disable Nvidia Driver on HOST

To prevent the host from loading the Nvidia driver, add the Nvidia modules to the kernel’s blocklist:

$ vim /etc/modprobe.d/blacklist.conf

Add the following lines:

blacklist nvidia
blacklist nvidia_drm
blacklist nvidia_modeset

Update the initramfs and reboot:

$ update-initramfs -u
$ reboot

Configure VM

Install Virt-manager

In this article, we used virt-manager as VM manager, first step is install virt-manager, suppose Ubuntu will install all relative packages.

$ apt install virt-manager 

If you’re using SSH with X11 forwarding (e.g., ssh -X host) or MobaXTerm on Windows, Virt-manager will display the remote X window. If neither method works, consider installing a KDE desktop on the host and accessing it via the BMC remote console.

(Optional) Install KDE Plasma Desktop:

(option)
$ apt install kde-plasma-desktop

Create VM image

Virt-manager creates fixed-size VM images by default. If you prefer dynamic allocation, create the image manually:

$ qemu-img create -f qcow2 ubuntu2204.qcow2 200G

Add Nvidia device to VM

If the host is configured correctly, Virt-manager will list all PCIe devices, including the Nvidia GPU. Add the GPU and its audio device (e.g., 0005:01:00.0 and 0005:01:00.1) to the VM’s hardware list.

After adding the devices, proceed with the Ubuntu 22.04 installation.

After add hardware

Now, it can run begin install to install ubuntu 22.04

Disable secure Boot in UEFI

By default, Virt-manager enables Secure Boot. However, Nvidia drivers may not work with Secure Boot enabled. Even though the Nvidia installer includes a driver signing feature, the driver may still fail to load. To avoid issues, disable Secure Boot in the VM’s UEFI settings.

During the VM’s boot process, press the DEL key to enter UEFI settings and uncheck the Secure Boot option.

Before installing the Nvidia driver, ensure the necessary development packages are installed:

$ apt install build-essential

Then, install the Nvidia driver and reboot the VM.

GPU Passthrough Test

If everything is set up correctly, running nvidia-smi should display the GPU’s status.

For testing, you can use Ollama with the Deepseek-R1 model. Install Ollama with:

curl -fsSL https://ollama.com/install.sh | sh

Pull the Deepseek-R1 model. Since the GPU has 16GB of memory, the 14B model is a good choice (it requires ~10GB):

ollama run deepseek-r1:14b

Ask a question like, “Why is the sky blue?” This will trigger the model’s Chain-of-Thought (CoT) reasoning.

Monitor the GPU’s status using nvidia-smi to ensure it’s functioning correctly.

The post GPU Passthrough on ARM64 with Libvirt/Virt-manager appeared first on richliu's blog.

In this article, I will skip other steps such as installing Ollama.

System Specifications:

• OS: Ubuntu 22.04 on an ARM64 virtual machine
• CPU: Ampere Altra
• DRAM: 32GB (allocated to the VM, though smaller models do not require this much memory)

At the time of writing, Infinity does not support the ARM64 platform, but that is not an issue since we do not need it. RAGFlow uses Elasticsearch as its default AI database engine, and Elasticsearch does support the ARM64 platform.

Building RAGFlow on ARM64

Build ragflow command [ref. Build a RAGFlow Docker Image]

git clone https://github.com/infiniflow/ragflow.git
cd ragflow/
docker build --build-arg LIGHTEN=1 -f Dockerfile -t <username>/ragflow:<version> .

docker build --build-arg LIGHTEN=1 -f Dockerfile -t user/ragflow:v0.16.0 .

If everything is set up correctly, the Docker images will display an output similar to the following:

docker images
REPOSITORY                                      TAG                            IMAGE ID       CREATED         SIZE
user/ragflow                                    v0.16.0                        8a71ac9cb2fa   3 hours ago     5.73GB

Note that this image does not include pre-built models. However, models can be installed separately using other methods.

Running RAGFlow

Before running the docker-compose command, assign the RAGFlow Docker image name to the docker-compose configuration.

export RAGFLOW_IMAGE=user/ragflow:v0.16.0
docker compose -f docker/docker-compose.yml up -d

If everything is set up correctly, you should see a message similar to the following:

[+] Running 10/10
  Network docker_ragflow      Created                                                                         0.2s
  Volume "docker_minio_data"  Created                                                                         0.0s
  Volume "docker_redis_data"  Created                                                                         0.0s
  Volume "docker_esdata01"    Created                                                                         0.0s
  Volume "docker_mysql_data"  Created                                                                         0.0s
  Container ragflow-mysql     Healthy                                                                        21.7s
  Container ragflow-minio     Started                                                                         1.1s
  Container ragflow-redis     Started                                                                         1.1s
  Container ragflow-es-01     Started                                                                         1.1s
  Container ragflow-server    Started                                                                        22.3s

Downloading Models with Ollama

Ollama needs to download certain models for system functionality. Here’s an easy way to download the models—you can choose the ones you prefer.

ollama pull deepseek-r1:14b
ollama pull deepseek-r1:32b
ollama pull smartcreation/bge-large-zh-v1.5:latest

After downloading the Ollama models, you need to add them in the “USER → Model Providers → Add Ollama Model” section. Below is an example of how to add the Deepseek model. Once you understand the process, you can add more models as needed.

At a minimum, you will need one chat model and one embedding model.

After pressing OK, you can configure the System Model Settings, which should look something like this:

Using RAGFlow

Now, you are ready to use the Knowledge Base and Chat features to explore RAGFlow.

For example, you can ask the engine to generate an ARM SIMD sample code based on the ARMv8 documentation.

Conclusion

RAGFlow is an easy-to-use RAG framework. Although it does not provide a default ARM64 image, you can still build it from source to obtain a functional version of RAGFlow.

The post Ragflow on ARM64 appeared first on richliu's blog.

no-dangling-assignment-gsl -std=c++20 -Wno-trigraphs -gsimple-template-names -fno-exceptio00:01:51 [472/19443]
nc++ -isystem../../third_party/libc++/src/include -isystem../../third_party/libc++abi/src/include -fvisibility
-inlines-hidden -O2 -pipe -march=armv8.2-a -Wno-unknown-warning-option -c ../../third_party/libyuv/source/comp
are_neon64.cc -o obj/third_party/libyuv/libyuv_neon/compare_neon64.o
../../third_party/libyuv/source/compare_neon64.cc:175:8: error: instruction requires: dotprod
  175 |       "udot        v4.4s, v0.16b, v6.16b         \n"
      |        ^
<inline asm>:14:1: note: instantiated into assembly here
   14 | udot        v4.4s, v0.16b, v6.16b
      | ^
../../third_party/libyuv/source/compare_neon64.cc:176:8: error: instruction requires: dotprod
  176 |       "udot        v5.4s, v1.16b, v6.16b         \n"
      |        ^
<inline asm>:15:1: note: instantiated into assembly here
   15 | udot        v5.4s, v1.16b, v6.16b
      | ^
../../third_party/libyuv/source/compare_neon64.cc:204:8: error: instruction requires: dotprod
  204 |       "udot        v4.4s, v0.16b, v0.16b         \n"
      |        ^
<inline asm>:10:1: note: instantiated into assembly here
   10 | udot        v4.4s, v0.16b, v0.16b
      | ^
../../third_party/libyuv/source/compare_neon64.cc:205:8: error: instruction requires: dotprod
  205 |       "udot        v5.4s, v1.16b, v1.16b         \n"
      |        ^
<inline asm>:11:1: note: instantiated into assembly here
   11 | udot        v5.4s, v1.16b, v1.16b
      | ^
4 errors generated.
[10940/63891] aarch64-unknown-linux-gnu-clang++-18 -MD -MF obj/third_party/libyuv/libyuv_neon/rotate_neon.o.d

Seems because some -march/-mcpu/-mtune drop ‘dotprod’ parameters, it needs to add the extension manuall. ‘+dotprod’.

Edit the file /etc/portage/make.conf, and change the COMMON_FLAGS to

COMMON_FLAGS="-mcpu=neoverse-n1+crc+crypto+ssbs+dotprod -O2 -pipe"

It can use the following command to get current CPU flag

gcc -v -E -x c /dev/null -o /dev/null -march=native -mcpu=native 2>&1 | grep /cc1

Thanks Gentoo user zukunf to provide those information.

The post Gentoo ARM64 Altra Compile chromium Issue appeared first on richliu's blog.

The Dynatron W3 package comes in a standard case.

includes a tower heatsink, one 120mm 4-pin fan, and thermal paste.

The base is not entirely copper, and its area seems somewhat small

other parts

While no documentation is provided, installation is straightforward for those familiar with DIY PC building

There is a gap between the DDR slots and the CPU fan, but larger DIMMs might require disassembling and reinstalling the tower fan.

In a preliminary test using stress-ng with 128 threads on an AltraMax 128-30 CPU and ASRock Rack ALTRAD8UD-1L2T.

Test command is

stress-ng --cpu 128 --vm 128 --vm-bytes 128M --timeout 180s --metrics-brief

the fan operated at 1400 RPM and maintained a quiet profile. The CPU temperature reached 64°C, staying below 70°C, with power consumption around 178W.

This performance suggests the tower fan is suitable for general purposes.

Notice: The CPU’s TDP is rated at 250W, this maximum power draw is only achievable under full system stress, such as utilizing all 8 DRAM controllers and all PCIe root controllers. Therefore, this figure should be considered as a reference rather than a precise measurement.

The post Dynatron W3 for Ampere AltraMax appeared first on richliu's blog.

In previous article on How to Install Stable Diffusion GUI on ARM64 Nvidia RTX platform , we installed the Nvidia driver and ran PyTorch on the platform. Now, this platform has Nvidia RTX gaming card, can we use it to play game?

The answer is yes, there are some windows x86 or x64 emulators on Linux, but the most stable solution comes from Ampere’s official release solution. Other emulators might not work as well on this combination, such as FEX, but I don’t know whether it is a driver issue or another issue.

Basically, following Ampere’s documentation makes it easy to install the driver and play games. Let’s combine both documents into one and simplify it in this article.

Hardware
CPU: Ampere Altra
Board: AsRock ALTRAD8UD
GPU: Nvidia RTX 4080

Reference Ampere’s document
NVIDIA GPU Accelerated Linux Desktop on Ampere
Steam-on-Ampere

Install Nvidia Driver

All the conditions are similar to those outlined in the article How to Install Stable Diffusion GUI on ARM64 Nvidia RTX platform, using Ubuntu 22.04 with a general kernel (not the HWE kernel, which Nvidia may address in the future).

I haven’t utilized the original document’s script; instead, I’ve copied some necessary commands from it.

sudo apt-get install build-essential
# Install Nvidia driver here
sudo rmmod nouveau
cat << EOF | sudo tee /etc/modprobe.d/blacklist-nouveau.conf
blacklist nouveau
options nouveau modeset=0
EOF

sudo nvidia-xconfig -a --cool-bits=31 --allow-empty-initial-configuration

# Install Desktop, personally I like kde. 
sudo apt install kde-plasma-desktop
# or it can install regular ubuntu desktop 
# sudo apt install ubuntu-desktop

Download Nvidia ARM64 (aarch64) Display Driver ; the current version, 535.154.05, is acceptable. After installing the Nvidia driver, simply reboot the system. You can then use the following command to check the driver status.

nvidia-smi

Disable aspm from kernel

I didn’t apply this option. Most of the time, it’s only necessary to do so if the system encounters the massive AER error issue. Here’s an example command from the original document:

sudo nano /etc/default/grub
# Add pcie_aspm=off to kernel parameters
# GRUB_CMDLINE_LINUX_DEFAULT="pcie_aspm=off"
sudo update-grub

Install Box86

Following commands show how to install Box86

sudo wget https://itai-nelken.github.io/weekly-box86-debs/debian/box86.list -O /etc/apt/sources.list.d/box86.list
wget -qO- https://itai-nelken.github.io/weekly-box86-debs/debian/KEY.gpg | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/box86-debs-archive-keyring.gpg
sudo dpkg --add-architecture armhf
sudo apt update
sudo apt install box86:armhf -y

Install Box64

Following command show how to install Box64, but original instruction have update , a package name was changed.

sudo wget https://ryanfortner.github.io/box64-debs/box64.list -O /etc/apt/sources.list.d/box64.list
wget -qO- https://ryanfortner.github.io/box64-debs/KEY.gpg | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/box64-debs-archive-keyring.gpg
sudo apt update 
sudo apt install box64-arm64 -y
sudo systemctl restart systemd-binfmt

Installing Steam

Because I am using the Ubuntu 22.04 ARM64 server version with minimal, so, some packages might not be installed, Before Install Steam, install necessary packages

sudo apt install lsof

sudo apt install zenity
sudo apt install alsa

Use the following to install steam.

git clone https://github.com/ptitSeb/box86
cd box86
./install_steam.sh

Now, it can use steam now.

/usr/local/bin/steam

Okay, log in and enjoy it. However, in my experience, when installing other emulators, it might damage the box86+Steam setup, necessitating a system reinstallation to fix it.

Furthermore, the game’s performance seems not as good as expected, possibly due to my choice of Vampire Survivors and enabling high performance

Xbox JoyStick controller

I have an Xbox controller, and ALTRAD8UD doesn’t have Bluetooth capability, so I cannot use wireless conntions. However, it can connect via USB.

There are several Xbox drivers available for Linux. After testing, I found that xpadano is compatible with this platform. Simply follow their instruction guide to install the Xbox driver.

# Install DKMS package
sudo apt-get install dkms linux-headers-`uname -r`

# download source code
git clone https://github.com/atar-axis/xpadneo.git
cd xpadneo
# use DKMS to compile the driver
sudo ./install.sh

The Xbox controller driver will be installed automatically, and it works well in my tests. You can install the jstest package to perform a joystick test.

The Xbox controller functions seamlessly on Steam and Vampire Survivors. I can easily use the Xbox controller to play Vampire Survivors.

The post Play Game On Ampere Altra ARM64 Platform appeared first on richliu's blog.

In this article, it will walk through the process of installiung Stable Diffusion WebUI on ARM64 + Nvidia RTX platform.
For such as hardware combinarion, it needs to solve some compatiability issues, so, here will describe how to install it and fix those issues.

Platform

CPU: Ampere Altra ARM64 CPU
Board: AsRock ALTRAD8UD
OS: Ubuntu 22.04

The platform is the most powerful ARM64 platform for a homelab that we can obtain today. There are some server models available, but they come with high-speed cooling fans that are not suitable for home use.

Nvidia has provided comprehensive library support for the ARM64 platform, including drivers, CUDA, and cuDNN. However, despite these efforts, there are still gaps in support. Even Nvidia’s Jetson product line lacks adequate support for ARM64 platforms in certain applications such as PyTorch. While there exists an individual PyTorch accelerator package for direct download on x64 platforms, the same cannot be said for ARM64 platforms. Although there is technically one package available, its availability is severely limited, rendering it practically useless.

The only viable solution to overcome these limitations is to utilize Nvidia’s Docker service to install Stable Diffusion.

Installation

Install Ubuntu 22.04 without selecting the HWE (Hardware Enablement) support. It appears that the current Nvidia driver version has issues, leading to potential soft-lock bugs. A general kernel version (5.15) is preferable for stability.

Next, install the necessary Ubuntu packages.

sudo apt-get install build-essential

Download Nvidia ARM64 (aarch64) Display Driver , current version is ok, when I write this article, current version is 535.154.05

Docker

Install Docker by following the Docker guide Install Docker Engine on Ubuntu,”. Also provide the necessary installation steps here. If already installed Docker or if these steps don’t work, please refer to the original website for updated instructions.

# Add Docker's official GPG key:
sudo apt-get update
sudo apt-get install ca-certificates curl
sudo install -m 0755 -d /etc/apt/keyrings
sudo curl -fsSL https://download.docker.com/linux/ubuntu/gpg -o /etc/apt/keyrings/docker.asc
sudo chmod a+r /etc/apt/keyrings/docker.asc

# Add the repository to Apt sources:
echo \
  "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/ubuntu \
  $(. /etc/os-release && echo "$VERSION_CODENAME") stable" | \
  sudo tee /etc/apt/sources.list.d/docker.list > /dev/null
sudo apt-get update

sudo apt-get install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin

Nvidia Container Toolkit

Follow the instructions for installing the NVIDIA Container Toolit. It can modify the Docker configuration using this toolkit. Alternatively, modifying the daemon.json file might achieve the same result, although I haven’t personally tested it. Below are the instructions:

curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
  && curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

sudo sed -i -e '/experimental/ s/^#//g' /etc/apt/sources.list.d/nvidia-container-toolkit.list

sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit

sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

It can check the file /etc/docker/daemon.json file, the configuration should be like this.

{
    "runtimes": {
        "nvidia": {
            "args": [],
            "path": "nvidia-container-runtime"
        }
    }
}

Run Nvidia pytorch docker image

Following is example command to run the Nvidia pytorch docker image

sudo docker run --gpus all -it --rm \
        -v /host/work:/container/work \
        -p 7860:7860 \
        nvcr.io/nvidia/pytorch:XX.XX-py3

/host/work means work directory on host, /container/work means work directory in container(docker). -p 7860 mean gui default port, it needs to forward to host machine. XX.XX means container versiom please see tags to get container version, or this URL . [NGC Pytorch]. Following is example command.

sudo docker run --gpus all -it --rm \
        -v /work:/work \
        -p 7860:7860 \
        nvcr.io/nvidia/pytorch:24.01-py3

Stable Diffustion WebUI

First steps clone Stable Diffustion WebUI source code.

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

Then, it needs to modify the settings file. Edit webui.sh, and adjust the parameters to the following numbers.

use_venv=0
can_run_as_root=1

As we are operating within a container, there’s no need for a virtual environment (venv), and the script can be run as root. Modify the file webui-user.sh and add COMMANDLINE_ARGS accordingly.

export COMMANDLINE_ARGS="--no-half --precision full --listen"

or run it directly

export COMMANDLINE_ARGS="--no-half --precision full --listen" ; bash webui.sh

Currently, Stable Diffusion will download some packages, but errors may occur during the installation process. We’ll address and fix these errors in the next few steps.

CV2

Error message :

    if __load_extra_py_code_for_module("cv2", submodule, DEBUG):
  File "/usr/local/lib/python3.10/dist-packages/cv2/__init__.py", line 28, in __load_extra_py_code_for_module
    py_module = importlib.import_module(module_name)
  File "/usr/lib/python3.10/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "/usr/local/lib/python3.10/dist-packages/cv2/typing/__init__.py", line 168, in <module>
    LayerId = cv2.dnn.DictValue
AttributeError: module 'cv2.dnn' has no attribute 'DictValue'

Modify file “/usr/local/lib/python3.10/dist-packages/cv2/typing/__init__.py”, comment out line 169 like following.[ref]

# LayerId = cv2.dnn.DictValue

CodeFormer

Error message:

"/work/stable-diffusion-webui/repositories/CodeFormer/facelib/utils/face_restoration_helper.py", line 7, in <module>
        from facelib.detection import init_detection_model
      File "/work/stable-diffusion-webui/repositories/CodeFormer/facelib/detection/__init__.py", line 11, in <module>
        from .yolov5face.face_detector import YoloDetector
      File "/work/stable-diffusion-webui/repositories/CodeFormer/facelib/detection/yolov5face/face_detector.py", line 20, in <module>
        IS_HIGH_VERSION = tuple(map(int, torch.__version__.split('+')[0].split('.'))) >= (1, 9, 0)
    ValueError: invalid literal for int() with base 10: '0a0'

This issue caused by pytorch.version will return illegeal number like 2.2.0a0+81ea7a4, so it will fail to parse this mesage, temperory solution is to remove the version number, edit file “stable-diffusion-webui/repositories/CodeFormer/facelib/detection/yolov5face/face_detector.py”. and modify line 20, IS_HIGH_VERSION to following version to following:[ref]

version_string = torch.version.split('+')[0]
version_string = ''.join(filter(str.isdigit, version_string))
IS_HIGH_VERSION = tuple(map(int, version_string.split('.'))) >= (1, 9, 0)

Now, the Stable Diffusion GUI should be working as well as expected

The post How to Install Stable Diffusion GUI on ARM64 Nvidia RTX platform appeared first on richliu's blog.

English please see : Install ARM64 Windows 11 on ARM64 QEMU System

現在 ARM64 的系統相對成熟了，最近 Windows 11 for ARM64 也可以在系統上安裝，自然也會有人想弄到 QEMU 上去，不過看來看去，大部份都沒有整理完畢的文章

這邊記錄一下需要準備的東西

準備

1) 系統：Ampere Altra CPU
RPI4 or RPI5 這種開發板應該也可以，因為這些板子只有有限的 CPU Power 和記憶體，就沒有安裝 qemu 的必要性，便宜一點的解決方案就是 Apple Mx 系列的 CPU [ref]，因為也是 ARM64 ，而且很多人也是買來安裝 Linux

2) ARM64 ISO Image
網路上應該很多，不過如果要相對合法取得 ISO image ，推薦 Windows-11-On-Ampere
這步驟有點複雜，不過重點是用 UUP Dump 抓回 Windows ISO ，然後用 Rufus 燒到 USB Flash ，在燒錄的時候 Rufus 可以解除一些 Windows 11 的限制，像是 TPM Support 等等這些在 ARM64 CPU 還不完全支援的設定

3) Windows Virtio Driver

下載 Windows virtio driver，然後放進去剛剛那一步做到的 USB Disk

Proxmox Windows VirtIO Drivers 這邊有下載點， download the latest stable or  download the most recent ，都可以下載到 VirtIO ISO，下載完將 ISO 解開放到 USB Disk

4) 將 USB Dump 成 Image
Linux user 應該很熟這個步驟

dd if=/dev/sda of=/storage/Win11_ARM64_INSTALL.img

為什麼要這樣做？最大的原因就是 QEMU 內 Windows 現在似乎沒辦法同時設定二個 ISO 設備，我之前這樣裝會卡住。所以就繞了一大串讓它可以裝，還可以吧

5) virt-manager
其他人能找到的範例都是 qemu command ，不過我這次實在是不想打一堆指令，只想點點點，所以這次的範例都是用 virt-manager ，不過我相信背後的思路都是一樣的

New Virtual Machine

Virt Type: KVM
Architecutre: aarch64
Machine Type: virt

下一步，ISO image 選剛剛做好的 Win11_ARM64_INSTALL.img，Choose the operaion system you are installing: 選 Microsoft Windows 11

CPU 和記憶體我用 16 cores 和 16GB，小一點應該也可以，不過我懶得試了

如果空間不夠，Disk Image 建議自己設定，virt-manager 設定會一開始就 allocate 設定的 disk space ，不省空間，Windows 11 最小需求空間是 52G ，可以設 60G 或是更大，看需求
這邊是範例命令，可以預先建好硬碟空間

qemu-img create -f qcow2 MyImage.qcow2 60G

在執行之前選擇 Customize configuration before install，有些東西要修改，記得要先開啟 virt-manager 內的 XML 修改功能，在主程式下的 Edit -> Perferences -> Enable XML editing ，請勾選

在 CPUs 的 XML 這一段，找到 hyperv 這一段，改成如下的設定

    <hyperv mode="custom">
      <relaxed state="off"/>
      <vapic state="off"/>
      <spinlocks state="off"/>
    </hyperv>

在 CPU 數量，要從 sockets 改成 Cores，以下就是 16 Cores 的範例

Add hardware ，加入以下設定
Input -> USB Keyboard
Input -> EvTouch USB Graphics Tablet
Graphics -> Spice server
Video -> Ramfb

Video 選項一定要用 Ramfb ，否則 Graphics 會出不來，qxl or virtio 都不能用，virtio-gpu-pci 沒試過，有人說可以用[ref]
最後是 Storage 增加 USB CDROM device，並且將剛剛的 Win11_ARM64_INSTALL.img 設定上去

在 Boot Options 這邊 enable USB CDROM 1的選項，然後就可以開機了

如果設定正確，那就可以正常開機到這個畫面

如果照著以上步驟卻在安裝過程卡住了，可以試著調整 boot option 切換一下開機的順序
先 disk 後 USB CD ROM .

Install Windows

正常狀況下，Windows 是找不到 disk 的，要選 load driver -> 25982_1000_ARM64_EN-US(C:) -> Virtio-driver -> viostor -> win11 -> ARM64 (註：這部份的標籤和目錄可能因不同的設定而異）

最後一段，NIC driver ，在初次進入 Windows 11 時，要求要連線到網路時，選 Install driver -> 25982_1000_ARM64_EN-US(D:) -> Virtio-driver -> NetKVM-> win11 -> ARM64

這樣就可以完成安裝 Windows 11 了

Happy Playing Windows 11 on ARM64 Qemu

ref.
virtio-win / virtio-win-pkg-scripts

The post 在 ARM64 QEMU System 上安裝 ARM64 Windows 11 appeared first on richliu's blog.

Chinese version Article is here (Ampere Altra/AltraMax 加AsRock ALTRAD8UD-1L2T 水冷方案)
2024/01/31 Update: Ampere will have 2U passive cooler Making Arm Desktops Viable Ampere Altra Noctua NH-D9 AMP-4926 4U and NH-U14S AMP-4926

The Ampere Altra/AltraMax CPU features a maximum of 128 ARM64 cores, typically found in cloud server. However, cloud server Care not suitable for placement in offices or homes due to the high-speed fans used for heat dissipation. The noise generated by these fans is generally intolerable for regular users. Additionally, adapting server motherboards to standard consumer motherboards is inconvenient. If there is a desire to utilize such a CPU in a different setting, the ADLink Ampere Altra Dev Platform presents a viable solution, complete with the option for liquid cooling.

Another solution is the AsRock ALTRAD8UD-1L2T, currently available on Newegg. Paired with an Ampere Altra Q64-22 CPU (64 cores, 2.2GHz), it comes with a complimentary 2U passive heatsink, priced at US$1,500, which can be considered affordable. However, it’s worth noting that the included heatsink is passive. If you prefer liquid cooling over installing a fan on the passive heatsink, then this article is the right guide for you.

The author have adding a declaration at the beginning of articles that may pose potential risks:

• This article is not an official endorsement by Ampere, Newegg, or any other mentioned merchants or vendors. It is solely a personal record.
• The author has a business relationship related to the production of the mentioned motherboard.
• The author is a first-time to use liquid cooling computer system , and the content may not be entirely accurate. Please be aware of potential risks.
• This article is not an official document from Ampere or AsRock.
• The original article is in Chinese, and the English translation is done using machine translation. Please forgive any inaccuracies in wording.
• I am not responsible for any consequences resulting from this article; for example, losses incurred due to improper implementation of instructions provided in this article.

The Ampere system primarily focuses on Cloud-native CPUs, and as of now, there is no manufacturer producing general-purpose cooling systems for it. The included 2U passive heatsink in this system is designed for server use, utilizing high-speed server fans for heat dissipation. While the Q64-22 included by Newegg has lower power consumption, its only cooling method involves direct airflow from an electric fan. Alternatively, users may resort to attaching an 8-inch fan onto it. Such cooling solutions are deemed unacceptable for enthusiasts and professionals like myself.

If you intend to upgrade the cooling system, typical tower-type coolers require replacing the fixture on the back of the motherboard. Moreover, compatibility with the fastening mechanisms of this specific motherboard can be challenging to find. Considering that only front-facing screws can be used, the only viable cooling solution is a liquid cooling system. Therefore, I have been experimenting with a liquid cooling solution that works for this scenario. I’m sharing this information for those adventurous individuals who might encounter similar challenges, and perhaps this approach can inspire some creative solutions.

In the upcoming sections of this article, I will guide you on how to set up a liquid cooling system specifically tailored for the Ampere Altra on the ALTRAD8UD-1L2T motherboard. In theory, since the screw hole positions are the same, this method should be applicable to other Ampere motherboards as well. However, the focus of this content will primarily be on the ALTRAD8UD-1L2T. The article will also cover any pitfalls encountered during assembly and the items needed for the setup(Chinese website).

Preparation

ALTRAD8UD-1L2T

You can find the specifications for this motherboard online, and I won’t repeat them here. However, what might confuse some people is the term “Deep mATX.” This is a specification unique to AsRock, where they increase the depth dimension of the mATX form factor while maintaining the standard width. The purpose is to provide more space to accommodate additional components. This particular specification is commonly seen in AsRock’s server-grade motherboards.

Chassis

Generally, most computer cases should work. However, I made a mistake on my end by choosing a case that was too small. The case I purchased is the Antec P110 Silent. While it has liquid cooling mounting points, it is not suitable for adding a liquid cooling system. The internal space is too limited, lacking a designated area for securing the water pump. Additionally, the Antec P110 liquid cooling radiator cannot be placed at the top as it interferes with the motherboard. Consequently, the radiator had to be positioned at the front of the case.

When making a purchase, it’s crucial to choose a case that is suitable for water cooling. In more extreme cases, considering a caseless system might be an option (hopefully, this can be updated in the next release).

Cases designed for liquid cooling systems typically have more significant internal space to accommodate a water pump, and there is ample room for fans without obstruction. Some all-in-one water cooling designs may be more compact and might not encounter this issue.

Liquid Cooling System

Liquid cooling systems have some considerations to take into account:

• Tubing type: Choose between soft or hard tubing. Hard tubing is suitable for experienced users, offering a longer lifespan and a more aesthetic appearance. Soft tubing is easier for assembly, making it suitable for beginners.
• All-in-one (AIO) or custom loop: Decide whether to go for a pre-assembled AIO liquid cooling system or build a custom loop system.

Since the initial focus was on researching liquid cooling, the combination chosen was a custom loop with soft tubing. Building a custom loop may encounter some issues, with one of the main challenges being the uncertainty about specifications.

The first crucial aspect is determining the size of the liquid cooling tubing, as additional components will depend on this size, limiting the options for procurement. Currently, the mainstream sizes are two-eighths (2/8’), three-eighths (3/8’), and four-eighths (4/8’) liquid cooling tubing.

Another parameter is the thickness of the tubing, thick tubing and thin tubing.
* thick tubing refer to 10/16mm or 3/8”-5/8” tubing.
* thin thbing refer to 10/13mm or 3/8”-1/2″ tubing.

Most common the inner diameter specification is G 1/4, when find some liquid colling componens like pressure relief valve needs it.

Once you’ve confirmed your inner and outer diameter, you can proceed to purchase the necessary liquid cooling components.

At a minimum, you’ll need:

• CPU water block *1
• Water Pump *1
• 240mm radiator *1
• 12cm fans *2
• Liquid coolant 500cc ~ 1000cc
• Tubing 2M
• Thermal paste
• Pressure relief valve
• Fittings: Quick connectors *6, 90-degree rotary fittings *(1~2)
• Tools for filling and draining: Filling tool, draining tool (watering can, large syringe, drainage hose)

Optional components:

• Flow meter, temperature gauge (note: buying these requires purchasing two additional quick connectors).

In practical testing, if focusing solely on CPU cooling, it appears that a 240mm radiator is sufficient. The thin three-eighths (3/8”) tubing provides a large enough water flow diameter. Choosing a smaller tube diameter may not be as suitable for this purpose.

CPU water block

This turned out to be quite troublesome. My original plan was to purchase a 50mmx50mm heatsink and connect it to the heatsink on the right. Then, I intended to secure it to the CPU socket using the fixtures shown in the image. However, as expected, unexpected issues arose – the world isn’t always that kind.

1. There was no water block with a 50mmx50mm connector available. So, I ended up buying a 40mmx40mm connector, one size smaller. This resulted in poorer heat dissipation performance. Fortunately, I never had the chance to test the inferior heat dissipation, so this step failed unnoticed.
2. Here’s where I made a mistake. Custom water block connectors, like the one shown, are typically 2/8″ fittings. However, the main liquid cooling system I bought has 3/8″ fittings, making them completely incompatible. The options were either to change the fittings or change the entire system. Changing the fittings was the more cost-effective choice, so I opted for that.

After connecting , it looked like the image on the right – and just looking at it made me worry that water would leak out.

So, I bought a two-eighths(2/8″) tubing to connect to this connector. However, while it could connect, it couldn’t link up with my three-eighths(3/8″) tubing in the water cooling system. I had to reluctantly abandon the custom water block connector.

The final solution had nothing to do with this connector.

Communical CPU water block

During the installation, I discovered that the Bykski CPU-XPH-T8 CPU water block, along with the provided adapter bracket, could be securely attached to the Ampere Altra’s socket. This was fantastic news and saved me a great deal of effort.

Although it’s designed for Intel LGA1700/LGA2066 sockets, the flexibility in its hole configuration allowed for adjustments, providing me with the necessary adaptability. As for whether other CPU brackets might be compatible, it’s likely that some could work. However, I haven’t had the chance to try others yet. If you have attempted this and found success, please let me know.

The base of this water block is quite substantial, and my intuition tells me that the cooling performance should be excellent. Therefore, I didn’t explore other compliant CPU water blocks.

However, another issue arises. The water block is 56mmx56mm, whereas the width of the Ampere Altra CPU socket is 55mm. The water block does not fit. At this point, there are two possible solutions:

Upon observing the CPU’s structure, the first solution seemed unfeasible. Cutting down the thickness of this water block would compromise its structural integrity, not to mention finding the tools or a professional service for the task.

Thus, the only viable option is the second solution. The simplest and most intuitive way to implement this is by using a thermal pad made of silicone to transfer heat upwards from the CPU.

Silicone Thermal Pad

But things couldn’t be as simple as I, a fool, initially thought. In conclusion, regular silicone thermal pads are not suitable for power monsters like the Ampere Altra. This type of thermal pad may be effective for low-power devices such as routers, switches, NVMe disks, etc., but it’s not useful for high-power CPUs.

In the image, there are 2mm thick silicone thermal pads, and you need two layers because the height is just that much. Essentially, this provides only a slight improvement over having no heat dissipation at all, so it’s not worth considering.

As for the high-efficiency silicone thermal pads available in the market, I don’t know if they would work. If you have them on hand, you can give them a try. However, I believe the chances of success are quite low.

Purple Copper Thermal Pad

Nowadays, a common method for cooling is using purple copper heat pipes. This type of heat dissipation is quite effective, and the 2U passive heatsink included with Newegg utilizes these heat pipes. Therefore, if we can customize the heatsink, wouldn’t that solve the problem? Luckily, the country right next to my country is the world’s largest manufacturing factory, capable of customizing just about anything. This made the solution very affordable and instantly resolved my problem.

The dimensions of the purple copper thermal pad are: Length x Width x Height = 64x54x4mm, with a rounded R5 edge.

Alternatively, you can measure the dimensions of the piece under the 2U passive heatsink to get the specifications. Placing it on the original heatsink, I can say the dimensions match perfectly. The craftsmanship is quite impressive, and there are no issues when it’s actually placed.

Assembly

First, place the purple copper thermal pad, and remember to apply thermal paste before doing so.

Apply thermal paste before placing CPU water cooling block.

Assemble the components.

Actually, I assembled the tubing on the left side incorrectly. I’ll need to purchase materials and quick connectors to replace it when I have the time. But for now, let’s leave it as is and make do.

Overall view.

This way, it’s evident that the case isn’t ideal. It’s too cramped, affecting airflow, and it cannot accommodate the components properly.

Shopping

If you are interested in replicating my setup, all the components were purchased from Taobao. I spent extra for testing and experimenting, but in reality, you should be able to set up the entire cooling system for around US$100 or NTD$3,000. This is comparable to the cost of an efficient dual-fan tower cooler.

Since this is a Chinese website, I’ll keep the text in Chinese and provide English translations afterward.

Liquid Cooling System

watertiger定制台式机电脑分体式CPU显卡水冷软管套装冷散热器 (“WaterTiger Custom Desktop Computer Split-Type CPU GPU Liquid Cooling Soft Tube Kit Radiator”)
Enter the URL choice “單 CPU 水冷 240 冷排” (“Single CPU Liquid Cooling 240 Radiator”)

At first, I chose this brand because the photos looked appealing, but it turned out to be a fortunate coincidence that saved me a lot of exploration time. This brand integrates various components into a set for sale, and the components they use are of decent quality. For example, they use the Enermax NEOChanger as the water pump, which is also a good pump. If you are familiar enough, you might not need these kits.

This kit does not include a pressure relief valve. You’ll need to purchase a pressure relief valve separately, and depending on your needs, you might want to buy additional quick connectors

Clamping Fittings

All the components can be purchased from the same place, saving the hassle of searching everywhere. It’s a good idea to buy a few extra parts to avoid trouble later on. The time cost is much higher than the cost of the components.

电脑水冷头扣具卡具CPU显卡南北桥固定笔记本散热器压条五金夹具 (“Computer Water Cooling Head Fastening Fixture, CPU GPU North-South Bridge Fixing, Laptop Heatsink Clamp, Hardware Clamp”)

Choice 台式机CPU冷头扣具(“Desktop CPU Cooling Head Fastening Fixture”) *2

M2 M3 M4不锈钢平头盘头螺丝钉水冷头水冷排机箱风扇安装长螺丝 (“M2 M3 M4 Stainless Steel Flat Head Countersunk Screws for Water Cooling Head, Radiator, Case Fan Installation – Long Screws”)

Fastening fixtures for securing CPU coolers and water blocks. The components can be increased based on requirements. The minimum required quantity includes
M3手拧螺母 (M3 thumb nuts)*4，M3*30 (M3 screws) *4 ，M3垫片(M3 washers) *4 ，M3六角螺母(M3 hex nuts) *4

电脑水冷水流计 液冷散热器流量器 流体温度观测器透明水流指示计 (“Computer Liquid Cooling Flow Meter, Liquid Cooler Flow Indicator, Fluid Temperature Monitor Transparent Flow Meter”) (option)

Choice: 3孔水流观测器 + 温度计 (“3-Port Flow Meter + Thermometer for Water Flow Observation”)

Nice to have items, but it can make me feel better.

Test Result

This system feels relatively quiet and not noisy.

The following tests are only for verifying the liquid cooling setup and are not formal performance tests. The results are provided for liquid cooling reference and do not include testing power consumption, etc.

Room temperature: 25 degrees Celsius
CPU : Ampere AltraMax 128-30
Run command : stress-ng –cpu 128 –iomix 128 –vm 128

Gentoo Linux kernel 6.6 ，CPU temperature is approximately，Power consumption CPU 150W/ IO 19W
Ubuntu 22.04 kernel 6.2(?)，CPU temperature is approximately 72 degrees，Power consumption CPU 178W/ IO 20W

In conclusion, I hope this article has been helpful for those eager to try it out.

Afterword

Newegg has also listed another model, which is only the motherboard version, only US$849.-
ASRock Rack ALTRAD8UD-1L2T Deep Micro-ATX Server Motherboard Singel Socket (LGA 4926) Ampere Altra Max/Ampere Altra processors Dual 10G & 1G

The post Ampere Altra/AltraMax AsRock ALTRAD8UD-1L2T Liquid Cooling System appeared first on richliu's blog.

English version article is here (Ampere Altra/AltraMax AsRock ALTRAD8UD-1L2T Liquid Cooling System)
2024/01/31 Update: Ampere will have 2U passive cooler Making Arm Desktops Viable Ampere Altra Noctua NH-D9 AMP-4926 4U and NH-U14S AMP-4926

Ampere Altra/AltraMax CPU 最高單顆 CPU 是 128 核 ARM64 的 CPU，一般都是在伺服器主機上，但是伺服器主機非常不適合放在公司內或是家用，因為一般伺服器主機都會用高轉速風扇散熱，一般人是不能忍受這樣的噪音。伺服器主機板也不方便改裝到一般家用主機板，如果真的想要用，ADLink Ampere Altra Dev Platform 是一個不錯的解決方案，也有水冷的選項。
另一個解決方案就是 AsRock ALTRAD8UD-1L2T ，目前在 newegg 上架了，配合一個 Ampere Altra Q64-22 (64 cores 2.2G) 的 CPU，附贈一個 2U Passive 的散熱片，價格 US$1,500.- ，應該算便宜。但是送的是被動散熱，如果不想要裝一顆風扇到被動散熱上，想要改成水冷散熱，那看我這篇文章就對了。

本人習慣，在有可能造成危險的文章，會在文章前面加上宣言
* 本篇文章並非 Ampere, Newegg 或是其他任何文章內提到的商城及店家業配文，純屬個人紀錄
* 本文作者和部份這塊板子的生產有利益關係，特此揭示
* 作者是第一次接觸電腦系統水冷，內容可能不盡正確，請自行注意風險
* 本文並非 Ampere 或是 AsRock 官方文件
* 本人不對本篇文章的造成的任何影響負責任；例如看本人文章操作不當造成的損失等等

Ampere 系統主要是專攻 Cloud nativate CPU，所以直至目前為止，並沒有廠商生產一般用途的散熱系統。這個系統附贈的 2U Passive 散熱片是 Server 專用的，使用的是 Server 高速風扇散熱；雖然 newegg 附的是 Q64-22 ，功耗比較低，但是唯一的散熱方式就是拿電風扇直接吹，要不然就是卡一個8吋風扇在上面，這種散熱方案像我這樣的專業玩家是無法接受的

如果要進階散熱，一般的塔型散熱系統需要將主機板背面的夾具換掉，而且也要能跟這個板子的扣具相容，找到相容的塔型散熱難度可能也不小。考量到只能在正面鎖螺絲，那唯一能用的散熱系統就是水冷式散熱。所以我這段時間摸索出了一條可以用的水冷散熱解決方案，在此分享給想要踩雷的朋友們，或許這個做法也可以激發各位的創意。

接下來會在這篇文章內介紹要如何在 ALTRAD8UD-1L2T 板子上，搭建屬於 Ampere Altra 的水冷系統，理論上因為螺絲孔位一樣，所以 Ampere 其他的板子也可以用，但是本篇內容主要以 ALTRAD8UD-1L2T 為主，而本文亦會提到踩過的地雷或是採購項目。

準備

ALTRAD8UD-1L2T

上網可以找到這塊板子的規格，這些就不再重覆，但是一般人會比較迷惑的是，什麼是 Deep mATX ？
這個是 AsRock 自己的規格，是增加 mATX 的深度尺寸，保持 mATX 的寬度規格，以增加更多空間容納更多元件。這個特別的規格常見於 AsRock server grade 的主機板上。

機殼

一般機殼應該都可以。不過我這邊犯了一個錯誤，機殼買太小，我買的機殼是 Antec P110 Silent ，雖然有水冷孔位，但是這個機殼不適合加裝水冷，內部空間太小，沒有固定水泵的地方，Antec P110 水冷排無法放在上方，會卡到主機板，所以最後水冷排是放機殼前方

採購時，請找適合水冷的機殼，更極端一點可以考慮無機殼系統（希望下次更新可以更新無機殼系統）。

水冷機殼內部空間需求更大，要有地方放水泵，還有風扇放下去不會卡住。
有些一體式水冷的設計會比較小，比較不會有這個問題

水冷設備

水冷散熱有一些需要考慮的
* 軟管或是硬管（硬管比較適合有經驗的，壽命長，較美觀；軟體組裝方便，適合新手）
* 一體式水冷（廠商已經組裝好）或是分體式水冷（自組水冷系統）

因為一開始是研究如何建置水冷，所以一定是分體式水冷加軟管的組合。自組水水冷又會碰到一些問題，最主要的是不知道規格為何？
第一件需要確定的是水冷管大小，因為加購設備就是看水冷管大小，這會限制採購品項
目前主流有 二分(2/8’)、三分(3/8’)、四分(4/8’)水冷管，但是實際上在找產品的時候，會標為像 6*10(二分) 9.5*12.7(三分，有些會標 10*13) 另外還有 8*12 不知道對應到什麼，數字小的是內徑，數字大的是外徑 。三分又有分三分薄和三分厚，三分薄就是 10/13 ，三分厚就是 10/16 ，如果不是老手很容易在這邊搞錯。本文用的全部都是三分薄的系統

以上有英制和公制，要特別小心注意規格，像洩氣閥內徑規格就要用 G 1/4 規格的（目前大都是此規格）
確定好自己的內外徑之後，就可以採購需要的水冷套件

最少需要的是
CPU水冷頭*1
水泵*1
240寬冷排*1
12CM風扇*2
水冷液 500cc ~ 1000cc
水管 2M
散熱膏
洩壓閥
接頭：快速接頭*6 ，90度旋轉彎頭*(1~2個)
注水工具，放水工具（加水壺，大針筒，放水管）
可選：
流量計，水溫計（註：買這個要多買二個快速接頭）

實測如果只針對 CPU 散熱，目前看起來 240 水冷排就夠，而三分薄提供了夠大的水流口徑。
如果管徑再小可以不太合適。

CPU水冷散熱頭

這個頗麻煩的，我原來的規劃是買一個 50mmx50mm 的散熱片接到像右邊的散熱頭上。然後用圖中的夾具固定在 CPU 座上面，但是不出意外的就出了意外，這世界上沒這麼好的事情

1 ) 沒有 50mmx50mm 的水冷頭接頭，所以買的是 40mmx40mm 的接頭，小了一圈，散熱性能較差，幸運的是我根本沒機會測到散熱性能較差這步就失敗了

2) 我在這邊犯了一個錯誤，這種像是定製的水冷接頭都是 2/8″ 頭，但是我買的主要水冷系統是 3/8″ 的系統，完全接不上去，要就是要換頭，或是換系統；當然換頭比較便宜，所以選擇換頭

接上去之後像是右邊這樣，這光看都覺得水會跑出來吧

所以我買了二分管想要接這個接頭，但是二分管是可以接上去沒錯，但是接不上我水冷的三分管系統。只能果斷放棄接定製水冷頭

最後的解法已經和這個接頭無關了

商用水冷頭

在裝的時候發現買的套件附的 Bykski CPU-XPH-T8 這顆 CPU 水冷頭加轉接夾具可以合上到 Ampere Altra 的底座，這簡直太棒了啊。省了非常多功夫

雖然是給 Intel LGA1700/LGA2066用的，所以它的孔位設計成非常彈性，這就給了我調整的空間。至於其他 CPU 的夾具有沒有可能可以用，應該也有可以用的，不過這邊我還沒機會試，如果你有試過可以告訴我。
而這個水冷頭的底座非常大，直覺告訴我散熱效果應該非常好，所以也並未去找其他合規的 CPU 水冷頭

但是接下來又有另一個問題，這個水冷頭是 56mmx56mm ，而 Ampere Altra CPU socket 寬度為 55mm，這個水冷頭放不下。這時有二個選擇方案
1) 切割水冷頭讓它變小（還沒考慮高度問題）
2) 將熱從 CPU 接到水冷頭

第一個方案觀察了一下 CPU 工藝，覺得不可行，因為這顆水冷頭的厚度切下去會影響結構，更別提還要想辦法找到工具或是店家切割。

那就只能用第二個方案了，這個方案最簡單並且非常直覺的方式就是用矽膠導熱墊將熱拉上來

矽膠導熱墊 (thermal pad silicone)

但是事情怎麼會像我這個笨蛋想的這麼簡單，先說結論，一般的矽膠導熱墊不適用於 Ampere Altra 這種 Power Monster ，所以完全不用考慮。這種矽膠導熱墊給小功率的 router/switch/nvme disk 等等或許有用，對大功率的 CPU 是沒有用的

圖中是 2mm 厚的矽膠導熱墊，而且要二層，因為高度就是這麼高，基本上這個只比不接散熱好一點點而已，所以別想了

至於市面上賣的高效率矽膠導熱墊有沒有用我就不知道了，如果手上有現成的可以試試，不過我認為成功機率很低

紫銅導熱片

那現在一般的散熱是怎麼散熱呢？一般都是紫銅導熱管，這個散熱效率還不錯，newegg 附贈的 2U passive heatsink 就是使用這種導熱管散熱。所以如果可以客製散熱片，這樣不就解決問題了嗎？幸好我們旁邊就是世界最大的製造工廠，任何怪東西都可以客製，這樣一片非常便宜，瞬間解決了我的問題

紫銅導熱片的尺寸規格是
長x寬x高 = 64x54x4mm ，另外要倒 R5 的角

也可以量 2U passive heatsink 下面那一塊得到規格，放在原廠的散熱片上只能說尺寸一樣
工藝還真不錯，實際放下去也沒有問題

組合

先放上紫銅導熱片，記得放之前塗散熱膏

放上去之前再塗散熱膏

合體

其實這張圖左邊的管子我裝錯了，有空要再去買材料和快速接頭來換，但是現在就先這樣吧，撐一下。

全貌

這樣就看得出來這機殼不好了，太擠太小影響風流，不能放在上面

購物

如果你是想學我的方案，我所有的零件都是從 Taobao 購入，我因為有測試還有多花錢
但是實際上應該 US$100, NTD$3,000 左右可以搞定整個散熱系統。
跟一個效率好的雙風扇塔扇差不多

水冷系統

watertiger定制台式机电脑分体式CPU显卡水冷软管套装冷散热器
進去後選單 CPU 水冷 240 冷排

當初是看照片比較好看所以買這家的，但是誤打誤撞節省我很多摸索的時間。
這家也是將各家設備整合弄一套出來賣，但是東西用的還不錯，像水泵用的是 Enermax NEOChanger ，也是還不錯的水泵。如果夠熟是可以不用這些套件。

這個套件沒有洩壓閥，另外要多採購洩壓閥，還有依需求可以多買快速接頭

夾具

以下都在同一家採購，不用到處找，零件可以多買一點，免得到時候要找麻煩，時間成本遠高於零件成本。

电脑水冷头扣具卡具CPU显卡南北桥固定笔记本散热器压条五金夹具

選台式机CPU冷头扣具 *2

M2 M3 M4不锈钢平头盘头螺丝钉水冷头水冷排机箱风扇安装长螺丝

固定 CPU 冷頭扣具和水冷頭的零件，這個可以依需求增加，以下是最低需求量
M3手拧螺母*4，M3*30 *4 ，M3垫片*4 ，M3六角螺母*4

电脑水冷水流计 液冷散热器流量器 流体温度观测器透明水流指示计 (option)

選: 3孔水流观测器 + 温度计

可有可無的東西，不過有比較好玩

測試結果

這套系統個人感覺相對安靜，不會吵

以下測試只是驗證水冷測試，並非正式的 performance test ，結果僅供建置水冷參考，並非測試 power consumption等結果

室溫 25 度
CPU : Ampere AltraMax 128-30
Run command : stress-ng –cpu 128 –iomix 128 –vm 128

Gentoo Linux kernel 6.6 ，CPU 溫度約在 67 度，Power consumption CPU 150W/ IO 19W
Ubuntu 22.04 kernel 6.2(?)，CPU 溫度約在 72 度，Power consumption CPU 178W/ IO 20W

最後希望這篇文章有幫到躍躍欲試的你

後記

Newegg 還有上架另一個型號，只有主機板的版本，只賣US$849.-
ASRock Rack ALTRAD8UD-1L2T Deep Micro-ATX Server Motherboard Singel Socket (LGA 4926) Ampere Altra Max/Ampere Altra processors Dual 10G & 1G

The post Ampere Altra/AltraMax 加AsRock ALTRAD8UD-1L2T 水冷方案 appeared first on richliu's blog.

而且我們現在單核心的 CPU core 數已經夠高了，沒有必要一定要搞 multi-socket system

像 Ampere AltraMax 已經到達 128 核心
Amazon Graviton 3 也有 64 核心

當然 Amazon 並沒有展示出他們的系統，但是他們圖上提到一件事

這是一個 3 核心的系統，當然一般都是用 CCIX 接起來，應該會是二的倍數，像是 2 顆或是 4 顆
但是 Amazon 的系統是用 Nitro Card 接起來，

Nitro Card 是一塊多功能 DPU Card ，負責資料傳輸加解密等等，看資料似乎也負責管理 edge VM 系統，還有當作 storage 系統

沒有想到 Amazon 衝的這麼快，真的就丟掉 multi-socket 架構，這個應該是整個系統都為了加速 Cloud native system 而生，軟體 CPU 跟不上的用硬體去推
也完全符合 Brendan Gregg 所說的情境

而像 Ampere 下一代的 CPU 更多核，那又更密集了，也符合 Brendan Gregg 的預測

未來 data center 可能就會走半板單 CPU 的架構吧，學 Amazon 的 triple-cores 不現實，一般真要搞 2U 4 systems 還是可以做到的，也比較相容工業標準

Ref.
AWS Graviton3 Hits GA with 3 Sockets Per Motherboard Designs
AWS Nitro the Big Cloud DPU Deployment Detailed

The post Multi-socket is doomed appeared first on richliu's blog.