• GPU Passthrough

  ZStack Cloud supports the GPU passthrough feature, which means to passthrough a physical GPU with all attached devices (GPU graphic cards, sound cards, and others) as group to a VM instance. Therefore, the VM instance gains access to the full and powerful parallel computing capabilities of the physical GPU.

  The GPU passthrough feature applies to the GPUs of following models.

  Vendor	Model
  NVIDIA	Nvidia RTX 6000Ada, Nvidia RTX A6000 GeForce RTX 5090, Geforce RTX 4090, Nvidia RTX 3090 Quadro RTX 8000, Quadro RTX 6000 M4000, P2000 GTX 1650/1660, GTX 1060ti H100, H800, H200, H20 Nvidia L40, Nvidia L20, Nvidia L4 Nvidia A100, A30 Nvidia A40, Nvidia A16, Nvidia A10, Nvidia T4 Tesla V100, Tesla P4/6/40/100, M6/10/60 K6000 More NVIDA models, see NVIDIA Official Documentation
  AMD	Radeon v620, Radeon RX5700 FirePro S7150, FirePro S7150X2
  HYGON	DCU Z100/Z100L, DCU K100 AI
  Huawei	Atlas 300i Pro 910B/910B4 Note: You can only passthrough Huawei GPUs installed on ARM hosts.
  Iluvatar CoreX	Zhikai 100, Tiangai
  Vastai	SV100、SG100 Note: We recommend that you do not passthrough a SG100 GPU to a VM instance directly. The recommended practice is to virtualize SG100 and attach vGPUs to VM instances.
  Enflame	S60 Note: If you attach more than 5 GPU devices of this type to a VM instance, enter the VM operating system and add event_timeout=300 in the configuration file /etc/udev/udev.conf. This extend the udev timeout to ensure all GPU devices are identified properly.
  Other	MetaX N100, MOORE TREADS, Cambricon, CloudBlazer

• GPU Virtualization

  ZStack Cloud supports the GPU virtualization feature, which means to divide a physical GPU device into multiple vGPUs based on the virtualization method and form a vGPU pool. You can use these vGPUs to create lightweight vGPU VM instances, improving the VM computing capabilities as well as reducing the cost.

  The GPU virtualization feature applies to the GPUs of following models.

  Vendor	Model
  NVIDIA	Nvidia RTX 6000Ada, Nvidia RTX A6000 Quadro RTX 8000, Quadro RTX 6000 H100, H800, H200, H20 Nvidia L40, Nvidia L20, Nvidia L4 Nvidia A40, Nvidia A16, Nvidia A10, Nvidia T4 Tesla V100, Tesla P4/6/40/100, M6/10/60 More NVIDA models, see NVIDIA Official Documentation
  AMD	FirePro S7150, FirePro S7150X2
  Vastai	SV100、SG100

• Recommended GPU Driver Version

  We recommend that you install GPU drivers of following versions to hosts and VM instances to make the GPU monitoring and vGPU feature work properly.

  GPU	Recommended Version for Host	Recommended Version for VM Instance
  NVIDA	GPU driver: NVIDIA-Linux-x86_64-510.47.03-grid.run vGPU driver: NVIDIA-Linux-x86_64-510.47.03-vgpu-kvm.run	Latest version recommended by NVDIA. For more information, see NVIDIA Official Documentation.
  AMD	rocm-smi 6.1.2 and later versions.	rocm-smi 6.1.2 and later versions. Note: If the VM instance uses the OS of RHEL7 series, make sure that the VM kernel is of 4.8.10 or later version.
  Hygon	rock-5.2.0-5.16.29-V01.13.run	/
  Huawei	Ascend-hdk-310p-npu-driver_24.1.rc1_linux-aarch64.run	Ascend-hdk-310p-npu-driver_24.1.rc1_linux-aarch64.run
  Iluvatar CoreX	x86: corex-installer-linux64-4.0.1_x86_64_10.2.run ARM: corex-installer-linux64-4.0.1_arm64_10.2.run	x86: corex-installer-linux64-4.0.1_x86_64_10.2.run ARM: corex-installer-linux64-4.0.1_arm64_10.2.run
  Enflame	nflame-x86_64-gcc-1.4.3.4-20250520100021.run or later	nflame-x86_64-gcc-1.4.3.4-20250520100021.run or later

• 3D Rendering

  Pre-rendering, a technology often used in filmmaking, demands high computational intensity which needs to be provided by a large amount of servers. Real-time rendering, a technology commonly used in 3D video games, usually relies on GPUs to execute the rendering process.

  Nowadays, with the rapid development of the GPU technology, a significant amount of 3D rendering is accomplished by using GPU server clusters. With the GPU passthrough feature provided by ZStack Cloud, you can make VM instances process these 3D rendering tasks with a low GPU performance loss (within 5%). Combined with intelligent monitoring software and the billing feature of ZStack Cloud, this feature provides you with a comprehensive, convenient, and efficient rendering farm solution.

  
  
  

• Artificial Intelligent

  GPU's powerful computations supports deep learning tasks. Since Google released its network tool, TensorFlow, GPUs have become infrastructures preferred by research organizations and companies. Take the NVIDIA P100 GPU as an example. After be passthroughed to a VM instance, the performance of the GPU is nearly the same with that of a host GPU and can perfectly meet the requirements of extensive model trainings.

  
  
  

• Cloud Gaming

  With the expansion of broadband networks and the widespread adoption of mobile devices, a new trend in gaming has emerged where the computational load is shifted to the cloud. In this model, the client device merely handles display and input, while the heavy lifting of 3D game rendering is performed on powerful cloud servers. Combined with GPUs, these cloud servers can encode each frame instantly and stream it to any devices connected to the network. This model can be realized by using the GPU passthrough or vGPU feature provides by ZStack Cloud, ensuring a more fluent game experience for users.

  
  
  

• Virtual Desktop Infrastructure
  GPUs have long been crucial components in VDI environments. They enhance visual experiences and take on primary computational roles in specialized applications, effectively replacing traditional PC workstations. This enables users to perform 3D design tasks in a more secure setting. Through the GPU passthrough or vGPU feature provided by ZStack Cloud, coupled with protocols such as RDP (Remote Desktop Protocol) or PCoIP (PC over IP), you can harness the full potential of the GPU devices, providing users with a VDI using experience as fluent as that in a physical environment.

  
  
  

{
    "inventories": [
        {
            "chooser": "None",
            "createDate": "Jul 15, 2024 3:54:32 PM",
            "description": "3D controller: NVIDIA TU102GL [Quadro RTX 6000/8000]",
            "device": "TU102GL [Quadro RTX 6000/8000] [1e78]",
            "deviceId": "1e78",
            "hostUuid": "363c8eab666547e29a7a74bbde9c1967",
            "iommuGroup": "363c8eab666547e29a7a74bbde9c1967_/sys/kernel/iommu_groups/39",
            "isDriverLoaded": true,
            "lastOpDate": "Jul 17, 2024 4:03:05 PM",
            "mdevSpecRefs": [],
            "memory": 48318382080,
            "metaData": {},
            "name": "NVIDIA_TU102GL [Quadro RTX 6000/8000]_4885da63",
            "pciDeviceAddress": "0000:17:00.0",
            "pciSpecUuid": "016a96fd5aa648b290d2e96bf49f7407",
            "power": 250,
            "serialNumber": "1325119076420",
            "state": "Enabled",
            "status": "System",
            "subdeviceId": "13d8",
            "subvendorId": "10de",
            "type": "GPU_3D_Controller",
            "uuid": "4885da636aba4e12afcc5901e12673a3",
            "vendor": "NVIDIA",
            "vendorId": "10de",
            "virtStatus": "SRIOV_VIRTUALIZABLE"
        }
    ],
    "success": true
}