Overview

ZStack Cloud allows you to attach a PCI device to a VM instance. For the current version, ZStack Cloud supports two methods, including GPU pass-through and virtual GPU (vGPU), both of which can be used to attach VM instances.

• ZStack Cloud supports the physical GPU passthrough feature. Specifically, physical GPUs can pass through entirely all its peripheral devices (GPU graphics cards, GPU sound cards, and other small devices) as a group to VM instances, thus allowing the VM instances to obtain the powerful GPU parallel computing of hosts. This feature is applied to high performance computing (HPC) scenarios in fields of 3D rendering, high definition decoder, and highly intensive computing, such as machine learning, medical imaging, data analysis for petroleum exploitations, and Bitcoin mining.

  ZStack Cloud supports the following physical GPU specifications.

  NVIDIA	AMD
  Tesla T4 Tesla M10/M60 Tesla P100/P40/P6/P4 Tesla V100 RTX 6000/8000 ...	FirePro S7150 FirePro S7150X2

  Note: ZStack Cloud supports multiple passthrough GPU specifications. Here, we only display a common set of the GPU specifications. For more information about GPU specifications, see NVIDIV Virtual GPU Software Documentation.

• ZStack Cloud supports the vGPU feature. Through the GPU virtualization technology, physical GPUs will be segmented to more fine-grained vGPUs, which will become vGPU resource pools. You can create lightweight vGPU VM instances by using vGPU specifications to achieve more flexible, scalable resource deployments, higher resource utilization, and lower cost. This feature is applied to the lightweight GPU computing scenarios in fields of cloud gaming, VDI, VR/AR, AI inference, and machine learning pedagogy.

  ZStack Cloud supports the following vGPU (segmented by physical GPU virtualization technology) specifications.

  NVIDIA	AMD
  Tesla T4 Tesla M10/M60 Tesla P100/P40/P6/P4 Tesla V100 RTX 6000/8000 ...	FirePro S7150 FirePro S7150X2

  Note: NVIDIA GPU supports multiple virtualization segmentation models. We only display a common set of GPU specifications. For more information about GPU models, see NVIDIV Virtual GPU Software Documentation.