On the orange area, the GPU is used in a traditional virtualized environment. It is also shared across multiple applications or users. The GPU virtualization solution is widely adapted in the existing infrastructure.
On the blue area, four key components are in a composable AI solution and these are the following: physical server, GPU chassis (GPU pool), PCIe fabric, and the management software. GPUs in GPU chassis are dynamically assigned to any connected physical server (through PCIe fabric) by management GUI or API to build up a composed server. GPU virtualization and composable AI solutions are complementary to each other.
GPUs are installed in a GPU chassis. The host adapter (PCIe Gen4 x16) is installed in the PCIe slot of the physical server. A GPU chassis is connected to 2-4 physical servers via cables running the PCIe protocol.
Users can use the management software of a single unit or H3 management center to assign (un-assign) a GPU to (from) a connected server.
The composed server is composed by using a physical server and a GPU chassis. For example, if 2 GPUs in a GPU chassis are assigned to the first connected physical server, the composed server is a 2 GPU server (physical server plug 2 GPUs). These two GPUs are in a composed server work as direct-attached GPUs.
Users can use GPUs in composed servers as normal GPUs. In AI or HPC, GPUs are dedicated to one VM as bare metal GPU servers. In a virtual desktop or development environment, the GPU is shared across multiple VMs. This part is exactly the same as which GPUs are consumed in the existing infrastructure.
Through a graphic user interface, H3 Center enables GPU provisioning as well as GPU chassis discovery, inventory, port configuration, diagnostics, monitoring, fault detection, utilization auditing, and performance.
The IT administrator can provision or redeploy GPU and configure PCIe ports in just a few seconds without service interruptions.
H3 Center discovers the GPU utilization and performance then performs continuous real-time analysis to the administrator for better resource utilization.
H3 Center gives IT professionals the insight and control they need to manage and mitigate issues to anticipate failure risks. It also presents actions the administrator can take to anticipate and avoid problems.
Dynamically assign, move, and scale GPU pools with greater flexibility and efficiency to deliver optimal value. Falcon 4016 and Falcon 4010 are designed to support your AI, HPC, and big data projects. You can tailor GPU configurations to your own requirements. This on-the-fly hardware capacity helps reduce stranded assets and over provisioning while greatly improving performance and efficiency.Learn More
If you want to apply for any product display, please write a form and we will contact you after receiving the message.