Hewlett-Packard has announced a new blade-server/thin-client setup designed to run virtual desktops more quickly than its earlier models, and let users run several applications across multiple screens.
The new products represent HP’s first that incorporate the Nvidia GRID graphics processing units (GPU), designed to add enough graphics power to a blade or server to either lower costs by running many more sessions per blade server, or give end users far richer graphics on each virtual machine.
The client hardware is the HP t820 Flexible Series Thin Client running Intel Core i5 microprocessors; it runs eight times faster than the earlier HP T610 Flexible thin client, and is designed to control large amounts of server-based data or applications simultaneously. It can run either 32- or 64-bit Windows Embedded Standard thin-client OSes , either Intel Graphics 4600 or AMD Radeon HD 750 MXM graphics cards and include DVI-D, HDMI and VGA adapters, all three of which can be active on machines with the AMD graphics card installed.
The server side is what drives the graphics, however.
HP’s HP WS460c Gen8 Graphics Server Blade is the first to include Nvidia’s virtual-desktop-friendly GRID GPUs, which allow each blade server to support heavy graphics for several users at a time. The higher-end GRID K1 and K2 adapters are designed to give thin-client users rich-media or high-performance graphics, and are also able to support several virtual clients on each blade running resource-intensive graphics.
The servers will be available later this year for list prices starting under $9,000, according to HP, which will show off the thin client at VMworld in San Francisco next week.
Nvidia’s GRID is a GPU combined with GRID VGX software (PDF)—a stack of software libraries designed for high-performance graphics accessible by thin clients in virtual-desktop infrastructures.
The processors are constructed on Nvidia’s Kepler virtual-graphics architecture, which was purpose-built to allow several users to share one GPU with high graphics performance, according to Nvidia. The GPUs come in PCIe cards with up to 16 GB of memory.
The lower-end GRID K1 chipset comes with up to 4 entry level Kepler GPUs and 768 CUDA cores; the higher-end K2 version comes with 2 GPUs and 3072 CUDA cores. Nvidia announced the GRID chips in March along with commitments from Dell, HP and IBM to build servers using them.
Dell, IBM and HP have all toyed with the idea of replacing desktop machines with thin-client versions, but have come up short partly due to the traditionally weak graphics capability of VDI and the cost of dedicating entire VMs to a single user in order to run a desktop with enough graphics capability to support more than one monitor or high-demand graphics.
Nvidia GRID comes in three pieces: the GPUs, the GRID VGX software and the GRID Visual Computing Appliance, an add-on that provides more power and graphics capability for the GPUs it supports than if it were configured as an ordinary plug-in PCIe-based graphics card.