Nvidia, OEM Partners Show How GPUs Are Used in Data Center

by Jeffrey Burt

Huang, on stage during his keynote address at the GTC, talked about the Digits DevBox, which is powered by four Titan X GPUs. (Image courtesy of Nvidia)

Nvidia’s Titan X, which is available now for $999 and is based on the Maxwell architecture, comes with 8 billion transistors, 3,072 CUDA cores, up to 12GB of memory and 7 teraflops of peak single-precision performance. (Image courtesy of Nvidia)

Nvidia had a DevBox in the exhibit hall for GTC attendees to take a look at. The appliance will start shipping in May at a cost of $15,000.

At the GTC, Dell officials announced the Precision Appliance for Wyse, a virtual desktop infrastructure (VDI) solution that leverages Nvidia’s Grid vGPU technology and is aimed at graphics-intensive workstation workloads. The appliance at the show is housed in this rack.

The Precision Appliance for Wyse enables applications such as this one to run in a virtualized environment.

Dell also uses Nvidia GPU accelerators in its servers, including the PowerEdge C4130 rack server optimized for GPUs (front) and the PowerEdge R730 system.

HP leverages Nvidia’s accelerators in a range of systems, including the Apollo 6000 HPC system powered by Intel processors.

HP’s ProLiant WS460c Gen 9 Graphics Server Blade enables customers to deliver desktop virtualization capabilities with workstation-class graphics.

Lenovo’s NextScale portfolio of systems, inherited last year when the company bought IBM’s x86 server business, sports Nvidia GPU accelerators.

Cisco is partnering with chassis maker Magma to enable customers to scale their GPU use by filing a Magma chassis (the black box on left) with Nvidia GPUs and connecting them to Cisco Unified Computing System (UCS) converged solutions, like the C240 systems, underneath and to the right of the chassis.

SuperMicro’s 1U (1.75-inch) SuperServer supports up to two Nvidia GPUs. It also can support Intel’s Xeon Phi x86 co-processors.

Quanta brought its Quanta Cloud Technology (QTC) solution to the GTC, which leverages Nvidia’s Grid vGPU technology. The QTC offering is currently in preview.

The day before the GTC kicked off, Cavium officials announced that the company’s ARM-based 64-bit ThunderX server system-on-a-chip (SoC) will support Nvidia’s Tesla GPU accelerators.

The two companies last year announced that VMware’s Horizon 6 solution would support Nvidia’s Grid vGPU technology, and it became available March 17. During the GTC show, the vendors competed in a friendly contest to see which could spin up 60 virtual desktops in 60 minutes first.