Cisco UCS delivers industry leading GPU density on blade servers
GPUs are becoming ubiquitous in the data center.
On one hand, if you’re running virtual desktops you know that modern apps and newer operating systems such as Windows 10 make use of a lot more graphic power than their predecessors. Which means that you need to provide more graphic power when you deliver these apps over a virtual desktop. This is why most modern VDI architectures include virtualized GPUs.
On the other hand, artificial intelligence, machine learning and deep learning projects are on every enterprise’s radar. Most AI or ML/DL applications require powerful GPUs to accomplish their complex calculations in a reasonable amount of time.
What this all means, is that your data center has to evolve in order to keep up with these new requirements.
Figure 1: The Cisco GPU-Accelerated Data Center
Cisco recently introduced its 5th generation Unified Computing System (UCS) servers to address evolution in the data center. With this recent release Cisco delivers the highest GPU density in the industry, ready to power workstation and application virtualization with even greater efficiency and end-user satisfaction. At the same time, the versatility of Cisco UCS servers also allow them to run data-intensive and compute-intensive workloads such as real-time analytics and in-memory computing.
Today NVIDIA announced a new generation of virtual GPU software, the Quadro Virtual Data Center Workstation (Quadro vDWS and GRID Virtual PC – GRID vPC) and its latest NVIDIA Tesla Pascal GPU, the P6 for blade servers. The Quadro vDWS turns Tesla GPU-powered Cisco UCS Servers into powerful workstations and allows organizations to virtualize both graphics and compute workloads. The new GRID vPC improves density and manageability to help optimize TCO for the digital workspace. This new generation of NVIDIA virtual GPU solutions provide an even better virtual desktop and workstation experience, which combined with Cisco UCS is cost effective to purchase, deploy and operate.
Cisco supports the new NVIDIA Tesla P6 GPU on Cisco UCS B200 M5 and B480 M5 blade servers.
Figure 2: Summary of NVIDIA GPU support on Cisco UCS M5
Blade servers – unmatched GPU density
The Cisco UCS B200 M5 supports up to two NVIDIA Tesla P6 mezzanine adapters. This supports up to double the user density for remote knowledge workers, task workers, and designers doing 3D design modeling and offers the highest density in the industry for a two-socket blade server.
The Cisco UCS B480 M5 supports up to four NVIDIA Tesla P6 mezzanine adapters which is unmatched in the industry for a four-socket blade server.
Rack servers – introducing support for the NVIDIA Tesla P40
The Cisco UCS C480 M5 supports up to six PCIe GPU adapters. We are introducing support for the NVIDIA Tesla P40. The Tesla P40 is for remote engineering workstations and application delivery via the data center. Additional support for the NVIDIA Tesla M10 for VDI and NVIDIA Tesla P100 for deep learning and HPC applications give the C480 M5 a robust GPU portfolio for your data intensive and VDI needs.
The Cisco UCS C240 M5 supports up to two PCIe GPU adapters and supports the Tesla P100, M10 and P40 GPUs. It is very well suited for VDI deployments.
If you need more GPUs but not more processor memory to justify purchasing another rack server, the Magma ExpressBox PCIe expander chassis allows the C240 M5 to support up to nine GPUs. The Magma ExpressBox will also be available for the C480 M5 at a later date.
Maximizing your investments
With the ability to switch profile to run both virtualized, accelerated graphics and compute (CUDA and OpenCL) workloads on the same Pascal GPUs, enterprises can maximize their investments. You could for example run virtual desktops in the day while most of your users need access to their apps and switch to CUDA mode at night to run your HPC projects.
We will be at VMworld Aug 27-31 in Las Vegas. Stop by the Cisco booth and talk to our engineers about deploying virtual desktops or workstations on VMware Horizon in a Cisco GPU-accelerated data center.