By: Michael Feldman

Cisco has unveiled the UCS C480 ML M5 Rack Server, the company’s first purpose-built offering for the burgeoning AI/machine learning market.

The ML server is a 4U box powered by two Intel Xeon-SP CPUs and up to eight NVIDIA V100 GPUs.  As such, it is essentially a machine learning-optimized variant of the UCS C480 M5 Rack Server, which offers six (lesser) GPUs and is primarily aimed at analytics, VDI and database processing. Thanks to the V100 GPUs and their integrated Tensor Cores, the ML server offers a peak petaflop of deep learning performance for either training or inferencing. Those V100s are NVIDIA’s latest and greatest, sporting 32 GB of HBM2 stacked memory and NVLink 2.0.

The ML server has plenty of place to put data, with up to 3 TB of main memory, 24 SAS/SATA SSDs and hard disk drives, and 6 NVMe drives. Network connectivity is provided by 4 PCIe 3.0 slots for 100G virtual interface cards (VICs), as well as integrated dual 10-Gbps Ethernet.

It might seem like Cisco is a little late to the artificial intelligence game, but the company points out that only about four percent of CIOs say they currently have AI projects in production – that according to a 2018 Gartner report. If so, that’s good news for Cisco, which, according to IDC is the fifth biggest server-maker in the world and thus is in position to capture a decent share of the commercial AI/machine learning market.

"Over the next few years, apps powered by artificial intelligence and machine learning will become mainstream in the enterprise. While this will solve many complex business issues, it will also create new challenges for IT," said Roland Acra, SVP and GM for Cisco's Data Center Business Group. "Today's powerful addition to the Cisco UCS lineup will power AI initiatives across a wide range of industries. Our early-access customers in the financial sector are exploring ways to improve fraud detection and enhance algorithmic trading. Meanwhile in healthcare, they're interested in better insights and diagnostics, improving medical image classification, and speeding drug discovery and research.”

Cisco recognizes that much of machine learning is and will be deployed in cloud and container setups and is spurring on its partners to validate the relevant software stacks, such as Anaconda, Hadoop, and Kubeflow, for these types of environments. Cisco Services will provide AI/machine learning support for all the relevant commercial application areas, namely data analytics, deep learning, and automation.

The Kubeflow support for the ML server is being done in collaboration with Google, which has a vested interest in customers who employ the software to run on top of Kubernetes. "We're pleased to see Cisco creating hybrid cloud solutions for machine learning, and also contributing code to the Google-led open source project, Kubeflow,” said David Aronchick, Product Manager at Google Cloud. "Organizations running Kubeflow on the new UCS C480 deep learning server will benefit from consistent machine learning tools that work great both on-premises and on Google Cloud."

The UCS C480 ML M5 Rack Server will be available through the company’s channel partners, starting in the fourth quarter of 2018.