Cisco Systems is adding to its converged infrastructure lineup a system aimed at the fast-growing and compute-intensive artificial intelligence and machine learning market.
Company officials on Sept. 10 unveiled the UCS C480 ML rack server, a system designed to accelerate deep learning workloads powered not only by two Intel “Skylake” Scalable Processors but also eight Tesla V100 GPUs from Nvidia that are connected by the chip maker’s NVLink interconnect. The system also comes with up to 24 disk drives offering up to 182TB of storage and up to six NVMe drives. Plus, it can support up to four 100 Gigabit Ethernet switches.
The tightly integrated offering will address the challenges facing enterprises as they try to drive competitive advantages from the massive amounts of data they’re generating, according to Cisco officials. AI and such subsets as machine learning and deep learning promise to accelerate the collection, processing and analysis of the data, giving business leaders deeper insights and the ability to make faster and better business decisions.
Enterprises are looking to tech vendors for help bringing these capabilities into their IT environments. Fueling the demand for the UCS 480C ML system is Cisco’s broad installed base of customers that are now turning to AI, Todd Brannon, senior director of data center marketing for Cisco, told eWEEK. Customers are looking for tools to run machine learning tasks at an industrial level, Brannon said.
AI is becoming a key technology in a rapidly changing IT world that is becoming less focused on systems and more on data. The rise of the cloud, the internet of things (IoT), greater mobility and other trends is fueling the growth in data, and businesses want to leverage that data to become more competitive and more efficient. Gartner analysts said that this year, the global business value of AI will hit $1.2 trillion, and grow to $3.9 trillion in 2022. That value will come from improved customer experience, new revenue and reduced costs, according to the analysts.
Cisco officials also noted that eight out of 10 businesses have adopted or plan to adopt AI as a customer service solution by 2020, and that by 2035, AI technologies will increase business productivity by as much as 40 percent. All of that is aided by the improvements in data collection technologies and compute power—from both CPUs and GPUs—as well as the growth of such AI frameworks as Caffe and TensorFlow.
However, there are myriad challenges for enterprises. The data is coming from multiple sources rather than only a core data center, and skill shortages in data science and IT are making it difficult to keep up with the rapidly evolving AI ecosystem, according to Cisco. In addition, businesses need to find ways to deal with the AI workloads as scale. Businesses can use a laptop to experiment with AI, but they need servers with more compute power and GPU capabilities to run large machine learning workloads and neural networks to run at an industrial level, Brannon said.
A broad array of tech vendors are developing systems aimed at making it easier for enterprises to adopt AI in their IT environments through simple installation, deployment and management. Cisco officials are making similar arguments for the UCS C480 ML, noting the highly integrated packaging of the technologies and the ability to easily fit into existing data centers. They also noted other accelerated computing offerings within the UCS family—including the C220 systems for inference workloads in regional and micro data centers and C240 M5 HyperFlex for test and development and machine learning in private clouds—as well as work with such ISV partners as Cloudera, Nvidia, Anaconda and Hortonworks. The company also is working with channel partners to help businesses adopt AI.
That’s a key differentiator between Cisco and other vendors who are looking for ways to democratize AI, according to Han Yang, senior product manager at Cisco.
“We’re not just giving you the hardware and saying, ‘Have a good time,’” Yang told eWEEK, adding that Cisco’s product portfolio addresses the entire lifecycle of the data, from when and where it’s generated to the collection and analysis. “We’ve been doing this kind of [data] collection, cleaning and correlating as part of our big data network for the past several years. So as far as dealing with data, we didn’t just wake up one day. … This is something that’s been going on for the last several years.”
It works with other UCS systems and leverages Cisco’s Intersight cloud-based management software, preventing the creation of operational silos as enterprises bring AI into their environments. The system will run Intel chips with up to 28 cores each.
The company has seen an 18-fold increase in the number of customers running big data workloads on UCS over the last four years, Yang said. Cisco is competing in a market that includes such heavyweights as Dell EMC, Hewlett Packard Enterprise, IBM and Lenovo. IDC analysts said that in the second quarter, Cisco saw server revenues grow to more than $1 billion, putting it in a three-way tie for fifth place with Chinese system makers Inspur and Huawei. Cisco’s revenue grew 22.4 percent over the same time last year.
The UCS C480 ML rack server uses Cisco’s Intersight management technology to help automate policy and operations in the compute infrastructure and leverages the company’s validated design to ensure it can be deployed at scale. It will run containerized applications and will be available through channel partners in the fourth quarter and will come with Cisco Services support around analytics, deep learning and automation.