In March of this year, Cisco introduced the Cisco UCS B480 M5 blade server. Based on this system, we
launched a new resource pool in St. Petersburg and are preparing to launch a similar pool in Moscow. But literally half a year later (in September), the IT giant
released another new product - the UCS C480 ML M5, “sharpened” to work with artificial intelligence systems. We tell what is inside, and why this is all data centers.
/ Wikimedia / Tibigc / CC
What's inside
New development of the company - a modified version of the server
C480 M5 . To speed up the processing of machine learning algorithms, two slots for the CPU in that server
were replaced with connectors for the GPU. In these slots placed graphics accelerators Tesla V100-32G from NVIDIA.
')
The total number of cards reaches eight pieces. The performance of each accelerator is 47 times
greater than the performance of traditional processors for deep learning tasks. Data exchange goes through the built-in interface
NVLink - this is 200 GB / s per GPU.
Also, the C480 ML M5 has two Intel Xeon processors, each of which supports up to 28 cores, and 3 TB of RAM. In this case, up to 24 HDD or SSD can be connected to the server via SAS or SATA interfaces.
Why do you need a new server
According to Cisco, the server should simplify work with machine learning models at all stages of their development: from data collection and analysis to output processing results. At the same time, it is suitable for working with pipelined applications in multi-cloud environments. These include the Hadoop and Spark frameworks, as well as TensorFlow and PyTorch. In addition, Cisco is negotiating cooperation with the Anaconda platform, and is also working with Kubeflow with Google.
“The new ML M5 will make it easier to work with artificial intelligence systems for cloud providers and companies owning data centers,” commented Sergey Belkin, head of development at 1cloud . - Machine learning technologies can be used to evaluate all the processes occurring in the data center. Such applications are already in the Cisco ecosystem. For example, Tetration Analytics collects telemetry data from software and hardware sensors and processes them using MO methods. ”
Cisco suggests that by 2019, 70% of hacker attacks will use traffic encryption, therefore, in their opinion, the additional capacity of the new server will also play an important role in protecting the IT infrastructure of data centers. For these purposes, the IT giant also has its own solution: Encrypted Traffic Analytics. The system detects vulnerabilities in the infrastructure and detects malicious patterns even in encrypted traffic.
/ Wikimedia / Global Access Point / PD
Note that the release of C480 ML M5 will also affect the work of organizations that process large amounts of data - for example, medical and financial. It will be easier for analysts and developers in these organizations to extract useful information from Big Data. The new Cisco server will speed up the processing of large sets of information and the search for patterns (for example, when analyzing medical images).
According
to Cisco, 80% of global businesses have already implemented services based on artificial intelligence technologies or will implement such solutions by 2020. And by 2035, MO technology will help
increase the productivity of organizations by 40%.
Prospects and plans for the future
Gartner Research Director Chirag Decat (Chirag Dekate) believes that the C480 ML M5 will be in demand with the company's old and new customers. At the same time, Scott Mohr, director of partner relations,
said that Cisco will continue to improve solutions for services based on neural networks, and customer experience will play an important role in this process. So now the new product is being tested by fifty Cisco partners.
Later the system will be made available to all certified partners of the company. Despite ongoing tests, it is not yet possible to say anything about the effectiveness of the new server. Cisco specialists (unlike NetApp) decided not to provide the results of Resnet or Alexnet performance tests, since, according to them, they do not reflect the state of things when working on real machine learning projects.
Order C480 ML M5 will be available later this year.
Several materials on the topic of corporate blog 1cloud