core things (ID: aichip001)
Author | Xinyuan
Editor | Moying
core things reported on June 1 that this afternoon, NVIDIA launched the cloud-based development center NVIDIA Base Command Platform at the Taipei International Computer Exhibition Computex 2021, helping enterprises quickly put AI projects into production from prototypes.
In addition, NVIDIA announced that the NVIDIA certification system plan will be further expanded, and now it has included more than 50 NVIDIA certification systems for enterprise data centers. Dozens of certified new servers can run NVIDIA AI enterprise-level software.
Advantech , Altos, Yongqing Electronics, Auss , Dell Technology, Gigabyte Technologies, HPE, Lenovo , Yunda Technology, Ultramicro and other companies are increasing in the number of NVIDIA certification systems, including some mainstream x86 servers used in mainstream data centers, providing support for AI applications in industries such as medical, manufacturing, retail and financial services.
The first batch of NVIDIA certification systems will use NVIDIA BlueField-2 DPUs to improve security. Several world-renowned server manufacturers have released new systems using NVIDIA BlueField-2 DPU today.
data processor NVIDIA BlueField-2 DPU
I. NVIDIA and NetApp provide subscription services and will launch various integrated
for public cloud infrastructure. "World-class AI development requires a strong computing infrastructure, and it is crucial that every company and their customers can use and obtain these resources, so that AI technology can be used for it." said Manuvir Das, head of enterprise computing at NVIDIA.
NVIDIA DGX is a system optimized for AI computing. Since the launch of DGX in 2016, many industry companies at the forefront of AI have been using DGX. Today, DGX is fused with hundreds of millions of hours of experience from thousands of systems around the world, and its AI performance peaks is the cluster DGX SuperPOD composed of multiple DGX nodes. The first step in
to achieve DGX popularity is to make this machine with the best performance of the same type easier to obtain. To make it easy to use, NVIDIA has created a software stack called Base Command Platform.
is a cloud hosting solution based on NVIDIA accelerated computing, designed for large-scale, multi-user, multi-team AI development workflow hosted on-premises or in the cloud. It can reduce the complexity of AI workflow management, allowing data scientists and researchers to spend more time developing their AI projects, reducing the time to manage their machines, and increasing productivity.
Manuvir Das said that for years, NVIDIA has been using Base Command Platform internally to share SuperPOD among thousands of data scientists who have run more than a million jobs.
Base Command Platform is now available through a premium monthly subscription jointly provided by NVIDIA and NetApp. With Base Command Platform with NetApp solutions, it will enable customers to deploy and apply AI to their work with record-breaking NVIDIA DGX SuperPOD AI supercomputers and NetApp data management services.
This subscription model is now in the early access stage. Customers can experience the SuperPOD feature for months at a time, or a smaller portion of SuperPOD features. NVIDIA hopes to use this model to enable more customers to experience the unique features of SuperPOD.
Amazon and Google Cloud both plan to add support for Base Command Platform to their cloud GPU instance cluster, and Google Cloud will provide customers with a truly hybrid AI experience later this year. Manish Sainani, Product Management Director of Google Cloud Machine Learning Infrastructure, mentioned: "This hybrid AI product allows enterprises to run anywhere with just one writing, and flexibly connect to multiple NVIDIA A100 Tensor Core GPUs. By using on-demand accelerated computing, enterprises can speed up AI development."
Base Command Platform provides a unified view and simple interface for the entire AI development. This view facilitates easy sharing of resources through a graphical user interface and command line API, as well as a comprehensive monitoring and reporting dashboard.
includes NVIDIA NGC's AI and analytics software catalog, APIs integrated with MLOps software, Jupyter notebooks, and more, a wide range of AI and data science tools that can help researchers plan and schedule workloads faster, refine models and gain insights.
2. The first batch of NVIDIA certification systems use BlueField-2 DPU to improve security
When popularizing DGX, NVIDIA's bigger goal is to decompose DGX into smaller computing instances optimized by AI, so that system providers can combine computing instances for different business scenarios to meet different computing needs, while adding unique value-added functions to customers.
First, NVIDIA transforms the GPU board product of multiple GPUs into a tightly interconnected computing structure, namely A100.
Next, NVIDIA further decomposes the A100 into a GPU with a smaller form factor, such as the A30, which has lower power consumption and cost, while still has strong acceleration performance.
Finally, NVIDIA has productized the BlueField-2 DPU in DGX SuperPOD and is now available in various servers.
NVIDIA BlueField DPU transfers infrastructure tasks from CPU to DPU, making more server CPU cores available to run applications, thereby increasing server and data center efficiency.
This DPU is equipped with each server with a "computer before the computer" to provide independent and secure infrastructure services and securely isolate from the server application domain, and implement proxy-free application isolation, secure isolation, storage virtualization, remote management and telemetry on virtualized and bare metal servers.
Today, NVIDIA announced that it will expand the NVIDIA certification program to add to . In the second half of this year, several world-leading system manufacturers will release a number of servers that have been approved by the new NVIDIA certified system for the first time. New authentication systems such as will achieve breakthroughs in network, storage and security performance through BlueField-2 DPU (data processor).
By uninstalling the work tasks of CPU, a single BlueField-2 DPU can provide data center services that can only be implemented with up to 125 CPU cores, freeing up server CPU cycles to support a variety of business-critical applications. In the future, DPU will become an important component of each server in data center and edge computing systems.
ASUS, Dell Technologies, Gigabyte , Vanda Technology and Ultramicro have all announced that they will provide servers accelerated with NVIDIA BlueField-2 DPU.
BlueField-2 DPU has also been widely supported by software infrastructure suppliers such as Red Hat and VMware. For example, to assist developers in building applications that support BlueField-2 DPUs, Red Hat provides developers with a subscription version of Red Hat Enterprise Linux operating system for free.
This year, NVIDIA released the first version of DOCA, the SDK of BlueField. Customers and software manufacturers can easily program BlueField DPUs using the NVIDIA DOCA SDK.
NVIDIA expects DOCA to DPU, just like CUDA to GPU, for millions of developers to smoothly use generations of BlueField through a persistent and consistent SDK.
This on-chip data center architecture simplifies application development and enables front-and-back compatibility, including compatibility with the NVIDIA BlueField-3 DPUs expected to be launched in 2022, as well as all BlueField DPUs in the future.
III. NVIDIA certification will be expanded to Arm CPU server next year.
NVIDIA also announced that the NVIDIA certification program will be expanded to accelerated systems using Arm CPU-based.
As CPU and DPU accelerators undertake more AI computing workloads, the host CPU is not only regarded as a computing engine, but also as an orchestrator.
To expand the Arm ecosystem to enterprises and implement high-performance AI computing, Gigabyte Technology and Weiying Technology will provide new servers that use Arm Neoverse-based CPUs, NVIDIA Ampere architecture GPUs and/or BlueField-2 DPUs. The
server is expected to be launched next year and will be submitted to NVIDIA for certification.
NVIDIA also cooperated with Gigabyte Technology to provide an Arm HPC developer suite, providing a platform for integrated hardware and software for HPC, AI and scientific computing application development.
The platform is NVIDIA-verified to meet the stringent HPC application requirements. It includes an Arm Neoverse-based Ampere Altra processor from Ampere Computing, two A100 GPUs, two BlueField-2 DPUs, and NVIDIA HPC SDK to prepare GPU-accelerated applications for Arm.
NVIDIA certification system with NVIDIA Ampere architecture GPU is now available. Systems with NVIDIA BlueField-2 DPU and systems with Arm CPU will be available later this year and 2022, respectively.
eligible developers can apply for the NVIDIA Arm HPC Developer Kit immediately.
4. Multi-class NVIDIA certification system aims at accelerated computing
With the NVIDIA certification system, enterprises can support various complex tasks in traditional data centers and hybrid cloud .
includes running NVIDIA AI Enterprise and data analytics software on VMware vSphere to deploy enterprise-level platforms that support AI to enable batch run of AI workloads; it also includes design collaboration, advanced simulation on NVIDIA Omniverse Enterprise, and AI deployment on Red Hat OpenShift platform. The system can also seamlessly integrate with Clouderah data engineering and machine learning, reducing the time to deliver models from hours to minutes.
NVIDIA certification system has been rigorously tested to continue NVIDIA's design best practices in terms of performance, security and scalability. These systems use NVIDIA A100, A40, A30 or A10 Tensor Core GPUs as well as NVIDIA BlueField-2 DPUs or NVIDIA ConnectX-6 series network cards, with varying prices and performance levels.
In the fields of advanced AI training and cloud computing services, Dell Technology, HPE, Ningchang and Super Micro are the first to launch a certified new server based on the NVIDIA HGX accelerated computing platform. Its products provide leading AI performance through 4 or 8 NVIDIA A100 GPUs, NVIDIA NVLink GPU interconnect technology, NVIDIA InfiniBand networks, and NVIDIA AI and HPC software stacks.
Conclusion: NVIDIA is focusing on promoting the popularization of AI
From supercomputers to servers, workstations and edge devices, from computing, graphics, virtual desktops to data center infrastructure, NVIDIA has established a complete ecosystem covering different product forms.
Based on these software and hardware products of different shapes, NVIDIA is promoting the popularization of AI in more industry scenarios through a wide ecosystem of system assemblers.
