The Channel Company

Sponsored by:

Visit AMD Visit Supermicro

Performance Intensive Computing

Capture the full potential of IT

Microsoft Azure’s More Capable Compute Instances Take Advantage of the Latest AMD EPYC™ Processors

Featured content

Microsoft Azure’s More Capable Compute Instances Take Advantage of the Latest AMD EPYC™ Processors

Azure HBv3 series virtual machines (VMs) are optimized for HPC applications, such as fluid dynamics, explicit and implicit finite element analysis, weather modeling, seismic processing, and various simulation tasks. HBv3 VMs feature up to 120 Third-Generation AMD EPYC™ 7v73X-series CPU cores with more than 450 GB of RAM.

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Azure

Increasing demands for higher-performance computing mean that the cloud-based computing needs to ratchet up its performance too. Microsoft Azure has introduced more capable compute virtual machines (VMs) that take advantage of the latest from AMD EPYC™ processors. This means that developers can easily spin up VMs that normally cost thousands of dollars if they were to purchase their physical equivalents.

 

This story's focus is on two of Azure's series: HBv3 and NVv4. In most cases, a single virtual machine is used to take advantage of all its resources. High-performance examples of Azure HBv3 series VMs are optimized for HPC applications, such as fluid dynamics, explicit and implicit finite element analysis, weather modeling, seismic processing, and various simulation tasks. HBv3 VMs feature up to 120 Third-Generation AMD EPYC™ 7v73X-series CPU cores with more than 450 GB of RAM. This series of VMs has processor clock frequencies up to 3.5GHz. All HBv3-series VMs feature 200Gb/sec HDR InfiniBand switches to enable supercomputer-scale HPC workloads. The VMs are connected and optimized to deliver the most consistent performance. Get more information about AMD EPYC™ and Microsoft Azure virtual machines.

 

A Dutch construction company, TBI, is using the Azure NVv4 to run computer-aided design and building modeling tasks on a series of virtual Windows desktops. The NVv4 VMs are only available running Windows powered by from four to 32 AMD EPYC™ vCPUs and offering a partial to full AMD Instinct™ M125 GPU with memory ranging from 2GB to 17GB. Previous generations of NV instances used Intel CPUs and NVIDIA GPUs that offer less performance.

 

TBI chose this solution because it was cheaper, easier to support and keep its software collection updated. Using virtual desktops meant that no client data was stored on any laptops, making things more secure. Also, these instances delivered equivalent performance, taking advantage of the SR-IOV technology.

 

Supermicro offers a wide range of servers that incorporate the AMD EPYC™ CPU and a number of servers optimized for applications that use GPUs. These servers range from 1U rackmount servers to high end 4U GPU optimized systems. Whether you’re using it on-prem or you’re building your own cloud, Supermicro’s Aplus servers are optimized for performance and technical computing applications and they run Azure and other systems well. Get more information about Supermicro servers with AMD’s EPYC™ CPUs.

Featured videos

Follow

Related Content

Supermicro SuperBlades®: Designed to Power Through Distributed AI/ML Training Models

Featured content

Supermicro SuperBlades®: Designed to Power Through Distributed AI/ML Training Models

Running heavy AI/ML workloads can be a challenge for any server, but the SuperBlade has extremely fast networking options, upgradability, the ability to run two AMD EPYC™ 7000-series 64-core processors and the Horovod open-source framework for scaling deep-learning training across multiple GPUs.

Learn More about this topic
  • Applications:
  • Featured Technologies:

Running the largest artificial intelligence (AI) and machine learning (ML) workloads is a job for the higher-performing systems. Such loads are often tough for even more capable machines. Supermicro’s SuperBlade combines blades using AMD EPYC™ CPUs with competing GPUs into a single rack-mounted enclosure (such as the Supermicro SBE-820H-822). That leverages an extremely fast networking architecture for these demanding applications that need to communicate with other servers to complete a task.

 

The Supermicro SuperBlade fits everything into an 8U chassis that can host up to 20 individual servers. This means a single chassis can be divided into separate training and model processing jobs. The components are key: servers can take advantage of the 200G HDR InfiniBand network switch without losing any performance. Think of this as delivering a cloud-in-a-box, providing both easier management of the cluster along with higher performance and lower latencies.

 

The Supermicro SuperBlade is also designed as a disaggregated server, meaning that components can be upgraded with newer and more efficient CPUs or memory as technology progresses. This feature significantly reduces E-waste.


The SuperBlade line supports a wide selection of various configurations, including both CPU-only and mixed CPU/GPU models, such as the SBA-4119SG, which comes with up to two AMD EPYC™ 7000-series 64-core CPUs. These components are delivered on blades that can easily slide right in. Plus, they slide out as easily when you need to replace the blades or the enclosure. The SuperBlade servers support a wide network selection as well, ranging from 10G to 200G Ethernet connections.

 

The SuperBlade employs the Horovod distributed model-training, message-passing interface to let multiple ML sessions run in parallel, maximizing performance. In a sample test of two SuperBlade nodes, the solution was able to process 3,622 GoogleNet images/second, and eight nodes were able to scale up to 13,475 GoogleNet images/second.


As you can see, Supermicro’s SuperBlade improves performance-intensive computing and boosts AI and ML use cases, enabling larger models and data workloads. The combined solution enables higher operational efficiency to automatically streamline processes, monitor for potential breakdowns, apply fixes, more efficiently facilitate the flow of accurate and actionable data and scale up training across multiple nodes.

Featured videos

Follow

Related Content

Supermicro and Qumulo Deliver High-Performance File Data Management Solution

Featured content

Supermicro and Qumulo Deliver High-Performance File Data Management Solution

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Qumulo

One of the issues that’s key to delivering higher-performing computing solutions is something that predates the PC itself: managing distributed file systems. The challenge becomes more acute when the applications involve manipulating large quantities of data. The tricky part is in how they scale to support these data collections, which might consist of video security footage, life sciences data collections and other research projects.

 

Storage systems from Qumulo integrate well into a variety of existing environments, such as those involving multiple storage protocols and file systems. The company supports a wide variety of use cases that allow for scaling up and out to handle Petabyte data quantities. Qumulo can run at both the network edge, in the data center and on various cloud environments. Their systems run on Supermicro’s all non-volatile memory express (NVMe) platform, the highest performing protocol designed for manipulating data stored on SSD drives. The servers are built on 24-core 2.8 GHz AMD EPYC™ processors.


 

Qumulo provides built-in near real-time data analytics that let IT administrators predict storage trends and better manage storage capacity so that they can proactively plan and optimize workflows.

 

The product handles seamless file and object data storage, is hardware agnostic, and supports single data namespace and burstable computing running on the three major cloud providers (AWS, Google and Azure) with nearly instant data replication. Its distributed file system is designed to handle billions of files and works equally well on both small and large file sizes.

 

Qumulo also works on storage clusters, such as those created with Supermicro AS-1114S servers, which can accommodate up to 150TB per storage node. Qumulo Shift for Amazon S3 is a feature that lets users copy data to the Amazon S3 native format for easy access to AWS services if the required services are not available in an on-prem data center. 

For more information, see the white paper on the Supermicro and Qumulo High-Performance File Data Management and Distributed Storage solution, powered by AMD EPYC™ processors.

Featured videos

Follow

Related Content

Red Hat’s OpenShift Runs More Efficiently with Supermicro’s SuperBlade® Servers

Featured content

Red Hat’s OpenShift Runs More Efficiently with Supermicro’s SuperBlade® Servers

The Supermicro SuperBlade's advantage for the Red Hat OCP environment is that it supports a higher-density infrastructure and lower-latency network configuration, along with benefits from reduced cabling, power and shared cooling features. SuperBlades feature multiple AMD EPYC™ processors using fast DDR4 3200MHz memory modules.

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Red Hat

Red Hat’s OpenShift Container Platform (OCP) provides enterprise Kubernetes-bundled devops pipelines. It automates builds and container deployments and lets developers focus on application logic while leveraging best-of-class enterprise infrastructure.

 

OpenShift supports a broad range of programming languages, web frameworks, databases, connectors to mobile devices and external back ends. OCP supports cloud-native, stateless applications and traditional applications. Because of its flexibility and utility in running advanced applications, OCP has become one of the go-to places that support high-performance computing.

 

Red Hat’s OCP comes in several deployment packages, including as a managed service running on the major cloud platforms, as virtual machines, and on “bare metal” servers, meaning a user installs all the software needed for the platform and is the sole tenant of the server.

 

It’s that last use case in which Supermicro’s SuperBlade servers are especially useful. Their advantage is that they support a higher-density infrastructure and lower-latency network configuration, along with benefits from reduced cabling, power and shared cooling features.

 

The SuperBlade comes in an 8U chassis with room to accommodate up to 20 hot-pluggable nodes (processor, network and storage) in a variety of more than a dozen models that support serial-attached SCSI, ordinary SATA drives, and GPU processor modules. It sports multiple AMD EPYC™ processors using fast DDR4 3200MHz memory modules.

A chief advantage of the SuperBlade is that it can support a variety of higher-capacity OCP workload configurations and do so within a single server chassis. This is critical because OCP requires a variety of server roles to deliver its overall functionality, and having these roles working inside of a chassis means performance  and latency benefits. For example, you could partition a SuperBlade’s 20 nodes into various OCP components such as administrative, management, storage, worker, infrastructure and load balancer nodes, all operating within a single chassis. For deeper detail about running OCP on the SuperBlade, check out this Supermicro white paper.

Featured videos

Follow

Related Content

Single-Root I/O Virtualization Delivers a Big Boost for Performance-Intensive Environments

Featured content

Single-Root I/O Virtualization Delivers a Big Boost for Performance-Intensive Environments

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Microsoft Azure, VMware, AMD, AWS

Single-root I/O virtualization (SR-IOV) is an interesting standard for performance-intensive computing because it lets a network adapter access resources across a PCIe bus, making it even higher performing. It lets data traffic be routed directly to a particular virtual machine (VM) without interrupting the flow of other traffic across the bus. It does that by bypassing the software switching layer of the virtualization stack, thereby reducing the input/output overhead and improving network performance, stability and reliability. (Get more information about SR-IOV in VMware and Microsoft contexts, for example.)

 

What this means, especially in GPU-based computing, is that each VM has its own dedicated share of the GPU and isn’t forced to compete with other VMs for its share of resources. The feature also helps isolate each VM and is the basic building block for modern VM hyperscale technologies.

 

Tests of SR-IOV have found big benefits, such as lowering processor utilization by 50% and boosting network throughput by up to 30%. This allows for more VMs per host and being able to run heavier workloads on each VM.
 

An excellent server for any virtualization platform is the Supermicro BigTwin® server. With up to 4 servers in just 2U, the Supermicro BigTwin is a versatile and powerful multi-node system that is environmentally friendly due to its shared components. Plus it can handle a wide range of workloads. Learn more about the Supermicro BigTwin model AS -2124BT-HTR.

 

Not a New Idea
 

The technology isn’t new: Scott Lowe wrote about it back in 2009 and SR-IOV was initially supported by Microsoft Windows Server 2012 and with AMD chipsets in 2016. This support has been extended with Azure NVv4 and AWS EC2 G4ad virtual machine instances, which are based on the AMD EPYC™ 7002 CPU and Radeon Pro™ GPU processor families.

The standard is supported by both VMware and Microsoft’s Hyper-V hosts and in various AMD EPYC™ CPU chipsets with MxGPU technology that is built into the actual silicon. This enables sharing a GPU’s power across multiple users or VMs but providing a similar performance level of a discrete processor.

The SR-IOV technology is a big benefit for immersive cloud-based gaming, desktop-as-a-service, machine learning models and 3D rendering applications.

Featured videos

Follow

Related Content

Build an Accelerated Data Center with AMD's Third-Gen EPYC™ CPUs

Featured content

Build an Accelerated Data Center with AMD's Third-Gen EPYC™ CPUs

“AMD EPYC™ processors are now a part of the world’s hyperscale data centers,” said Lisa Su, AMD’s CEO. Meta/Facebook is now building its servers with powerful third-generation AMD EPYC™ CPUs.

Learn More about this topic
  • Applications:
  • Featured Technologies:

If you're making plans to build a high-performance data center, be sure to take a close look at the latest version of AMD's EPYC™ CPU chipsets, which were code-named “Milan X.”

 

Servers that employ AMD’s third-generation EPYC™ CPUs are so powerful that Meta/Facebook is now building its servers with them, using the new single-socket cloud-scale design, which is a part of their Open Compute Project. “AMD EPYC™ processors are now a part of the world’s hyperscale data centers,” said Lisa Su, AMD’s CEO, in the presentation at which she debuted the processors.

 

This latest generation of AMD EPYC™ CPUs uses an innovative packaging option of 3D stacking of chiplets for high-performance computing applications. Higher density cached memory is stacked on top of the processor to deliver more than 200 times the interconnected density of prior chiplet packaging designs. “It is the most flexible active-on-active silicon technology available in the world,” Su said. “It consumes much less energy and fits into existing CPU sockets, too.” AMD's latest chipsets satisfy the higher demands of cloud computing and electronic circuit design applications.

 

Jason Zander, EVP Microsoft Azure, said that Microsoft's partnership with AMD has let the cloud computing company deliver cloud instances that can run up to 12 times the speed of earlier offerings. “That rivals some supercomputers,” he said. Azure has configured some of the most powerful virtual instances, which are running on the latest AMD EPYC™ processors. They are available from 16 cores up to 120 cores and can share 448 GB of memory and 480 MB of L3 cache among the processors. For deeper information, see this Microsoft blog.

 

Circuit design demands the fastest processors. “The next step for AMD is to deliver more differentiation in value with a focus on performance per core,” said Dan McNamara, general manager of AMD’s Server Business Unit. “In our tests comparing Synopsys VCS chip-design simulation software running on older and newer AMD EPYC™ CPUs, engineers were able to complete 66% more jobs in the same elapsed time, thanks to having a larger L3 cache. This means that more data can be kept closer to the processor for better performance.” These faster product design lifecycles mean faster times to market since designers can save time in the testing process.

Featured videos

Follow

Related Content

Fast Supermicro A+ Servers with Dual AMD EPYC™ CPUs Support Scientific Research in Hungary

Featured content

Fast Supermicro A+ Servers with Dual AMD EPYC™ CPUs Support Scientific Research in Hungary

Learn More about this topic
  • Applications:
  • Featured Technologies:

The Budapest Institute for Computer Science and Control (known as SZTAKI) conducts a wide range of scientific research spanning the fields of physics, computer science, industrial controls and intelligent systems. The work involves medical image processing, autonomous vehicles, robotics and natural language processing, all areas that place heavy demands on computing equipment and a natural use case for performance-intensive computing.


SZTAKI has been in operation since 1964 and has more than 300 full-time staff, with more than 70 of them holding science-related degrees. It works with both government and other academic institutions jointly on research projects as well as contract research and development of custom computer-based applications.

The institute also coordinates similar types of work done at Hungary’s AI national lab. For example, there are several projects underway to develop AI-based solutions to process the Hungarian language and build computational-based models that can be more effective and not require as much training as earlier models. They are also working on creating more transparent and explainable machine learning models to make them more reliable and more resilient in preserving data privacy.

SZTAKI has been using Supermicro’s A+ 4124GO-NART servers with GPUs that are configured with two AMD EPYC™ 7F72 CPUs. “Our researchers are now able to advance our use of AI and focus on more advanced research," said Andras Benczur, scientific director at the AI lab. One challenges they face is keeping up with the advanced algorithms that its researchers have developed. Having the Supermicro servers, which operate at 20x the speed of previous servers, means that researchers can execute coding and modeling decisions far more quickly.

Featured videos

Follow

Related Content

Performance-Intensive Computing Helps Lodestar Computer Vision ‘Index’ Video Data

Featured content

Performance-Intensive Computing Helps Lodestar Computer Vision ‘Index’ Video Data

Lodestar is a complete management suite for developing artificial intelligence-based computer vision models from video data. It can handle the navigation and curation of a native video stream without any preparation. Lodestar annotates and labels video, and using artificial intelligence, creates searchable, structured data.

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Lodestar.ai, Juypter.org

Lodestar doesn’t call it indexing, but the company has a product that annotates video, and using artificial intelligence (AI), creates searchable, structured data. Lodestar offers a complete management suite for developing AI-based computer vision models from video data. The company’s technology includes continuous training of its AI models along with real-time active learning and labeling.

 

The challenge for computer vision efforts before Lodestar's technology came into the picture was the sheer amount of data contained in any video stream: an hour of video contains trillions of pixels. The result was a very heavy computational load to manipulate and analyze. That meant video had to be pre-processed before anyone could analyze the stream. But thanks to performance-intensive computing, there are new ways to host more capable and responsive tools.

 

That's where Lodestar comes into play, handling the navigation and curation of a native video stream without any preparation, using the video as a single source of truth. Metadata is extracted on the fly so that each video frame can be accessed by an analyst. This is a highly CPU-intensive process, and Lodestar uses Supermicro A+ servers running Jupyter’s data science applications across a variety of containers. These servers have optimized hardware that combines AMD CPU and GPU chipsets with the appropriate amount of memory to make these applications function quickly.

 

By harnessing this power, data scientists can now collaborate in real time to validate the dataset, run experiments, train models and guide annotation. With Lodestar, data scientists and domain experts can develop a production AI in weeks instead of months.

 

That’s what a leading European optical and hearing aid retailer did to help automate its in-store inventory management processes and keep track of its eyewear collection. Before the advent of Lodestar, each store’s staff spent 10 hours a month manually counting inventory. That doesn’t sound like much until you multiply the effort by 300 stores. With Lodestar, store inventory is completed in minutes. Given that the stores frequently update their product offerings, this has brought significant savings in labor, and more accurate inventory numbers have provided a better customer experience.

Featured videos

Follow

Related Content

CERN Parses Hadron Collider Data with 900 Supermicro Computers and AMD CPUs

Featured content

CERN Parses Hadron Collider Data with 900 Supermicro Computers and AMD CPUs

CERN is trying to discover what happened in the nanoseconds following the Big Bang that created all matter. It is manipulating data flows with custom AMD circuitry that slices up the Large Hadron Collider data into smaller pieces. “You need to get all the data pieces together in a single location because only then can you do a meaningful calculation on this stuff,” said Niko Neufeld, a CERN project leader. The effort entails rapid data processing, high-bandwidth access to lots of memory and very speedy I/O among many servers.

Learn More about this topic
  • Applications:
  • Featured Technologies:
  • Featured Companies:
  • Gigabyte

Supermicro has delivered 900 of its BigTwin® A+ 2124BT-HNTR computers employing AMD EPYC™ processors to the European Organization for Nuclear Research (better known as CERN) to support the organization’s research. The systems are for running batch computing jobs related to physics-event reconstruction, data analysis and simulations. Many of CERN's discoveries have had a powerful effect on aspects of everyday life, in areas such as medicine and computing.

CERN is home to the largest physics project on earth, the Large Hadron Collider (LHC). It can collect data on subatomic particle interactions at the rate of 40TB per second. This means the lab needs high-performance computers to sift through the massive amount of data and find the most relevant interactions that will give scientists the data needed to support the right conclusions.

“It is a messy I/O challenge, and has huge data requirements,” said Niko Neufeld, a project leader for online computing at CERN. Neufeld oversees a project investigating the properties of a quark called the beauty particle. The project is attempting to determine what happened in the nanoseconds just after the Big Bang that created all matter. The data flows are manipulated with custom AMD circuitry that slices up the LHC data into smaller pieces. “You need to get all the data pieces together in a single location because only then can you do a meaningful calculation on this stuff," Neufeld said. The effort entails rapid data processing, high-bandwidth access to lots of memory and very speedy I/O among the many Supermicro servers.

There are at least three reasons that CERN gravitated to its eventual choice of servers and storage systems supplied by Supermicro and AMD. One is that CERN has been an AMD customer through many processor generations. Another is that there was support for 128 PCIe Gen 4 data paths that let networking cards run with minimal bottlenecks. The third reason was the capability of the CPUs and servers to support the 512GB RAM installed on each server, so the servers can collectively keep pace with the data driving at them at 40TB/sec.

“This generation of the AMD EPYC™ processor platform offers an architectural advantage, and there isn’t any current server that offers as much power and slots,” he said. Finally, because of the vast compute power of the Supermicro BigTwins, CERN was able to reduce its overall server count by a third, making the project more energy efficient and providing the space to add more servers should that be needed. “We could eventually double our capacity and occupy the same physical space,” he said. “It gives us a lot of headroom.”

Featured videos

Follow

Related Content

Innovations from Supermicro and AMD Help Create Visual Effects for Blur Studio

Featured content

Innovations from Supermicro and AMD Help Create Visual Effects for Blur Studio

Blur Studio calculated it could replace a competitor's 500-node server farm with just 56 Supermicro A+ servers equipped with AMD EPYC™ CPUs, getting equivalent processing power.

Learn More about this topic
  • Applications:
  • Featured Technologies:

The latest computer graphics images in movies and TV require the latest computing innovations. The scenes are getting more realistic, and that means taking advantage of Supermicro A+ computers using AMD EPYC™ 7742 CPUs with 64 cores, 129 threads and loads of DDR4 memory. “These have the necessary horsepower to render the visual effects,” said Shawn Wallbridge, the head of IT for Blur Studio.

 

Blur is a major animation and visual effects house begun by Tim Miller, the director of “Deadpool.” The studio has produced game cinematics, commercials and complex visuals such as scenes for the latest Halo Wars, League of Legion and “Terminator: Dark Fate.”

 

Animation can benefit from AMD’s advanced CPUs with higher core densities and clock speeds, supporting higher frame rates and scene interactions.

 

Blur originally used a 500-node server farm with a competitor’s CPUs. It switched to the AMD EPYC™ processors when it had to work on three very demanding films concurrently. Rendering times that previously would have taken 75 hours to complete took only 10 hours with the AMD EPYC™ CPU-powered computers. There were also significant workflow improvements because the graphic artists could see the results overnight rather than having to wait days. Blur was able to create more complex action scenes that were both frenetic and highly believable to audiences.

 

The studio calculated it could provide the equivalent processing power by replacing its 500-node server farm with just 56 Supermicro A+ servers equipped with AMD EPYC™ CPUs. Additional advantages included lower software licensing fees, reduced power consumption and lower cooling expenses.

 

“Considering the CPU marketplace right now, there is just no competition. It’s just mind blowing how fast the effects are,” said Blur's Wallbridge.

 

For more information about this story, see the AMD case study on Blur Studio.

 

Featured videos

Follow

Related Content

Pages