turbocharge your applications€¦ · turbocharge your applications nvidia tesla gpus in dell emc...

18
Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: CHAPTER 4: Deep Learning / Neural Networks CHAPTER 2: Dell EMC and NVIDIA CHAPTER 6: Data Analytics NEXT STEPS CHAPTER 1: Benefits of GPU-Accelerated Workloads CHAPTER 5: High-Performance Computing CHAPTER 3: Machine Learning CHAPTER 7: Virtual Desktop Infrastructure

Upload: others

Post on 26-Jun-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1

Turbocharge Your ApplicationsNVIDIA Tesla GPUs in Dell EMC PowerEdge servers

INTRODUCTION:

CHAPTER 4:

Deep Learning / Neural Networks

CHAPTER 2:

Dell EMC and NVIDIA

CHAPTER 6:

Data Analytics

NEXT STEPS

CHAPTER 1:

Benefits of GPU-Accelerated Workloads

CHAPTER 5:

High-Performance Computing

CHAPTER 3:

Machine Learning

CHAPTER 7:

Virtual Desktop Infrastructure

Page 2: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

2

IntroductionFor three decades, the microprocessor industry reflected Moore’s Law, with the number of transistors on a chip doubling about every two years. This rate brought tremendous growth, but it couldn’t last forever. The glory days of 50% growth per year are behind us. Today, CPU performance advances at about an average of 10% each year.

Fortunately, another game-changing technology is positioned to kickstart the industry. GPU computing has given the industry another huge push forward. NVIDIA®, for example, is on track to deliver a 1,000X performance increase over CPUs alone by 2025.1 It’s an exciting time, and the potential benefits are mind-boggling.

A GPU is a graphics processing unit designed to accelerate compute performance. Early GPUs were primarily used to quickly render high-resolution images and videos for PCs and game consoles. Because they can perform parallel processing, GPUs are now commonly adopted for many uses, including MRI scans, seismic processing, data analysis and visualization, and modeling and simulation. Today, GPUs play a leading role in high-performance computing (HPC), artificial intelligence (AI), and virtual desktop infrastructure (VDI).

High-Performance Computing (HPC)

Artificial Intelligence (AI)

Virtual Desktop Infrastructure (VDI)

= 100x increase

Page 3: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

3

They also continue to evolve to facilitate machine learning (ML) and deep learning (DL), both for training and inference. GPUs today can accelerate more than 600 applications with additional ones coming available monthly. GPUs are used inside everything from desktops to supercomputers, including seven of the TOP500’s ten fastest supercomputers.

When it comes to server design, Dell EMC’s strong partnership with NVIDIA provides a significant advantage. The Dell EMC PowerEdge portfolio includes servers specifically designed for optimal GPU performance, providing ultimate performance for accelerator- optimized workloads.

This eBook provides an overview of GPU-accelerated computing, then dives into specific accelerator-optimized workloads. We’ll discuss the most important requirements and use cases for each workload and provide recommendations for specific Dell EMC PowerEdge servers, highlighting these options in a simple “good, better, best” format. Armed with this information, you’ll be able to customize your data center to meet your objectives—today and into the future.

Page 4: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

4

Benefits of GPU-Accelerated Workloads

CHAPTER 1:

The main difference between GPUs and CPUs is in how they process tasks. However, GPUs offload tasks from the CPU—they do not replace them. GPUs have a massively parallel architecture consisting of thousands of smaller cores designed for handling multiple tasks simultaneously. As a result, GPUs can deliver dramatic performance improvements—anywhere from one teraFLOP (TFLOP) to over 100. Even the most sophisticated CPUs available offer only one teraFLOP, and most offer much less.

The Difference between a CPU and GPU

Additionally, GPUs have a different architecture and processing paradigm than CPUs. This allows for more than 50% growth in performance each year, which gives GPUs an advantage in keeping up with today’s big data demands. The growth of GPU speed is exceeding the rate of data growth, which is why GPUs hold promise to deliver next- generation platforms for accelerating analytics at scale.

GPU-accelerated computing is the use of a GPU together with a CPU, in order to accelerate processing-intensive workloads such as AI, ML, DL, business analytics, and engineering applications. It enables a divide-and-conquer approach that offloads compute-intensive portions of an application to the GPU while the remainder of the code runs on the CPU, as well as freeing the CPU to handle disk and network I/O.

GPUCPU

Page 5: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

5

GPU-accelerated computing deployment is growing in popularity due to the large variety of potential applications, including AI, robots, drones, and self-driving cars. With GPU- accelerated servers, you can easily handle large and extremely fast data streams from sources such as the Internet of Things (IoT), clickstreams, and business transactions. They can run repetitive and complex queries in milliseconds. Traditional analytical solutions just can’t keep up.

IT managers all face the same challenge: how to meet the demand for computing resources that often exceed available cycles in the system. GPUs can dramatically boost the throughput of your data center with fewer nodes, completing more jobs and improving data center efficiency.

The benefits of GPU-acceleration include:

Thousands of GPU processing cores and high-bandwidth memory can fit on a single card.

A light IT footprint:

GPUs are twice as efficient as CPUs.Innovation velocity:

GPUs enable real-time data processing.

Analytic efficiency:

-10% of code

Compute intensive operations

GPU

Sequential CPU code

Multi-core CPU

Application Code

GPU Acceleration

Page 6: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

6

Dell EMC and NVIDIACHAPTER 2:

One of the biggest challenges facing businesses today is determining how to deal with massive amounts of data. Data is often considered an organization’s most important asset, but handling and processing that data is not easy. Today’s data-intensive work-loads require lightning-fast calculations and processing, and they need the fastest hardware including GPU power to get the job done.

NVIDIA launched the modern-day GPU in 1999, which was used in the Microsoft Xbox console released in 2001. Over the next ten years, NVIDIA GPU technology expanded into broader computing purposes across all industries. Ongoing performance innovations have now brought NVIDIA GPUs into the forefront of high-performance computing and beyond.

Dell EMC works closely with NVIDIA to develop solutions for today’s accelerator- optimized workloads. Based on your organization’s workloads and specific needs, Dell EMC PowerEdge servers can help you reach your business objectives.

NVIDIA Tesla® V100 is the most advanced data center GPU ever built, offering the performance of up to 100 CPUs in a single GPU. Equipped with 640 Tensor Cores, each Tesla V100 delivers up to 125 teraFLOPS of deep learning performance. Dell EMC PowerEdge servers, including the R740xd, R7425, and R940xa, can be boosted with up to four NVIDIA Tesla V100 GPU accelerators. The PowerEdge C4140 delivers that acceleration interconnected with NVIDIA NVLink™ fabric for up to 300GB/s to unleash extreme application performance on a single server.

Dell EMC PowerEdge + NVIDIA Tesla V100

Page 7: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

77

The new NVIDIA T4 Tensor Core GPU is an accelerated computing platform for diverse workloads, from the data center to the edge, across training, inferencing, virtualization, and machine learning. The small-form-factor, 70-watt design makes the NVIDIA T4 ideal for enterprise mainstream servers, providing 240x more energy efficiency than CPUs. NVIDIA Turing Tensor Core technology delivers up to 16x higher performance than CPUs on training of neural networks, and up to 59x higher performance on inference. The NVIDIA T4 is now available in Dell EMC PowerEdge R740, R740xd, T640, and R7425 servers.2

Dell EMC PowerEdge + NVIDIA T4

Page 8: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

8

Machine LearningCHAPTER 3:

GPU acceleration can speed the processing of large data sets in machine learning environments such as Kinetica, MapD, BlazingDB, Brytlyt, OmniSci, and SQream, analyzing massive amounts of information and returning results in milliseconds. When it comes to GPU acceleration for ML, memory and storage are top priorities, and it’s important to have a well-matched CPU-to-GPU ratio. Based on these requirements, Dell EMC recommends the following PowerEdge servers:

PowerEdge R7425 is a 2U, AMD two-socket rack server that delivers outstanding TCO for data analytics, hybrid cloud, and scale-up software-defined deployments. The system can be easily expanded with extreme memory and storage capacity for low latency, data-intensive workloads. With up to 64 cores, 128 PCIe lanes, and up to 32 DIMMs, the R7425 provides the right balance of compute to I/O that gives you the freedom to take on challenging, data-intensive projects.

Good PowerEdge R7425

The PowerEdge R740 server has the perfect balance of accelerator cards, storage, and compute resources in a 2U, two-socket platform. The R740 offers up to 16x 2.5” drives or 8x 3.5” drives and iDRAC9, so you can scale to meet demands and simplify the entire IT lifecycle.

Better PowerEdge R740

Page 9: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

9

The PowerEdge R940xa accelerates applications to deliver real-time decisions. It combines four CPUs with four GPUs in a powerful 1:1 ratio to drive database acceleration. With up to 6TB of memory and four-socket performance, the R940xa delivers consistent and fast response times.

Best PowerEdge R940xa

The DSS 8440 is a new 4U, two-socket server optimized for expansion. With a wall of fans at the front of the chassis, it supports up to 10 GPUs. In the center, it includes the latest 2nd Gen Intel® Xeon® Scalable Processors with Intel Optane™ DCPMM support. At the rear of the chassis is an array of ten drives, either NVMe or SAS/SATA. And there are four 2.4kW power supplies as well as a massive array of PCIe slots for fabric and other devices.

The Dell EMC DSS 8440: Purpose-Built for Machine Learning

Read More See the Video

Page 10: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1010

Deep Learning / Neural NetworksCHAPTER 4:

Developing AI applications starts with the training of deep neural networks with large datasets. GPU-accelerated deep learning platforms like BrightML, Bitfusion, and NVIDIA NGC offer the flexibility to design and train custom deep neural networks and provide interfaces to commonly-used programming languages.

Top priorities to consider with deep learning/neural network workloads include GPU density, high network bandwidth I/O (IB/OPA/100Gb or above), and peer-to-peer GPU connectivity. Based on these requirements, Dell EMC recommends the following PowerEdge servers:

The PowerEdge T640 is a versatile, powerhouse two-socket server ideal for mid-sized offices, remote sites, and data centers. The T640 combines powerful performance and massive internal storage capacity in a rack or tower platform. Winner of the 2018 IT Pro Best Tower Server award, the T640 delivers fast insights with up to 8 NVMe drives and 2x 10GbE connections.3

Good PowerEdge T640

Page 11: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

11

Give challenging cognitive computing workloads what they need with this high-density, accelerator-optimized server. The PowerEdge C4140 provides maximum density and outstanding thermal efficiency for scale-out performance. Speed up your applications with four NVLink-interconnected GPUs in a 1U, two-socket form factor, and harness the full potential of complex datasets.

Best PowerEdge C4140

PowerEdge R7425 is a 2U, AMD two-socket rack server that delivers outstanding TCO for data analytics, hybrid cloud, and scale-up software-defined deployments. Easily add extreme memory and storage capacity for low latency, data-intensive workloads. With up to 64 cores, 128 PCIe lanes, and up to 32 DIMMs, the R7425 provides the balance of compute to I/O that gives you the freedom to take on challenging, data-intensive projects.

The PowerEdge R740 server has the perfect balance of accelerator cards, storage, and compute resources in a 2U, Intel two-socket platform. The R740 offers up to 16x 2.5” drives or 8x 3.5” drives and iDRAC9, so you can scale to meet demands and simplify the entire IT lifecycle.

Better PowerEdge R7425 or R740

“Increasingly, deep learning is a strategic imperative for every major technology company, permeating every aspect of work. Specifically, artificial intelligence is being driven by leaps in GPU computing power that defy the slowdown in Moore’s law. The work we are doing to advance GPU computing alongside Dell EMC will empower AI developers as they race to build new frameworks to tackle some of the greatest challenges of our time.”

–Ian Buck, NVIDIA4

Page 12: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1212

High-Performance ComputingCHAPTER 5:

GPU computing is the most accessible and energy-efficient path to high-performance computing (HPC) in the data center. The NVIDIA CUDA® programming model for general computing on GPUs offers a language-based solution for programmers who want to fine-tune their applications for the best possible performance. CUDA supports more than 600 GPU-accelerated applications, including the top 15 HPC applications.

When it comes to HPC, be sure to consider high GPU density, high network bandwidth I/O (IB/OPA/100Gb or above), and peer-to-peer transfers between GPUs. Based on these requirements, Dell EMC recommends the following PowerEdge servers:

PowerEdge R7425 is a 2U, AMD two-socket rack server that delivers outstanding TCO for data analytics, hybrid cloud, and scale-up software-defined deployments. Easily add extreme memory and storage capacity for low latency, data-intensive workloads. With up to 64 cores, 128 PCIe lanes, and up to 32 DIMMs, the R7425 provides the balance of compute to I/O that gives you the freedom to take on challenging, data-intensive projects.

Good PowerEdge R7425

Page 13: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1313

With the convergence of AI and HPC, the PowerEdge R940xa accelerates applications to power real-time decisions. It combines four CPUs with four GPUs in a powerful 1:1 ratio to drive database acceleration. With up to 6TB of memory and four-socket performance, the R940xa delivers consistent and fast response times.

Better PowerEdge R940xa

Give challenging cognitive computing workloads what they need with this high-density, accelerator-optimized server. The PowerEdge C4140 provides maximum density and outstanding thermal efficiency for scale-out performance. Speed up your applications with four interconnected GPUs in a 1U, two-socket form factor. The PowerEdge C4140 delivers that acceleration with NVIDIA NVLink™ fabric for up to 300GB/s.

Best PowerEdge C4140

Page 14: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1414

Data AnalyticsCHAPTER 6:

Data analytics technologies help companies examine data, draw conclusions, and make decisions. Data analytics workloads have a wide variety of business uses across industries—for example, fraud detection in financial services, biomedical analysis in healthcare, customer insights in retail, and demand forecasting in manufacturing. Analytics may involve structured or unstructured data—or both.

The top considerations for data analytics include memory bandwidth, high I/O storage (NVMe/SSDs), and CPU sizing for an accelerated solution. Based on these requirements, Dell EMC recommends the following PowerEdge servers:

The PowerEdge T640 is a versatile, powerhouse two-socket server ideal for mid-sized offices, remote sites, and data centers. The T640 combines powerful performance and massive internal storage capacity in a rack or tower platform. Winner of the 2018 IT Pro Best Tower Server award, the T640 delivers fast insights with up to 8 NVMe drives and 2x 10GbE connections.3

Good PowerEdge T640

Page 15: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1515

The PowerEdge R940xa accelerates applications to deliver real-time decisions. It combines four CPUs with four GPUs in a powerful 1:1 ratio to drive database acceleration. With up to 6TB of memory and four-socket performance, the R940xa delivers consistent and fast response times.

Best PowerEdge R940xa

PowerEdge R7425 is a 2U, AMD two-socket rack server that delivers outstanding TCO for data analytics, hybrid cloud, and scale-up software-defined deployments. Easily add extreme memory and storage capacity for low latency, data-intensive workloads. With up to 64 cores, 128 PCIe lanes, and up to 32 DIMMs, the R7425 provides the balance of compute to I/O that gives you the freedom to take on challenging, data-intensive projects.

The PowerEdge R740 server has the perfect balance of accelerator cards, storage, and compute resources in a 2U, Intel two-socket platform. The R740 offers up to 16x 2.5” drives or 8x 3.5” drives and iDRAC9, so you can scale to meet demands and simplify the entire IT lifecycle.

Better PowerEdge R7425 or R740

Page 16: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

16

Virtual Desktop InfrastructureCHAPTER 7:

With virtual desktop infrastructure (VDI), a user desktop runs inside a virtual machine that lives on a server in the data center. VDI gives end-users the mobility and freedom to access virtual desktops anytime, anywhere, from any device. VDI can also help organizations streamline their processes and lower costs by consolidating and centralizing management. NVIDIA has the industry’s highest user-density solution with support for up to 32 virtual desktops per GPU.

When it comes to VDI, the most important factors are CPU-to-GPU ratio, storage capacity, and memory. Based on these requirements, Dell EMC recommends the following PowerEdge servers:

The PowerEdge T640 is a versatile, powerhouse two-socket server ideal for mid-sized offices, remote sites, and data centers. The T640 combines powerful performance and massive internal storage capacity in a rack or tower platform. Winner of the 2018 IT Pro Best Tower Server award, the T640 delivers fast insights with up to 8 NVMe drives and 2x 10GbE connections.3

Good PowerEdge T640

PowerEdge R7425 is a 2U, AMD two-socket rack server that delivers outstanding TCO for data analytics, hybrid cloud, and scale-up software-defined deployments. Easily add extreme memory and storage capacity for low latency, data-intensive workloads. With up to 64 cores, 128 PCIe lanes, and up to 32 DIMMs, the R7425 provides the balance of compute to I/O that gives you the freedom to take on challenging, data-intensive projects.

Better PowerEdge R7425

Page 17: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

1717

Dell EMC VDI Complete offers a superior solution stack for simple end-to-end desktop and application virtualization at an exceptional total cost of ownership. Dell EMC VDI Complete combines the power of industry-leading Dell EMC VxRail appliances and vSAN Ready Nodes with VMware Horizon virtualization software. With Optional NVIDIA GPU technology and purpose-built end-user devices, Dell EMC VDI Complete provides everything you need for a powerful VDI environment in one fully validated bundle.

Dell EMC VDI Complete: Virtual Desktop Infrastructure, Evolved

Learn More Watch the Video

Dell EMC VxRail™ Hyperconverged Infrastructure appliances, the standard in hyperconverged infrastructure (HCI) jointly developed by Dell EMC and VMware, are ideal for a range of VDI environments, including those that start small and grow, or require GPU acceleration. HCI solutions benefit VDI customers by providing predictable scaling for organization growth and simplifying infrastructure management via VxRail Manager.

Using NVIDIA GRID Quadro Virtual Data Center Workstation licenses in the VxRail V Series with up to 6 NVIDIA Tesla T4 GPUs, you can create the creative designer workstation experience to meet your team’s needs. Powered by Dell EMC PowerEdge servers, VxRail appliances simplify deployment of your software-defined data center, enhance the VDI user experience, and reduce the cost of managing hardware.

Best Dell EMC VxRail Hyperconverged Infrastructure

Page 18: Turbocharge Your Applications€¦ · Turbocharge Your Applications NVIDIA Tesla GPUs in Dell EMC PowerEdge servers INTRODUCTION: ... This eBook provides an overview of GPU-accelerated

18

Ready to learn more? Take a test drive in one of our worldwide Dell EMC Customer Solution Centers, or visit Dell EMC to learn more about accelerator technology.

Explore the Customer Solution Centers »

Next Steps

Visit Dell EMC

“Supercharged computing,” NVIDIA. https://www.nvidia.com/en-us/about-nvidia/ai-computing

Kharya, Paresh, “Intel Highlighted Why NVIDIA Tensor Core GPUs Are Great for Inference,” NVIDIA, May 21, 2019. https://blogs.nvidia.com/blog/2019/05/21/intel-inference-nvidia-gpus/

“The IT Pro Product of the Year Awards 2018,” IT Pro, December 27, 2018. https://www.itpro.co.uk/hardware/32573/the-it-pro-product-of-the-year-awards-2018

“Dell EMC Demonstrates Significant Momentum in Advancing the High Performance Computing Community at ISC 2017,” Dell EMC, June 20, 2017. https://www.dell.com/learn/us/en/ph/press-releases/2017-06-20-dell-emc-startegic-agreement-with-nvidia

Sources:

1

2

3

4