Category

Uncategorized

Newsletter – NVIDIA GV100

By | Uncategorized

NVIDIA GV100

continued

The Quadro GV100 and its companion product, Quadro vDWS for the data center, address the growing demands of the world’s largest businesses to rapidly deploy deep learning-based research and development, accelerate deep learning-enhanced applications, enable photoreal VR and provide secure, anytime, anywhere access.

According to NVIDIA, the new Quadro GV100 packs 7.4 TFLOPS double-precision, 14.8 TFLOPS single-precision and 118.5 TFLOPS deep learning performance, and is equipped with 32GB of high-bandwidth memory capacity. Two GV100 cards can be combined using NVIDIA NVLink interconnect technology to scale memory and performance, creating a massive visual computing solution in a single workstation chassis.

Other benefits of the GV100 include:

  • Easy implementation of deep learning development – Access the NVIDIA GPU Cloud container registry with GV100 or other high-end Quadro GPUs for a comprehensive catalog of GPU-optimized software tools for deep learning and high performance computing on any workstation.
  • Accelerated deep learning training and inferencing on a desktop workstation – Dedicated Tensor Cores and the ability to scale two GV100s for up to 64GB of HBM2 memory with NVIDIA NVLink provide the performance required for demanding deep learning training and inferencing applications.
  • Supercharged rendering performance – Deep learning-accelerated denoising performance for ray tracing provides fluid visual interactivity throughout the design process.
  • Ability to run complex 3D simulations – Fast double-precision coupled with the ability to scale memory up to 64GB accelerates solver performance in computer-aided engineering workflows.
  • Collaborate, design and create in immersive VR – Support for advanced VR features and massive on-board memory capacity means designers can use physics-based, immersive VR platforms such as NVIDIA Holodeck to conduct design reviews and explore complex photoreal scenes and products at scale.

With newly added support for the NVIDIA Tesla V100 GPUs, Quadro vDWS has the power to address increasingly compute-intensive workflows and securely deliver workstation-class performance to any connected device.

Talk to our sales team about how to incorporate this into your system.

Newsletter – DGX-2

By | Uncategorized

The NVIDIA® DGX-2™

The new DGX-2 is powered by NVIDIA® DGX™ software with a scalable architecture built on NVIDIA NVSwitch – it allows you to take on the most complex AI challenges.

 

  1. Unbeatable Compute Power for Unprecedented Training
    As AI becomes increasingly complex and demands unprecedented levels of compute power NVIDIA® brings us the DGX-2. It is packed with the power of 16 of the world’s most advanced GPUs to accelerate new AI model types that were previously untrainable. It also enables groundbreaking GPU scalability, allowing you to train 4X bigger models on a single node with 10X the performance of an 8-GPU system.
  2. A Revolutionary AI Network Fabric
    Model complexity and size are no longer constrained by the limits of traditional architectures with the DGX-2. You can take advantage of model-parallel training with the NVIDIA NVSwitch networking fabric. This is the technology behind the world’s first 2-petaFLOPS GPU accelerator with 2.4 TB/s of bisection bandwidth, delivering a 24X increase over prior generations.
  3. AI Scale on a Whole New Level
    DGX-2 delivers a ready-to-go solution that offers the fastest path to scaling up AI. DGX-2 simplifies the speed of scaling-up AI with flexible networking options for building the largest deep learning compute clusters, combined with virtualization features with improved user and workload isolation in shared infrastructure environments. With an accelerated deployment model and an architecture purpose-built for ease of scale, you’ll spend more time driving insights and less time building infrastructure.
  4. Enterprise-Grade AI Infrastructure
    DGX-2 is purpose-built for RAS to reduce unplanned downtime, streamline serviceability and maintain operation continuity. DGX-2 is enterprise-grade, supported by NVIDIA expertise, and built for the rigor of around-the-clock operations to keep your most important AI endeavors running.

Talk to us about how you can explore the powerful performance of DGX-2 for yourself.

Newsletter

By | Uncategorized

The most powerful desktop CPU ever made?

continued

All four of the new CPUs feature AMD’s new boosting algorithms including features such as Precision Boost and Precision Boost Overdrive and sport boost frequencies of up to 4.4GHz.  AMD teamed up with Coolermaster to develop a solid air-cooling solution for the new CPUs, which run up to 250W TDP.  RAVE Computer is also offering customers factory-sealed AIO liquid cooling solutions in our line of RENDERBEAST workstations.

Threadripper 2990WX Threadripper 2970WX Threadripper 2950X Threadripper 2920X
Cores/Threads 32/64 24/48 16/32 12/24
Max Boost Freq 4.2GHz 4.2GHz 4.4GHz 4.3GHz
Base Freq 3GHz 3GHz 3.5GHz 3.5GHz
L3 Cache 64MB 64MB 32MB 32MB
TDP 250W 250W 180W 180W
Price $1799 $1299 $899 $649
Release Date 13th August October 31st August October

 

With a very aggressive price/performance ratio these processors offer an attractive value proposition to workstation users in need of higher core counts and/or more PCIe lanes (64) than competing processors available today.

Karl Rosenberger, Director of Product Development at RAVE Computer commented, “We are seeing excellent performance from these new AMD processors in well-threaded applications which benefit from their high number of cores.  With 64 PCI lanes and 64 thread concurrent processing, these processors provide outstanding value for power users.”

Excellence in Innovation – October

By | Uncategorized

Excellence in Innovation

continued

RAVE achieved the goal of their client by reducing the footprint and power requirements of the on-ship datacenter nodes with a compact 1U rugged server featuring Intel’s latest multi-core processors.  Now, the client is able to do more with significantly smaller computer hardware footprint.

Ruggedization and consolidation were achieved without bottlenecks thanks to bench testing followed by real world testing of first article. After implementing RAVE’s solution, the client enjoyed better overall performance with far fewer systems to transport, power and maintain. As a result, they gained space, weight and power budgets for other technologies where previously they did not have the capacity.

 

Data Processing – February

By | Uncategorized

A look into data processing of the future

continued

Traditional data centers are built for transactional workloads with limited computing needs.  They use many commodity servers interconnected with complex network infrastructures.  A GPU-accelerated data center is designed for workloads with infinite computing needs. I uses fewer, lightning fast nodes equal to the performance of thousands of commodity servers for simpler network infrastructure. Moreover, GPUs today are more efficient and powerful than ever, and there are massive multi-core processors from multiple sources including Intel, AMD, ARM, and NVIDIA. There is open source software available today that can be used to mine information from data. DIGITS is an NVIDIA project contributing to, leveraging and improving access to open source projects devoted to deep learning and machine learning.  

Currently, human analysts are buried with parsing through raw imagery to find the useful data within a set of data. Examples of that are: object identification for a self-driving car; detecting the right face in an ultra-large crowd in a city center to prevent the terrorist from deploying an IED; detecting persons in a video stream and accurately predicting their intentions; locating crop disease or underdeveloped areas of crops for a farmer’s field; or identifying cracks, corrosion, or other undesirable features for miles and miles of utility lines. There are systems that are very good at gathering the data that can be used to help make these decisions. Now with deep learning, high end machines with deep neural networks (DNNs) can be trained using known, tagged data (such as images, sound, text, etc.) and then can use an inference accelerator to plow through raw incoming data to identify subjects similar to those which the AI was trained on.  

Many-core processors such as GPUs and purpose-built FPGAs can perform inferencing extremely efficiently – humans do this very well, but super slow. You could now provide an analyst with a trained DNN and the output of the data wouldn’t just be a stitched together map or model, it could be reporting of an image found or matching facial features or object identification, etc. Recently, RAVE built a GPU server system for enterprise video processing and 3D model rendering for a customer. Implementation of the server has resulted in 75% decreased processing time per render. You’ll definitely hear more on this subject in the future. To learn more about how to get started in deep learning or how to grow your AI capabilities, contact RAVE today.

AUVSI XPONENTIAL RECAP

By | Uncategorized

AUVXI XPONENTIAL Recap

continued

Processors today are more efficient and powerful than ever. Vast amounts of data are available to train deep neural networks (DNNs). Mature algorithms which take advantage of parallel processing have been released in open source packages.

Many-core processors such as GPUs (like NVIDIA’s Volta architecture) can perform inferencing extremely efficiently. Humans do this very well, but very slowly.

Human analysts are currently bogged down parsing through raw imagery to find the useful information within a set of data. Some examples could be: detecting the right face in an ultra-large crowd in a city center to prevent the terrorist from deploying an IED; detect persons in a video stream and accurately predict their intentions; object identification for a self-driving car; crop disease or underdeveloped areas of crops for a farmer’s field; or cracks, corrosion, or other undesirable features for miles and miles of utility lines. Unmanned systems are very good at gathering the data that can be used to help determine this. Now, with deep learning, high end machines with DNNs can be trained using known, tagged data (such as images, sound, text, etc) and then can use an inference accelerator to plow through raw incoming data to identify subjects similar to those which the AI was trained on.

You could now provide an analyst with a trained DNN and the output of the data wouldn’t just be a stitched together map or model, it could be reporting of image found or matching facial features or object identification.

The end goal in all of this is to leverage advanced computing to analyze large amounts of data to deliver actionable information in order to maximize the performance of the humans who are making decisions.

Applications

By | Uncategorized

GPU Accelerated Defense Applications

Accelereyes Arrayfire

  • 2-20x Speed up*

Supported Features

  • Fast, easy-to-use GPU software library
  • Supports hundreds of most commonly used GPU functions
  • Available for C, C++, Fortran, and other languages

DigitalGlobe Advanced Ortho Series

  • 50x Speed up

Supported Features

  • Image orthorectification

Eternix Blaze Terra

  • 50x Speed up

Supported Features

  • 3D visualization of geospatial data

Harris – Exelis (ITT) ENVI

  • 70x Speed up

Supported Features

  • Image orthorectification (custom builds only)

GAIA

Supported Features

  • Built for GPUs
  • Scales to many machines and many cards without end user query modification or pre-meditation

GeoWeb3d Desktop

  • Requires GPU

Supported Features

  • 3D visualization of geospatial data

Incogna GIS

  • 50x Speed up

Supported Features

  • Image processing on Tesla cloud servers Object recognition

Intergraph Motion Video Analyst

  • 5-7x Speed up

Supported Features

  • Full motion video ortho mosaic processing

Intuvision Panoptes

  • 12x Speed up

Supported Features

  • Object recognition and change detection

MotionDSP Ikena

  • 5x Speed up

Supported Features

  • Video stabilization, filtering, mosaic’ing, transcofing

MrGeo

Supported Features

  • Terrain analytics

Nervve ViD SrX

  • 100x Speed up

Supported Features

  • Object recognition and tracking

NerVve Visual Search Solution (NVSS)

Supported Features

  • Object recognition and tracking

OpCoast SNEAK

  • 100x + Speed up

Supported Features

  • Ray tracing, DTED and remote sensing inputs

PCI Geomatics GXL

  • 20-60x Speed up

Supported Features

  • Image orthorectification and additional image processing

Ansys Savant (Delcross)

  • 20-50x Speed up

Supported Features

  • High-frequency solver

AMD Threadripper

By | Uncategorized

AMD’s Threadripper Series

On August 13th AMD lifted the lid on its 2nd Generation Threadripper CPUs with four new models starting with the new flagship 32-core processor, the Threadripper 2990WX.  These new 12nm processors use the same CPU socket as the first gen Threadripper CPUs and are supported by the majority of existing X399 chipset motherboards as long as a BIOS update is applied prior to installation.  The new models are split into two categories – the X-series, which succeed the previous Threadripper CPUs and a brand-new WX workstation-class series for professional workstations and power users.

All four of the new CPUs feature AMD’s new boosting algorithms including features such as Precision Boost and Precision Boost Overdrive and sport boost frequencies of up to 4.4GHz.  AMD teamed up with Coolermaster to develop a solid air-cooling solution for the new CPUs, which run up to 250W TDP.  RAVE Computer is also offering customers factory-sealed AIO liquid cooling solutions in our line of RENDERBEAST workstations.

AMD Ryzen Threadripper 2990WX

  • 32 Cores/64 Threads
  • 64MB L3 Cache
  • 64 PCIe Gen 3 Lanes
  • 250W TDP
  • Quad DDR4-2933
  • Supports ECC Memory
  • Supports both NVIDIA and AMD GPU’s
  • Release Date: 8/13/18

AMD Ryzen Threadripper 2970WX

  • 24 Cores/48 Threads
  • 64 MB L3 Cache
  • 64 PCIe Gen 3 lanes
  • 250W TDP
  • Quad DDR4-2933
  • Supports ECC Memory
  • Supports both NVIDIA and AMD GPU’s
  • Release Date: October 2018

AMD Ryzen Threadripper 2950X

  • 16 Cores/32 Threads
  • 32MB L3 Cache
  • 64 PCIe Gen 3 Lanes
  • 180W TDP
  • Quad DDR4-2933
  • Supports ECC Memory
  • Supports both NVIDIA and AMD GPU’s
  • Release Date: 8/31/18

AMD Ryzen Threadripper 2920X

  • 12 Cores/32 Threads
  • 32MB L3 Cache
  • 64 PCIe Gen 3 Lanes
  • 180W TDP
  • Quad DDR4-2933
  • Supports ECC Memory
  • Supports both NVIDIA and AMD GPU’s
  • Release Date: October

 

 

 

 

 


Threadripper 2990WX
 

Threadripper 2970WX

 

Threadripper 2950X

 

Threadripper 2920X

Cores/Threads 32/64 24/48 16/32 12/24
Max Boost Freq 4.2GHz 4.2GHz 4.4GHz 4.3GHz
Base Freq 3GHz 3GHz 3.5GHz 3.5GHz
L3 Cache 64MB 64MB 32MB 32MB
TDP 250W 250W 180W 180W
Price $1799 $1299 $899 $649
Release Date 13th August October 31st August October

With a very aggressive price/performance ratio these processors offer an attractive value proposition to workstation users in need of higher core counts and/or more PCIe lanes (64) than competing processors available today.

Karl Rosenberger, Director of Product Development at RAVE Computer commented, “We are seeing excellent performance from these new AMD processors in well-threaded applications which benefit from their high number of cores.  With 64 PCI lanes and 64 thread concurrent processing, these processors provide outstanding value for power users.”

AI-Medical Imaging – February

By | Uncategorized

AI Closes in on Medical Imaging

continued

The AI revolution is here – it extends traditional HPC by allowing researchers to analyze large volumes of data for rapid insights where simulation alone cannot fully predict the real world. Specifically in areas like medical imaging, bioinformatics, drug discovery, and high-energy physics. Tesla V100 is the computational engine driving the AI revolution and enabling HPC breakthroughs.

Over 500 HPC applications are already GPU optimized in a wide range of areas including quantum chemistry, molecular dynamics, bioinformatics, and more. A study by Intersect360 Research shows that 70% of the most popular HPC applications, including all of the top 10, have built-in support for GPUs. With most popular HPC applications and all deep learning frameworks GPU-accelerated, every researcher would see the majority of their data center workload benefit from GPU-accelerated computing.

Data center managers face the challenge of how to meet the demand for research computing resources that often exceed available cycles in the system.  A single server node with V100 GPUs can replace up to 50 CPU nodes. With lower networking, power, and rack space overheads, accelerated nodes provide higher application throughput at substantially reduced costs.

2018 Top Ten Innovations

By | Uncategorized

RAVE Top Ten Innovations for 2018

1. RENDERBEAST

RAVE’s mission was to provide the maximum desktop GPU capability whether needed for rendering, GPU compute or AI applications. It needed to be ‘well behaved,’ tame, yet quietly offering wicked performance for any GPU application.

The result was the next generation RENDERBEAST. This high-performance workstation is optimized for the design, rendering and analysis of large models on your desktop. It will improve product design, reduce development time and maximize efficiency – powerfully accelerating even the most challenging workflows.

RAVE integrated it with a choice of the latest liquid-cooled Intel® Xeon® Processors or the AMD Threadripper 2™. Packed with up to four of the latest NVIDIA Quadro GPUs, this system puts incredible capability into a very modest size.

The newest RENDERBEAST now gives the user the power to dive into the latest workloads including rendering and visualization, heavily parallel processing GPU compute or even AI/Deep Learning. It is the perfect solution for content creators, designers, engineers, architects and scientists, all in the same cool package.



2. NVMe High Speed Data Recording Server

RAVE’s customer was challenged with providing a rackmount solution for high-speed data recording in a military reconnaissance and surveillance aircraft. They needed the fastest possible read/write drive speeds and all drives were required to be in a removable carrier. They needed to build and support these systems for an extended-life service period, and operate in environments up to 55°C.

Using simulation analysis and physical testing, RAVE developed a solution delivering extended operation and performance in extreme temperatures.

RAVE developed a ruggedized, lightning fast, and compact NVMe server – fully utilizing available PCIe lanes for balanced and optimized throughput and maximizing performance. Managing for thermals was a challenge: 55°C ambient temperature is a full 20°C higher than most COTS products are certified. This solution also had to meet military aircraft shock and vibration requirements. In addition, the networking, architecture, and packaging for this very dense 5U were an engineering feat. This unique server allows users to take data center NVMe performance out in the field in more rugged environments…pumping data at breakneck speeds in comparison to SATA.

RAVE’s innovative and collaborative design approach delivered a data acquisition and storage solution for needed applications utilized on military aircraft. RAVE’s solution solved an incredibly challenging operating environment and space claim requirements well under the customer’s outlined budget.



3. miniBEAST

RAVE’s challenge was to design a system that delivered the same computing performance and visual output as an existing system but in half the size. We actually did it in ¼ the footprint! RAVE integrated an MXM-format graphics adapter designed for use in portable computers into an ultra-small form factor package with an Intel desktop-grade CPU, providing a significant step in performance per cubic inch of volume for users across simulation & training, design & engineering, and scalable visualization.

This machine can be incorporated into several configurations including single and dual-node 2U rackmount, wall mount or desktop. A locked BOM and long-life availability ensure compatibility for several years. The miniBEAST offers exceptional performance in a space-saving and adaptable package.



4. The Quiet Box

A Department of Defense customer needed a datacenter quality server solution, but it needed to be an isolated data center – segregated from public internet access.  It also needed to be extremely quiet given requirements of its intended use case environment.  Therefore, a robust, stand alone, security hardened system solution was required.

RAVE designed a dual CPU, 8 GPU, 4U system solution driving a performance level at the highest end of the customer’s needs spectrum. RAVE investigated significantly robust sound dampening options to create a near silent solution.

RAVE’s innovative and collaborative design approach delivered an artificial intelligence system with the potential to teach itself how to recognize and remove external interference from radar signals. RAVE’s solution delivered more than the customer originally thought possible while cutting nearly a third of the original target budget.

Quote from the customer: “I am happy to say that the system is up and running with hardly a hiccup! It is beautifully designed and well put together, and that cabinet is a real winner. Can’t even tell that it’s on from a sound perspective. Thanks to you and everyone there for a job well done.”



5. 4U Short Depth Rugged Signal Processing Server

A longtime Department of Defense customer executing signals intelligence was in need of a technology refresh to their current systems and desired a condensing of systems to produce a smaller hardware footprint. They also required removable hard drives for security reasons but needed a system that increased ease of the “remove and replace” (R/R) function of the hard drives.

RAVE designed a 4U server rackmount chassis solution powered by Intel Xeon embedded processors. RAVE engineers designed an easy to use 8-bay hard drive housing which allows for both individual hard drive hot swap as well as complete housing R/R. In addition to the storage feature, RAVE delivered a rugged system for use in transit cases, on aircraft, or co-located with forward operating units, able to handle more extreme temperature environments, shock and vibration considerations, and rougher handling. This server addresses the need for a wide bandwidth system that performs low latency, real-time, 100% duty cycle time and frequency signal processing in a short depth form factor utilizing high performance computer technology.

6. Autonomous Vehicle Development Appliance

RAVE’s customer was developing autonomous vehicles for large-scale operations. These robotic and autonomous systems (RAS) operations require real-time data processing and decision making. Throughout normal operations, RAS accumulate a large amount of sensor data that needs to be processed quickly to allow for RAS decision making. Due to the large number of variables that can affect the RAS, all of the data must be treated with equal importance. This combination of factors can make mobility testing of RAS both difficult and time consuming, which is especially true of real-world testing.

The real-world testing of RAS faces many challenges including the difficulty in conducting it efficiently and effectively, and in a repeatable fashion. It is also expensive and time consuming. Many of these challenges can be overcome by using a virtual world to conduct the testing.

RAVE designed High-Performance Computing (HPC) servers for the task – several inter-related aspects made the system design challenging, including the combination of several computer hardware requirements not often found in a COTS engineering server. Among the competing requirements were: a number of networked computers; space constraints of the installation location; the potential for 24/7 processing; high fidelity visual output from the simulation; and optimized system responsiveness with minimum latency.

RAVE’s HPC solution has the computing power to support the introduction of newer versions of the leader-follower autonomy software with more advanced autonomy behaviors into the real-time simulation environment. This HPC hardware solution was specifically designed for optimized performance of this software. This is superior to using a one-size-fits-all solution for this problem because of the customer’s need to be able to support running one vehicle simulation per computer for the autonomous system under test.

RAVE’s design approach reduced system footprint by half.



7. Multi-GPU High Throughput Server

RAVE’s mission was to provide the customer a dual-CPU, quad-GPU desktop high-performance computer with 16 discreet USB controllers. This represents the first application of this number of discreet inputs on a single motherboard, all packaged in a workstation configuration.

RAVE’s HPC workstation solution supports 4x double-width GPUs with 3 remaining open expansion slots for networking, storage, or other needs. Closed loop liquid cooling to the 2 high end Intel Scalable Xeon CPUs delivers increased performance and reliability and decreased acoustic output. Custom hardware was designed and tested in order to integrate the required discreet USB controllers and the liquid cooling system, all while ensuring the mechanical stabilization needs of the various components. A custom BIOS was required to support the entire configuration, which RAVE facilitated via our close partnership with Supermicro. This elegant system solution delivers multi-GPU configuration while also delivering high-throughput storage, network, and other I/O options – all packaged in one chassis housing instead of expansion or break-out boxes.



8. Compact Portable Training Solution

A defense and first responder training device developer asked RAVE to co-design a fully-integrated portable training device that could be set up in minutes.

The solution was to be completely self-contained, integrated into its own rolling case and incorporating a short-throw projector, professional daylight and low-light cameras on an adjustable mount with 3-axis calibration capability, a high-performance mini-PC, converged wireless and wired network capabilities, surge-suppression and power distribution equipment and a laptop computer for training control and management. RAVE’s engineering and technical staff tackled overall integration issues like size, weight and power while simultaneously leading PC development.

The final product delivers a powerful user experience capable of deploying anywhere and launching in minutes. This fully integrated and optimized hardware solution becomes a firearm use trainer when we install the client’s proprietary software applications on the hardware.



9. 3U Short Depth Intelligence Servers

RAVE designed a technology refresh system for one of our longtime prime contractor customers, delivering near real time on-scene intelligence collection, analysis, and dissemination capabilities on-board military reconnaissance and surveillance aircraft. This required size, weight, and power considerations in addition to ruggedization and long service life requirements.

RAVE’s solution is a 3U short depth, ultra-lightweight aluminum based system optimized specifically for use in both aircraft and ground stations. RAVE delivered a rugged system able to handle more extreme temperature environments, shock and vibration considerations, unique power requirements, and support longer life needs – 5-7 years after launch.

Quote from the customer: “I am looking forward to continue working with all of you as we continue to provide our war fighters with new technology. I am sure this will be another great year full of challenges and accomplishments. As always I appreciate the quality and professionalism all of you at RAVE Computer show in being able to provide the very best for our men and women who keep us free and safe.”



10. Private, Portable, On-Premise Cloud

RAVE co-designed a portable, local, on-premise cloud solution for a global science and technology solutions company. RAVE packaged a massive amount of compute into a chassis solution capable of being carried in a portable, rolling, hardened carrying case. This high powered system solution housed both large storage capability as well as multi GPU access for both visualization rendering and GPU compute.

RAVE’s solution delivered the key hardware component as part of an infrastructure for cloud and virtualized machines that could move simulation elements to an environment that would provide simple on-demand access to modeling and simulation applications anywhere in the world.