Entries tagged as: OpenCL

ArrayFire v2.0 adds specific tuning for AMD FirePro GPUs, APUs and Radeon GPUs

Posted by Tony DeYoung on December 18, 2013

ArrayFire is an OpenCL library designed for maximum speed without the hassle of writing OpenCL device code. The library contains hundreds of functions for math, signal processing, image processing, and algorithms so developers can maximize productivity and performance.

ArrayFire V2.0 for OpenCL adds:

  • Support for all of ArrayFire’s function library (with a few exceptions)
  • Same API as ArrayFire for CUDA enabling seamless interoperability
  • Just-In-Time (JIT) compilation of kernels for top performance
  • Specific tuning for AMD FirePro GPUs, APUs and Radeon GPUs
  • Specific tuning for Intel Xeon Phi coprocessors
  • Accelerated algorithms for image processing, signal processing, visualization

Magnetar T25 workstation with AMD FirePro S10000 - ideal machine for GPU Compute, HPC, 3D rendering

Posted by Tony DeYoung on December 02, 2013

Armari Ltd has just released the new Magnetar T25 workstation with AMD FirePro S10000.  The high performance workstation is an ultimate OpenCL developer or application server system.  If interested, contact .(JavaScript must be enabled to view this email address).




AMD FirePro S10K 12 GB Edition at SC13 - up to 3 times faster performance

Posted by Tony DeYoung on November 20, 2013

From Supercomputing 2013, the Glen Mathews of the AMD FirePro team talks about the new passively cooled server GPU, the FirePro S10K 12 GB Edition, with up to 3 times greater performance than the previous 6GB model.

Also at the show is a large focus on OpenCL, particularly with Apple’s Mac Pro announcement using dual AMD FirePro GPUs.

Pics from the AMD FirePro booth #1113 at Supercomputing 2013 (SC13)

Posted by Tony DeYoung on November 18, 2013

Some photos from Supercomputing 2013 - SC13 demonstrating OpenCL and HSA computing solutions.  AMD is also debuting the new FirePro S10000 12GB Edition for HPC servers (announced last Friday)  and hosting several HPC talks in the AMD booth #1113 theater,





Tags: Events, HSA, OpenCL

AMD to host 11 HPC talks at SuperComputing 2013 - Booth #1113

Posted by Tony DeYoung on November 14, 2013

Fresh off the heels of this week's successful AMD Developer Summit (APU13) focused on heterogenous computing, as well as the announcement of the world's first 12GB GPU-accelerator built for servers - the AMD FirePro S10000 12GB Edition, AMD will also be hosting some killer talks in Booth #1113 at Supercomputing 2013 .

Spotlighting the most advanced scientific and technical applications in the world, SC13 will bring together the international supercomputing community for an exceptional program of technical papers, tutorials and talks.

AMD SC13 Theater Talk Schedule - Booth #1113

Date/Time Title Speaker
Tuesday Nov 19th
11am-noon Distributed Symmetric Multiprocessing - Shattering Affordability and Scalability barriers to large shared memory HPC Richard Anderson, CEO of Symmetric Computing
2-3pm Portals for Visualizing Graphics Intensive Big Data from HPC Bill Nitzberg, CTO of PBS Works
3-4pm OpenACC on the AMD APU and Discrete GPU Michael Wolfe, Compiler Engineer at PGI
4-5pm Case Study: "Accelerating Full Waveform Inversion via OpenCL on AMD GPUs" Dan Cyca, Chief Technology Officer, Acceleware
5-6pm Michael Waltrip Racing uses Penguin Computing Servers running AMD processors for a Competitive Advantage Donour Sizemore, Director of Trackside Systems at Michael Waltrip Racing
Wednesday Nov 20th
11am-noon AMD Software Ecosystem Update Margaret Lewis, Dir. of Product Mktg., Server Software at AMD
1-2pm Distributed Symmetric Multiprocessing - Shattering Affordability and Scalability barriers to large shared memory HPC Richard Anderson, CEO of Symmetric Computing
2-3pm A Vision for Collaborative Research with AMD Jay Owen, Dir. of External Research Office (AMD Research)
3-4pm Delivering Economical Big Data and Hadoop solutions with AMD and Penguin Computing David Ingersoll, WW VP of Sales Solutions at Penguin Computing
4-5pm Case Study: "Accelerating Full Waveform Inversion via OpenCL on AMD GPUs" Dan Cyca, Chief Technology Officer, Acceleware
5-6pm Introduction to HSA: HSAIL, Memory Model and Queuing Model Vinod Tipparaju, Software Engineer at AMD

AMD FirePro S10000 Passively Cooled Server Graphics Card with 12GB memory

Posted by Tony DeYoung on November 14, 2013

The new AMD FirePro S10000 12GB Edition is the world’s first server graphics card to feature 12GB of memory. Announced just in time for next week’s SC13, the card is designed for ‘big data’ high-performance computing (HPC) workloads delivering the highest peak single and double precision floating point performance in a single server graphics card product. It fully supports PCIe 3.0, ECC memory and DirectGMA.

With this card, memory hungry applications can now take advantage of the GPU for compute workloads and OpenCL acceleration. Each of the 56 Compute Units on the 12GB Edition (28 per GPU) are designed to handle two compute and one graphics operation at the same time, enabling independent scheduling and work item dispatch for more efficient multi-tasking.

Performance benefits will also accrue to applications that require high-speed, high capacity video memory such as Genetic Sequencing, Computational Fluid Dynamics, Structural Mechanics, Numeral Analytics, Reservoir Simulation, Automated Reasoning, and Weather Forecasting.

Importantly the GPU features passive cooling for high-density datacenter computing integration. You can see the AMD FirePro S10000 12GB at Supercomputing 13, booth #1113. The card is slated for general availability in spring 2014. 

100 megatriangle model streamed & manipulated real-time over web using OpenCL, AMD FirePro, Cintoo3D

Posted by Tony DeYoung on November 13, 2013

The photo is a demo of Cintoo3D SmartMesh at the APU13 Developer Summit. Cintoo3D provides a rich set of technologies using OpenCL to process 3D massive data in real-time including an level of detail (LOD) generator, codec, multi resolution textures, viewers and streaming.

The photo shows a 100 million triangles model of a scanned locomotive engine (courtesy of GOM ATOS 3D scan and analysis software) being streamed and manipulated in real-time over the web.


Tags: Events, OpenCL

Key ISV presentations at APU 13 Professional Graphics Track

Posted by Tony DeYoung on November 11, 2013

Today begins the AMD Developer Summit - APU 13 in San Jose, CA.  In addition to the keynotes and general sessions there are a series of AMD FirePro professional session given by key ISV partners. See full agenda here.

Tues     10:00 – 10:45     PG-4118        
Steve Hoeg       Adobe Systems          Advanced GPU
Accelerated Video Editing

Tues     11:00 – 11:45     PG-4037        
Leonard Hoffnung       Siemens PLM Software
Fast modal analysis with NX Nastran and GPUs

Wed     4:00 – 4:45       PG-4120        
Gideon Shmuel eyeSight       
Bringing next generation gesture technology to digital devices

Tues     4:00 – 4:45       PG-4039        
Dmitry Kozlov   AMD   
RapidFire API

Tues     5:00 – 5:45       PG-4035        
Wociech Tarnawski       Vratis 
Virtual Microscopy in the cloud

Wed     10:00 – 10:45     PG-4036        
Jakub Pola, Lukasz Miroslaw     Vratis 
Computational Fluid Dynamics of the blood flow in the cloud
Wed     11:00 – 11:45     PG-4119        
Jacques Lefaucheux     Cintoo3D       
3D Geometry Compression on GPU

Wed     2:45 – 3:30       PG-4034        
Karl Hillesland   AMD   
Using OpenGL and DirectX for Heterogeneous Compute

Wed     4:00 – 4:45       PG-4084        
Luis Crivelli       Dassault Systemes Simulia Corp        
Accelerating Abaqus Using AMD GPUs            

AMD FirePro W Series vs Quadro K Series in battery of real-world benchmarks via CGChannel

Posted by Tony DeYoung on November 11, 2013

CGChannel ran a battery of real-world and synthetic benchmarks comparing AMD FirePro W Series vs Quadro K Series cards. The benchmarks are broken into 3 categories: viewport and display performance FPS when carrying out basic manipulation operations on a 3D model or scene the viewport,  GPU computing and synthetic benchmarks.


The AMD FirePros came out on top more often than Nvidia’s Quadros in viewport display performance, and they are typically less expensive. With GPU computing, the Quadro cards take the lead for applications that require CUDA, and therefore Nvidia hardware. (Note however, for the growing number of applications that already support OpenCL like Premiere Pro, After Effects, SpeedGrade and DaVinci Resolve, this advantage will be reversed as demonstrated by the LuxMark OpenCL benchmark CGChannel results in which all of the AMD cards outperform all of the Nvidia cards, with the exception of the Quadro 6000, which falls in between the FirePro W7000 and W5000).

Notable standout where AMD FirePro cards take the performance lead:

  • 3ds Max 2013
  • Maya 2013
  • Softimage 2013
  • Mudbox 2013
  • Cinema 4D R14
  • Unreal Development Kit using DX11
  • CryEngine 3 using DX11

Notable standouts where Quadro takes the lead:

  • Modo 701
  • V-Ray RT 2.2
  • Blender 2.68

CUDA-only tests:

  • iray
  • Octane Render 1.20


Complimentary Access to APU 13 Keynotes - while supplies lasts

Posted by Tony DeYoung on November 06, 2013

Complimentary access to all APU 13 keynote sessions is available by registering online, in limited numbers while quantities last. Notable keynote speakers include: Dr. Lisa Su, senior vice president and general manager, Global Business Units, AMD; Johan Andersson, technical director, DICE; Dominic Mallinson, vice president, research and development, Sony; and Mark Papermaster, chief technology officer, AMD.

AMD FirePro Graphics Track

The AMD FIrePro Graphics track at APU 13 will feature sessions focused on areas of interest to parties developing professional graphics and computing applications as well as techniques for utilizing CPUs and multiple GPUs to get the most out of available graphics and computing power.

Session ID Speaker(s) Organization Title
PG-4032 Duoduo Liao Computing for Geospatial Research Institute GPU-Based Fast Generating Volumetric Models for 3D Printing
This paper proposes an approach to fast converting 3D solid objects into multivalued voxel-based models based on Graphics Processing Unit (GPU) accelerating for 3D printing or any other areas using 3D voxel-based models. In particular, by dynamically applying different transfer functions, any surface-closed geometric object can be voxelized into a solid volumetric representation with any kind of interior materials, such as rainbow, marble, wood, translucent jade, etcThis approach significantly improves the voxelization speed for solid geometric objects in real time. Moreover, with this method, other limitations, such as missing voxels, binary-only voxels, and non-transparency, are avoided as well. Due to slice-independent voxelization processing, parallel computing can be applied to this approach to improve overall performance.
PG-4035 Wociech Tarnawski, Jakub Poâa, Lukasz Miroslaw Vratis Sp. z o.o. Virtual Microscopy in the cloud
Image data generated in microscopy are is hard to maintain due to high number of data formats, various image processing workflows, data storage and complicated and time-consuming post-processing. High content screening and automated image processing is a current trend but the complexity of both technologies prevent non-specialists to use them in their work. High number of parameters to be set, necessity to use different software for different task rises both the work effort and the cost of the overall system. VIRTUM is a image-processing platform that addresses these issues by integrating various technologies in one place and providing the user a web-based access to the system. The images are stored in the cloud and analyzed on specialized hardware equipped in CPU and GPU nodes.
PG-4037 Leonard Hoffnung Siemens PLM Software Fast modal analysis with NX Nastran and GPUs
Industrial finite element analysis of large structures can consume a large amount of hardware resources and run time, particularly in the computation of eigenvalues and eigenvectors (also known as modal analysis). The RDMODES method in NX Nastran speeds up modal computation by a factor of ten or more, while preserving sufficient accuracy for most engineering purposes. We present performance improvements in RDMODES achieved by solving relatively small and dense subproblems on the GPU using the clMAGMA library.
PG-4084 Luis Crivelli Dassault Systemes Simulia Corp Accelerating Abaqus Using AMD GPUs
We present recent work done at SIMULIA to increase the performance of Abaqus by using GPU accelerators. In particular, we discuss the implementation of Abaqus Direct Sparse Solver and analyze the main technical issues relevant to the implementation of Explicit. Our main objective is to reduce the solver time in the context of multiple CPU cores combined with multiple GPUs. We demonstrate the performance gains of Abaqus/Standard for a number of realistic, customer-based, large problems run on last generation CPUs and the benefits obtained when GPUs and CPUs collaborate and complement each other.
Tags: Events, OpenCL
Page 3 of 13 pages  <  1 2 3 4 5 >  Last »