At NAFEMS World Congress 2015 (June 21-24), Leonard Hoffnung from Siemens PLM Software, will discuss how GPUs and heterogeneous computing can be used to accelerate NX Nastran modal frequency response analysis for a Noise, Vibration, and Harshness application typical in the automotive industry.
Heterogeneous computing, or the use of GPU accelerators together with conventional single- or multiple-CPU systems, has become increasingly prevalent over the last decade. Heterogeneous systems have gained popularity, as they offer much higher theoretical flop rates compared with conventional workstations. However, successfully utilizing heterogeneous systems to improve the performance of engineering applications can be challenging.
In this paper, we examine a new method to utilize heterogeneous computing with multiple GPUs in a typical auto industry NVH (Noise, Vibration, and Harshness) application, modal frequency response analysis. Demands for improved fidelity have led to large and computationally expensive modal frequency response jobs over large frequency ranges. As problem sizes have increased, the dominant cost in a modal frequency response job has shifted from eigensolution to the frequency response calculation itself. Accordingly, we present a method to improve performance using multiple GPUs to target the frequency response calculation. The method takes advantage of multiple GPUs while minimizing inter-process communication for best efficiency. Two examples demonstrate the effectiveness of the method. The first illustrates the speedup potential with multiple GPUs in the response calculation specifically. The second example shows the performance impact of the improved response calculation, in the context of a complete industrial frequency response analysis.
If you can’t make it to SOLIDWORKS World 2015 to check out the AMD FirePro booth 501, at least you can check out this video.
This is the hang glider designed, visualized, sim-tested and manufactured by Wills Wing using SOLIDWORKS on AMD FirePro powered systems.
Specifically they used SolidWorks for 3D mechanical and electrical design, as well as finite element analysis simulation. They also use a CNC machine for producing various free form parts. Wills Wing hang gliders are typically 10 meters wide with crucial components with tolerances of a fraction of a mm. Being able to visualize them in real time, with ambient occlusion and OIT, all in detail is very important.
In the video Wills Wing President and Designer Steven Pearson flying the AMD FirePro T2C above Crestline, CA.
The Academy of Motion Picture Arts and Sciences today announced that the creator of the OpenCL-based Bullet physics engine, Erwin Coumans, will be honored with an Academy Technical Achievement Award.
The Oscar for Technical Achievement award goes to Erwin Coumans for the development of the Bullet physics library, and to Nafees Bin Zafar and Stephen Marshall for the separate development of two large-scale destruction simulation systems based on Bullet. These pioneering systems demonstrated that large numbers of constrained rigid bodies could be used to animate visually complex, believable destruction effects with minimal simulation time.
Unlike other Academy Awards to be presented this year, achievements receiving Scientific and Technical Awards need not have been developed and introduced during 2014. Rather, the achievements must demonstrate a proven record of contributing significant value to the process of making motion pictures.
Congrats to Erwin!, former Principal Physics Engineer at AMD where he lead the Bullet physics engine development, created an OpenCL rigid body pipeline running entirely on GPU and helped incorporate Bullet into a Siemens simulation software package and into Autodesk Maya.Effects.
AMD FirePro team was at the Siemens PLM Software CAE Symposium on board the Queen Mary in Long Beach demonstrating OpenCL-enabled NX Nastran finite element analysis solver accelerated by an AMD FirePro W9100.
The availability of fast approximate eigensolvers such as RDMODES makes it possible to compute a large number of modes over a wide frequency range economically. However, the modal frequency response calculation can be resource intensive for a large modal space. This project accelerated this computation in NX Nastran 9.1 using OpenCL on the AMD FirePro W9100 card.
The results of the FirePro/OpenCL acceleration for NX Nastran 9.1 Modal Frequency Response:
Up to 25x faster than serial
Up to 4x faster than the top of the line 24-core CPU run time
This year’s SIGGRAPH is amazing in terms of the demos. If you weren’t able to attend the show or make it to the AMD FirePro booth, below are 3 videos demonstrating a few of the remarkable technologies enabled by OpenGL and OpenCL running on AMD FirePro W8100 and W9100s. I have to note that each of these demos blew me away. This is leaps and bounds beyond what I saw last year.
Lumiscaphe P3D Design 50 fps photorealistic rendering using AMD FirePro W9100
It looks like a video playing on screen but it is actually a real-time 50 fps photorealistic rendering of a 4.1 million polygon model using P3D Design V6 running on an AMD FirePro W9100.
Cintoo 3D SmartMesh compression/decompression of massive 3D models on AMD FirePro using OpenCL
Cintoo 3D SmartMesh running on AMD FirePro W8100 using OpenCL to process 3D massive data in real-me. This example shows a 20 million triangles model with 20 GB texture compressed 1:50 and decompressed on the fly.
Ray tracer renderer proof of concept demo using OpenCL running on 4 AMD FirePro W8100s
AMD’s Takahiro Harada shows a path rendering / ray tracing demo using a prototype Maya 2015 plugin, targeting accelerated adoption of OpenCL 2.0 in advanced photorealistic rendering engines running on 4 AMD FirePro W8100.
At this year’s AMD FirePro SIGGRAPH booth 1023, Lumiscaphe will be demoing real-time rendering of complex photorealistic scenes at 4K using P3D running on HP workstation equipped with an AMD FirePro W9100. So just how impressive is this?
Check out the screen grabs below (scaled to 18% of original size and jpg compressed) of the images that were rendered in real-time at 4K. These images can be rendered and manipulated at 30-40 FPS at full 4K resolution using the AMD FirePro W9100. This will be a demo worth seeing live!
If you are attending Siemens PLM World next week, you will want to check out at least two of the demos in the AMD FirePro booth #14.
NX 9 Ray Traced Studio harnesses AMD FirePro graphics cards for GPU accelerated real-time ray tracing using OpenGL and/or OpenCL. The software checks the system configuration and uses the GPU if FirePro cards are available. With Ray Traced Studio, designers get photorealistic ray traced image results including global illumination, inner objet reflection and refraction and HDR based image base lighting.
NX Nastran is widely used finite element analysis (FEA) solver. It serves in different industries to support a wide range of engineering analyses including stress, vibration, structural failure, heat transfer, acoustics and aeroelasticity. NX Nastran is now GPU accelerated on AMD FirePro W9000 workstation cards and S9000 server cards. Widely used in aerospace and automotive, it provides support for modal frequency response.
The AMD FirePro team is at the 2014 SIMULIA Customer Conference in running May 20 - 22. AMD FirePro GPUs are certified for Abaqus technology in SIMULIA on 64-bit Windows and Linux platforms.
The photo below from the conference demonstrates SIMULIA Abaqus accelerated 2 to 4 times (over CPU only) by dual AMD FirePro W9100 GPU and OpenCL running on an HP Z820. The acceleration applies to both routine and sophisticated engineering simulation and the card can simultaneously drive analytics and visualization across multiple displays.
SIMULIA Abaqus accelerated using OpenCL on AMD FirePro W9100
Bernard Charlès, CEO of Dassault presenting at the conference
This article on 3DTotal describes how the OpenCL Bullet Physics plug-in for Maya, MayaBullet, has exploded the potential for game developers and visual effects teams to design more realistic dynamic and kinematic simulations.
From the article:
Using rigid-body, soft-body and collision-detection libraries, the MayaBullet Physics plug-in allows you to create small to large-scale realistic simulations for films, games and visualisations. Taking advantage of OpenCL, the MayaBullet plug-in allows you to very quickly create, view and edit simulations in pretty much real-time (depending the scene complexity of course) right inside the Maya viewport.
The AMD FirePro does not simply enhance the Bullet experience, but also allows you to push Maya further in certain ways.
1) It allows you to deal with heavier scene files, heavier models and more complex rigs. Therefore the animator can work with a model at a resolution that allows him to get a clearer idea of how the mesh will deform at render time.
2) By utilizing Viewport 2.0 and the DirectX 11 shader, you can truly get a better representation of your mesh with displacement maps, normal maps applied and more - all in the Maya viewport.
3) By taking advantage of the Alembic GPU Cache, you can bring a huge number of polygons (and by huge, we mean into the millions) into a scene and it still handles like a trooper.
FireUser.com is a community resource for CAD, visualization, 3D, video and engineering professionals to learn about the latest acceleration and display technologies and news with a focus on the AMD FirePro workstation graphics line.