DOCSLIB.ORG

Matt Sandy Program Manager Microsoft Corporation

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Matt Sandy Program Manager Microsoft Corporation Matt Sandy Program Manager Microsoft Corporation Many-Core Systems Today What is DirectCompute? Where does it fit? Design of DirectCompute Language, Execution Model, Memory Model… Tutorial – Hello DirectCompute Tutorial – Constant Buffers Performance Considerations Tutorial – Thread Groups Tutorial – Shared Memory SIMD SIMD SIMD SIMD CPU0 CPU1 SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD CPU2 CPU3 SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD L2 Cache L2 Cache CPU GPU APU CPU ~10 GB/s GPU 50 GFLOPS 2500 GFLOPS ~10 GB/s ~100 GB/s CPU RAM GPU RAM 4-6 GB 1 GB x86 SIMD 50 GFLOPS 500 GFLOPS ~10 GB/s System RAM Microsoft’s GPGPU Programming Solution API of the DirectX Family Component of the Direct3D API Technical Computing, Games, Applications Media Playback & Processing, etc. C++ AMP, Accelerator, Brook+, Domain Domain D3DCSX, Ct, RapidMind, MKL, Libraries Languages ACML, cuFFT, etc. Compute Languages DirectCompute, OpenCL, CUDA C, etc. Processors APU, CPU, GPU, AMD, Intel, NVIDIA, S3, etc. Language and Syntax DirectCompute code written in “HLSL” High-Level Shader Language DirectCompute functions are “Compute Shaders” Syntax is C-like, with some exceptions Built-in types and intrinsic functions No pointers Compute Shaders use existing memory resources Neither create nor destroy memory Execution Model Threads are bundled into Thread Groups Thread Group size is defined in the Compute Shader numthreads attribute Host code dispatches Thread Groups, not threads z 0,0,1 1,0,1 2,0,1 3,0,1 4,0,1 5,0,1 6,0,1 7,0,1 8,0,1 0 0 1 1 0 1 2 0 1 x 0,0,0 1,0,0 2,0,0 3,0,0 4,0,0 5,0,0 6,0,0 7,0,0 8,0,0 8,1,1 y 0,1,0 01,1,0 0 0 2,1,0 3,1,0 14,1,00 0 5,1,0 6,1,0 27,1,0 0 0 8,1,0 8,2,1 [numthreads(3,2,1)] 2 1 1 0,2,0 1,2,0 2,2,0 3,2,0 4,2,0 5,2,0 6,2,0 7,2,0 8,2,0 8,3,1 0,3,0 01,3,0 1 0 2,3,0 3,3,0 14,3,01 0 5,3,0 6,3,0 27,3,0 1 0 8,3,0 8,4,1 pContext->Dispatch(3,3,2); 0,4,0 1,4,0 2,4,0 3,4,0 4,4,0 5,4,0 6,4,0 7,4,02 28,4,0 1 8,5,1 0,5,0 01,5,0 2 0 2,5,0 3,5,0 14,5,02 0 5,5,0 6,5,0 27,5,0 2 0 8,5,0 Direct3D Runtime Manages all interactions with the GPU Executes Compute Shaders Handles device memory allocation Compilation HLSL Code HLSL Compilation Options HLSL Compiler Offline (fxc.exe) Runtime (D3D extensions library) Intermediate Direct3D runtime consumes IL Language (IL) Driver Hardware Native Code Memory Model DirectCompute partitions memory into Resources Buffers for arbitrary data formats Textures for media formats Resources are accessed using Resource Views Conveys usage intent, permitting access optimizations Enables utilization of fixed-function translation and sampling Can create multiple views of a single resource Runtime connects Views to Compute Shaders Compute Shaders operate on views of resources Common Buffer Views Shader Resource View (SRV) Provides read-only access to a resource Unordered Access View (UAV) Provides read-write access to a resource Shader Resource View Compute Resource Shader Object Unordered Access View D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV UAV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV UAV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV UAV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); context->CSSetShader(…); context->CSSetShaderResources(…); context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV UAV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory D3D11CreateDevice(…); APU device->CreateComputeShader(…); device->CreateBuffer(…); SIMD Engine device->CreateBuffer(…); device->CreateShaderResourceView(…); device->CreateUnorderedAccessView(…); SIMD context->CSSetShader(…); context->CSSetShaderResources(…); SIMD context->CSSetUnorderedAccessViews(…); SimpleCS context->Dispatch(…); SIMD SIMD StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; SRV UAV [numthreads(256,1,1)] void SimpleCS( uint3 DTid : SV_DispatchThreadID ) Buffer0 Buffer1 { myOutput[DTid.x] = sqrt( myInput[DTid.x] ); } Device Memory Thread Groups 1x1x1 32x8x1 SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD SIMD Shared Memory StructuredBuffer<float> myInput; RWStructuredBuffer<float> myOutput; groupshared float temp[256]; [numthreads(256,1,1)] void main( uint3 DTid : SV_DispatchThreadID, uint GI : SV_GroupIndex ) { temp[GI] = myInput[DTid.x]; // copy device memory value to shared memory
Load more

Recommended publications
  • Comparison of Technologies for General-Purpose Computing on Graphics Processing Units
    Master of Science Thesis in Information Coding Department of Electrical Engineering, Linköping University, 2016 Comparison of Technologies for General-Purpose Computing on Graphics Processing Units Torbjörn Sörman Master of Science Thesis in Information Coding Comparison of Technologies for General-Purpose Computing on Graphics Processing Units Torbjörn Sörman LiTH-ISY-EX–16/4923–SE Supervisor: Robert Forchheimer isy, Linköpings universitet Åsa Detterfelt MindRoad AB Examiner: Ingemar Ragnemalm isy, Linköpings universitet Organisatorisk avdelning Department of Electrical Engineering Linköping University SE-581 83 Linköping, Sweden Copyright © 2016 Torbjörn Sörman Abstract The computational capacity of graphics cards for general-purpose computing have progressed fast over the last decade. A major reason is computational heavy computer games, where standard of performance and high quality graphics con- stantly rise. Another reason is better suitable technologies for programming the graphics cards. Combined, the product is high raw performance devices and means to access that performance. This thesis investigates some of the current technologies for general-purpose computing on graphics processing units. Tech- nologies are primarily compared by means of benchmarking performance and secondarily by factors concerning programming and implementation. The choice of technology can have a large impact on performance. The benchmark applica- tion found the difference in execution time of the fastest technology, CUDA, com- pared to the slowest, OpenCL, to be twice a factor of two. The benchmark applica- tion also found out that the older technologies, OpenGL and DirectX, are compet- itive with CUDA and OpenCL in terms of resulting raw performance. iii Acknowledgments I would like to thank Åsa Detterfelt for the opportunity to make this thesis work at MindRoad AB.
    [Show full text]
  • A Qualitative Comparison Study Between Common GPGPU Frameworks
    A Qualitative Comparison Study Between Common GPGPU Frameworks. Adam Söderström Department of Computer and Information Science Linköping University This dissertation is submitted for the degree of M. Sc. in Media Technology and Engineering June 2018 Acknowledgements I would like to acknowledge MindRoad AB and Åsa Detterfelt for making this work possible. I would also like to thank Ingemar Ragnemalm and August Ernstsson at Linköping University. Abstract The development of graphic processing units have during the last decade improved signif- icantly in performance while at the same time becoming cheaper. This has developed a new type of usage of the device where the massive parallelism available in modern GPU’s are used for more general purpose computing, also known as GPGPU. Frameworks have been developed just for this purpose and some of the most popular are CUDA, OpenCL and DirectX Compute Shaders, also known as DirectCompute. The choice of what framework to use may depend on factors such as features, portability and framework complexity. This paper aims to evaluate these concepts, while also comparing the speedup of a parallel imple- mentation of the N-Body problem with Barnes-hut optimization, compared to a sequential implementation. Table of contents List of figures xi List of tables xiii Nomenclature xv 1 Introduction1 1.1 Motivation . .1 1.2 Aim . .3 1.3 Research questions . .3 1.4 Delimitations . .4 1.5 Related work . .4 1.5.1 Framework comparison . .4 1.5.2 N-Body with Barnes-Hut . .5 2 Theory9 2.1 Background . .9 2.1.1 GPGPU History . 10 2.2 GPU Architecture .
    [Show full text]
  • Gpgpu Processing in Cuda Architecture
    Advanced Computing: An International Journal ( ACIJ ), Vol.3, No.1, January 2012 GPGPU PROCESSING IN CUDA ARCHITECTURE Jayshree Ghorpade 1, Jitendra Parande 2, Madhura Kulkarni 3, Amit Bawaskar 4 1Departmentof Computer Engineering, MITCOE, Pune University, India [email protected] 2 SunGard Global Technologies, India [email protected] 3 Department of Computer Engineering, MITCOE, Pune University, India [email protected] 4Departmentof Computer Engineering, MITCOE, Pune University, India [email protected] ABSTRACT The future of computation is the Graphical Processing Unit, i.e. the GPU. The promise that the graphics cards have shown in the field of image processing and accelerated rendering of 3D scenes, and the computational capability that these GPUs possess, they are developing into great parallel computing units. It is quite simple to program a graphics processor to perform general parallel tasks. But after understanding the various architectural aspects of the graphics processor, it can be used to perform other taxing tasks as well. In this paper, we will show how CUDA can fully utilize the tremendous power of these GPUs. CUDA is NVIDIA’s parallel computing architecture. It enables dramatic increases in computing performance, by harnessing the power of the GPU. This paper talks about CUDA and its architecture. It takes us through a comparison of CUDA C/C++ with other parallel programming languages like OpenCL and DirectCompute. The paper also lists out the common myths about CUDA and how the future seems to be promising for CUDA. KEYWORDS GPU, GPGPU, thread, block, grid, GFLOPS, CUDA, OpenCL, DirectCompute, data parallelism, ALU 1. INTRODUCTION GPU computation has provided a huge edge over the CPU with respect to computation speed.
    [Show full text]
  • Pny-Nvidia-Quadro-P2200.Pdf
    UNMATCHED POWER. UNMATCHED CREATIVE FREEDOM. NVIDIA® QUADRO® P2200 Power and Performance in a Compact FEATURES Form Factor. > Four DisplayPort 1.4 1 The Quadro P2200 is the perfect balance of Connectors performance, compelling features, and compact > DisplayPort with Audio form factor delivering incredible creative > NVIDIA nView™ Desktop experience and productivity across a variety of Management Software professional 3D applications. It features a Pascal > HDCP 2.2 Support GPU with 1280 CUDA cores, large 5 GB GDDR5X > NVIDIA Mosaic2 on-board memory, and the power to drive up to four PNY PART NUMBER VCQP2200-SB > NVIDIA Iray and 5K (5120x2880 @ 60Hz) displays natively. Accelerate SPECIFICATIONS MentalRay Support product development and content creation GPU Memory 5 GB GDDR5X workflows with a GPU that delivers the fluid PACKAGE CONTENTS Memory Interface 160-bit interactivity you need to work with large scenes and > NVIDIA Quadro P2200 Memory Bandwidth Up to 200 GB/s models. Professional Graphics NVIDIA CUDA® Cores 1280 Quadro cards are certified with a broad range of board System Interface PCI Express 3.0 x16 sophisticated professional applications, tested by WARRANTY AND Max Power Consumption 75 W leading workstation manufacturers, and backed by SUPPORT a global team of support specialists. This gives you Thermal Solution Active > 3-Year Warranty the peace of mind to focus on doing your best work. Form Factor 4.4” H x 7.9” L, Single Slot Whether you’re developing revolutionary products > Pre- and Post-Sales or telling spectacularly vivid visual stories, Quadro Technical Support Display Connectors 4x DP 1.4 gives you the performance to do it brilliantly.
    [Show full text]
  • Gainward GT 440 1GB DVI HDMI
    Gainward GT 440 1GB DVI HDMI GPU Core Memory Capacity Type Outputs GT 440 810 MHz 1600 MHz 1024 MB GDDR5 / 128 bits DVI, VGA, HDMI PRODUCT SPECIFICATIONS CHIPSET SPECIFICATIONS BUNDLED ACCESSORIES TM z Superior Hardware Design z NVIDIA GeForce GT 440 • Gainward QuickStart Manual z Gainward’s award winning High-Performance/ z Microsoft DirectX 11 Support: DirectX 11 Wide-BandwidthTM hardware design powered GPU with Shader Model 5.0 support by NVIDIA’s GeForceTM GT 440 GPU (40nm) designed for ultra high performance in the integrating 1024MB/128bits high-speed new API’s key graphics feature, GPU- GDDR5 memory which offers enhanced, accelerated tessellation. leading-edge performance for the 3D TM z NVIDIA PhysX : Full support for NVIDIA enthusiasts. PhysX technology, enabling a totally new z 810MHz core clock, 1600MHz (DDR 3200 class of physical gaming interaction for a memory clock. more dynamic and realistic experience with • Driver CD and Application S/W GeForce. z High performance 2-slot cooler. TM Driver for Windows 7/Vista/XP z NVIDIA CUDA Technology: CUDA z DVI (resolution support up to 2560x1600), technology unlocks the power of the GPU’s VGA and HDMI support. processor cores to accelerate the most demanding system tasks such as video z Full integrated support for HDMI 1.4a transcoding, physics simulation, ray tracing including xvYCC, Deep color and 7.1 digital and more, delivering incredible performance surround sound. over traditional CPUs. z Dual-link HDCP Capable: Designed to z Microsoft Windows 7 Support: Windows 7 meet the output protection management is the next generation operating system that (HDCP) and security specifications of the will mark a dramatic improvement in the way Blu-ray Disc formats, allowing the playback the OS takes advantage of the GPU to of encrypted movie content on PCs when provide a more compelling user experience.
    [Show full text]
  • 3D Graphics for Virtual Desktops Smackdown
    3D Graphics for Virtual Desktops Smackdown 3D Graphics for Virtual Desktops Smackdown Author(s): Shawn Bass, Benny Tritsch and Ruben Spruijt Version: 1.11 Date: May 2014 Page i CONTENTS 1. Introduction ........................................................................ 1 1.1 Objectives .......................................................................... 1 1.2 Intended Audience .............................................................. 1 1.3 Vendor Involvement ............................................................ 2 1.4 Feedback ............................................................................ 2 1.5 Contact .............................................................................. 2 2. About ................................................................................. 4 2.1 About PQR .......................................................................... 4 2.2 Acknowledgements ............................................................. 4 3. Team Remoting Graphics Experts - TeamRGE ....................... 6 4. Quotes ............................................................................... 7 5. Tomorrow’s Workspace ....................................................... 9 5.1 Vendor Matrix, who delivers what ...................................... 18 6. Desktop Virtualization 101 ................................................. 24 6.1 Server Hosted Desktop Virtualization directions ................... 24 6.2 VDcry?! ...........................................................................
    [Show full text]
  • Gainward Geforce 210 512MB DVI HDMI
    Gainward GeForce 210 512MB DVI HDMI GPU Core Memory Capacity Type Outputs GeForce 210 589 MHz 400 MHz 512MB DDR2 / 64 bits DVI‐I, VGA, HDMI PRODUCT SPECIFICATIONS CHIPSET SPECIFICATIONS BUNDLED ACCESSORIES z Superior Hardware Design z NVIDIA unified architecture: Fully • Gainward QuickStart Manual unified graphics core dynamically z Gainward’s award winning High-Performance/ allocates processing power to geometry, TM Wide-Bandwidth hardware design powered vertex, physics, or pixel shading TM by NVIDIA’s GeForce 210 GPU integrating operations, delivering superior graphics 512MB high-speed DDR2 memory which processing power. offers enhanced, leading-edge performance for the 3D enthusiasts. z NVIDIA CUDA Technology: CUDA technology unlocks the power of the z 589MHz core clock and 400MHz (DDR GPU’s processor cores to accelerate the 800) memory clock speed. most demanding system tasks – such as video transcoding – delivering incredible z Passive cooling. performance improvement over traditional z Provides DVI-I (Dual-link DVI, support up to CPUs. 2560x1600), HDMI and VGA (D-sub 15) z Microsoft Windows 7 Support: • Driver CD and Application S/W connectors. Windows 7 is the next generation Driver for Windows 7/Vista/XP z Provides HDMI functionality up to 1080p operating system that will mark a resolution. dramatic improvement in the way the OS takes advantage of the GPU to provide a z Supports two monitors simultaneously, more compelling user experience. By enabled by the NVIDIA nView™ Multi- taking advantage of the GPU for both Display Technology for office applications, graphics and computing, Windows 7 will 3D gaming and professional applications not only make today’s PCs more visual such as CAD, DTP, animation or video and more interactive but also ensure that editing.
    [Show full text]
  • Matrox Imaging Library (MIL) 9.0
    ------------------------------------------------------------------------------- Matrox Imaging Library (MIL) 9.0. MIL 9.0 Update 35 (GPU Processing) Release Notes (July 2011) (c) Copyright Matrox Electronic Systems Ltd., 1992-2011 ------------------------------------------------------------------------------- Main table of contents Section 1 : Differences between MIL 9.0 Update 35 and MIL 9.0 Update 30 Section 2 : Differences between MIL 9.0 Update 30 and MIL 9.0 Update 14 Section 3 : Differences between MIL 9.0 Update 14 and MIL 9.0 Update 3 Section 4 : Differences between MIL 9.0 Update 3 and MIL 9.0 Section 5 : MIL 9.0 GPU (Graphics Processing Unit) accelerations ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- Section 1: Differences between MIL 9.0 Update 35 and MIL 9.0 Update 30 Table of Contents for Section 1 1. Overview 2. GPU acceleration restrictions 3. New GPU functionalities and improvements 3.1. GPU accelerated image processing operations 3.1.1. MimHistogram 4. GPU specific examples 4.1. MilInteropWithCUDA (Updated) 4.2. MilInteropWithDX (Updated) 4.3. MilInteropWithOpenCL (New) 5. Is my application running on the GPU? 5.1. Deactivate MIL Host processing compensation 5.2. Windows SysInternals Process Explorer v15.0 (c) tool 6. Do MIL GPU results and precision change between updates? 6.1. MIL GPU algortihms 6.2. Graphics card firmware 7. Fixed bugs 7.1. All modules (DX9, DX10, DX11) 8. GPU boards 8.1. List of tested GPUs 8.2. GPU board limitations ------------------------------------------------------------------------------- 1. Overview - New DirectX 11 and DirectCompute support (shader models 5.0) - Minimal requirement: Microsoft Windows Vista with SP2 (or later), Windows 7 Note: Make sure that Windows Update 971512 (Windows Graphics, Imaging, and XPS Library) is installed to use DirectX 11 and Direct Compute in Windows Vista with SP2.
    [Show full text]
  • NVIDIA Parallel Nsight™
    NVIDIA Parallel Nsight™ Jeff Kiel Agenda: NVIDIA Parallel Nsight™ Programmable GPU Development Presenting Parallel Nsight Demo Questions/Feedback Programmable GPU Development More programmability = more power, more control and cooler effects! BUT more power = longer programs…how do I debug this code? How do I harness this amazing hardware in an environment I am familiar with? Image property of Unigine Corp., used by permission Programmable GPU Development My scene should look like this… Image property of Emergent Game Technologies Inc., used by permission Programmable GPU Development …but instead looks like this ! How do I…debug my skinning shader? Image property of Emergent Game Technologies Inc., used by permission Programmable GPU Development How do I… …figure out what model led to which draw call that produced some fragment that was or wasn’t blended properly to produce this broken pixel!?!? …understand why my performance tanks in this room or when a certain character enters the scene? …and on…and on…and on. Image property of Emergent Game Technologies Inc., used by permission Programmable GPU Development 2-4 cores 256-512 or more cores 6-12 concurrent threads 1000s…10000s concurrent threads Fundamental problem: Scaling from CPU to GPU is immense and we need tools to make this possible! Presenting Parallel Nsight™ GTX480 + Parallel Nsight + Visual Studio One Killer DirectX Development Environment Integrated into Visual Studio Powerful and familiar user interface Hardware-based shader debugger All shader types, including tesselation and compute No game modifications required Full DirectX10/11 frame debugging & profiling Including Pixel History, with debugger integration Combined CPU/GPU performance tools CPU and GPU activities on the same correlated timeline Parallel Nsight Environment Host Target TCP/IP Target Application Visual Studio CUDA DirectX Nsight Nsight Monitor Run remotely for Shader Debugging (GPU halts at breakpoint) Run locally or remotely for Frame Debugger and Profiling/Tracing Demo: Launching 1.
    [Show full text]
  • Nvidia-Quadro-P620-V2.Pdf
    UNMATCHED POWER. UNMATCHED CREATIVE FREEDOM. NVIDIA® QUADRO® P620 Powerful Professional Graphics with FEATURES > Four Mini DisplayPort 1.4 Expansive 4K Visual Workspace Connectors1 > DisplayPort with Audio The NVIDIA Quadro P620 combines a 512 CUDA core > NVIDIA nView® Desktop Pascal GPU, large on-board memory and advanced Management Software display technologies to deliver amazing performance > HDCP 2.2 Support for a range of professional workflows. 2 GB of ultra- > NVIDIA Mosaic2 fast GPU memory enables the creation of complex 2D PACKAGE CONTENTS PNY PART NUMBER VCQP620V2-PB and 3D models and a flexible single-slot, low-profile > NVIDIA® Quadro® P620 form factor makes it compatible with even the most Professional Graphics Board SPECIFICATIONS space and power-constrained chassis. Support for > Aattached full-height (ATX) GPU Memory 2 GB GDDR5 up to four 4K displays (4096x2160 @ 60 Hz) with HDR bracket color gives you an expansive visual workspace to view > Unattached low-profile (SFF) Memory Interface 128-bit bracket Memory Bandwidth Up to 80 GB/s your creations in stunning detail. > Four mDP to DP adapters NVIDIA CUDA® Cores 512 Quadro cards are certified with a broad range of > Printed Quick Start Guide > Printed Support Guide System Interface PCI Express 3.0 x16 sophisticated professional applications, tested by leading workstation manufacturers, and backed by WARRANTY AND SUPPORT Max Power Consumption 40 W a global team of support specialists, giving you the > 3-Year Warranty Thermal Solution Active peace of mind to focus on doing your best work. > Pre- and Post-Sales Technical Form Factor 2.713” H x 6.061” L, Whether you’re developing revolutionary products or Support Single Slot, Low Profile > Dedicated Field Application Display Connectors 4x Mini DisplayPort 1.4 telling spectacularly vivid visual stories, Quadro gives Engineers you the performance to do it brilliantly.
    [Show full text]
  • A Fast Fluid Simulator Using Smoothed-Particle Hydrodynamics
    A Fast Fluid Simulator Using Smoothed-Particle Hydrodynamics by Gustavo Figueroa A Thesis Presented in Partial Fulfillment of the Requirements for the Degree Master of Science Approved October 2012 by the Graduate Supervisory Committee: Gerald Farin, Chair Ross Maciejewski Yalin Wang ARIZONA STATE UNIVERSITY December 2012 ABSTRACT This document presents a new implementation of the Smoothed Particles Hydrodynamics algorithm using DirectX 11 and DirectCompute. The main goal of this document is to present to the reader an alternative solution to the largely studied and researched problem of fluid simulation. Most other solutions have been implemented using the NVIDIA CUDA framework; however, the proposed solution in this document uses the Microsoft general-purpose computing on graphics processing units API. The implementation allows for the simulation of a large number of particles in a real-time scenario. The solution presented here uses the Smoothed Particles Hydrodynamics algorithm to calculate the forces within the fluid; this algorithm provides a Lagrangian approach for discretizes the Navier- Stockes equations into a set of particles. Our solution uses the DirectCompute compute shaders to evaluate each particle using the multithreading and multi-core capabilities of the GPU increasing the overall performance. The solution then describes a method for extracting the fluid surface using the Marching Cubes method and the programmable interfaces exposed by the DirectX pipeline. Particularly, this document presents a method for using the Geometry Shader Stage to generate the triangle mesh as defined by the Marching Cubes method. The implementation results show the ability to simulate over 64K particles at a rate of 900 and 400 frames per second, not including the surface reconstruction steps and including the Marching Cubes steps respectively.
    [Show full text]
  • Languages, Apis and Development Tools for GPU Computing San Jose, California | 30 September 2009
    Languages, APIs and Development Tools for GPU Computing San Jose, California | 30 September 2009 Will Ramey Product Manager, GPU Computing NVIDIA Confidential GPU Computing Overview Broad Adoption Over 150,000,000 installed CUDA- Architecture GPUs GPU Computing Applications Over 90,000 GPU Computing Developers (9/09) Windows, Linux and Direct CUDA OpenCL Fortran Python, MacOS Platforms Compute supported C/C++ Java, .NET, … 1st GPU demo Microsoft API for PGI Accelerator PyCUDA GPU Computing spans Over 90,000 developers Shipped 1st OpenCL GPU Computing PGI CUDA Fortran jCUDA HPC to Consumer Running in Production since 2008 Conformant Driver Supports all CUDA- NOAA Fortran bindings CUDA.NET Architecture GPUs Public Availability OpenCL.NET SDK + Libs + Visual (DX10 and DX11) FLAGON 250+ Universities Profiler and Debugger teaching GPU Computing on the CUDA Architecture NVIDIA GPU with the CUDA Parallel Computing Architecture OpenCL is a trademark of Apple Inc. used under license to the Khronos Group Inc. © 2009 NVIDIA Corporation GPU Computing Application Development Your GPU Computing Application Application Acceleration Engines (AXEs) Middleware, Modules & Plug-ins Foundation Libraries Low-level Functional Libraries Development Environment Languages, Device APIs, Compilers, Debuggers, Profilers, etc. CUDA Architecture © 2009 NVIDIA Corporation Languages & APIs © 2009 NVIDIA Corporation CUDA C/C++ Update 2007 2008 2009 July 07 Nov 07 April 08 Aug 08 July 09 Nov 09 CUDA “Nexus” CUDA Toolkit 1.0 CUDA Toolkit 1.1 CUDA Toolkit 2.0 CUDA Toolkit
    [Show full text]