TEGRA K1 による GPU コンピューティング

Size: px

Start display at page:

Download "TEGRA K1 による GPU コンピューティング"

Clifton Cameron
5 years ago
Views:

1 TEGRA K1 による GPU コンピューティング COMPUTE WITH TEGRA K1 馬路徹シニアソリューションアーキテクト NVIDIA

2 AGENDA Introducing Tegra K1 Tegra K1 Compute Software Capabilities OpenGL GLSL OpenCL CUDA/Unified Memory Google Renderscript

3 Tegra K1 Kepler Architecture ISA Compatible to GeForce, Quadro, Tesla Tesla In Super Computers A15 A15 LP A15 A15 A15 64kB L1 Cache and Shared Memory Quadro In Work Stations 192 CUDA Cores 128kB L2 Cache GeForce In PCs Mobile Kepler In Tegra

4 TEGRA K1 DEVELOPMENT PLATFORMS Coming to Android K1 Devices soon JETSON TK1 GigE, USB3.0, HDMI running Linux4Tegra JETSON X3 (TK1 PRO) GigE, USB3.0, HDMI, 8 x Cameras, CANBUS running Vibrante Linux AUTOMOTIVE GRADE

5 SOFTWARE FOR COMPUTE Tegra K1 can accelerate Renderscript OpenGL /OpenGL ES with Compute Shaders NPP cufft, cublas, cusparse OpenCV OpenCL full profile CUDA and a whole list of libraries enabling compute on the GPU

6 TEGRA K1 FOR OPENGL/GLSL Kepler Architecture 192 CUDA Cores Cortex-A15 4-Plus-1 Shared Physical Memory 2D Engine / ISP

3+ implementations And now OpenGL ES 3.1!

7 COMPUTE SHADERS Standard OpenGL API Execute algorithmically general purpose GLSL shaders Operate on buffers, images and textures Process graphics data in the context of the graphics pipeline Easier than interoperating with a compute API for graphics apps Standard part of all OpenGL 4.3+ implementations And now OpenGL ES 3.1! Image processing AI Simulation Ray Tracing Wave Simulation Global Illumination

OPENGL COMPUTE SHADERS From Application From Application Element Array Buffer b Vertex Puller Dispatch Indirect Buffer b Dispatch Draw Indirect Buffer b Vertex Shader Image Load / Store t/b Compute

8 OPENGL COMPUTE SHADERS From Application From Application Element Array Buffer b Vertex Puller Dispatch Indirect Buffer b Dispatch Draw Indirect Buffer b Vertex Shader Image Load / Store t/b Compute Shader Vertex Buffer Object b Tessellation Control Shader Atomic Counter b Tessellation Primitive Gen. Shader Storage b Tessellation Eval. Shader Geometry Shader Texture Fetch t/b Transform Feedback Buffer b Transform Feedback Uniform Block b Legend Rasterization From Application Fixed Function Stage Programmable Stage Fragment Shader Pixel Assembly Pixel Unpack Buffer b b Buffer Binding Per-Fragment Operations Pixel Operations Texture Image t t Texture Binding Arrows indicate data flow Framebuffer Pixel Pack Pixel Pack Buffer b

9 TEGRA K1 FOR COMPUTE Kepler Architecture 192 CUDA Cores Cortex-A15 4-Plus-1 Shared Physical Memory 2D Engine / ISP

from desktop Higher precision, higher limits Awesome performance

10 TEGRA K1 FOR OPENCL OpenCL 1.2 Full profile support (OpenCL and OpenCL Embedded) True portability from desktop Higher precision, higher limits Awesome performance Related Session: US S Real-Time Facial Motion Capture and Animation on Mobile Emiliano Gambaretto

11 TEGRA K1 CUDA 6 AND SHARED PHYSICAL MEMORY Kepler Architecture 192 CUDA Cores Cortex-A15 4-Plus-1 Shared Physical Memory 2D Engine / ISP

CUDA REQUIRES MEMORY COPY Programmers are forced do perform

copy data from/to host to/from device Conventional Discrete

int i = blockidx.x*blockdim.x + threadidx.

cudamemcpy(d_x, x, N, cudamemcpyhosttodevice);

SAXPY on 1M elements saxpy<<<4096,256>>>(n, 2.

12 CUDA REQUIRES MEMORY COPY Programmers are forced do perform additional work to allocate memories both in host/device and copy data from/to host to/from device Conventional Discrete GPU global void saxpy(int n, float a, float *x, float *y) { } int i = blockidx.x*blockdim.x + threadidx.x; if (i < n) y[i] = a*x[i] + y[i]; PCIe 1 int N = 1<<20; cudamemcpy(d_x, x, N, cudamemcpyhosttodevice); cudamemcpy(d_y, y, N, cudamemcpyhosttodevice); 1 // Perform SAXPY on 1M elements saxpy<<<4096,256>>>(n, 2.0, d_x, d_y); Host Memory (CPUMemory) 2 Device Memory (GPU Memory) 2 cudamemcpy(y, d_y, N, cudamemcpydevicetohost);

13 CUDA UNIFIED MEMORY (FROM CUDA 6) Developer View Today Developer View With Unified Memory System Memory GPU Memory Unified Memory Dramatically Lower Developer Effort Faster performance on SoC

14 WHAT IS HAPPENING IN THE BACKGROUND

15 Audio Processor ARM7 TEGRA K1 PHYSICALLY SHARED MEMORY Physically CPU Quad Cortex-A15 + Shadow LP C-A15 CPU HD Video Processor Unified Memory without the need of Data SATA2 x1 USB 2.0 x3 PCIe G2 x4 + x1 Image Processor GPU Kepler 192 CUDA Cores Migration USB 3.0 x2 UART x4 I2C x5 SPI x4 SDIO/MMC x4 Display x2 HDMI edp/lvds CSI x4 + x4 NOR Flash DDR3 Ctlr 64b Security Engine DAP x5 (1 2 S/TDM) Unified Memory

16 CUDA UNIFIED MEMORY (FROM CUDA 6) Developer View Today Developer View With Unified Memory System Memory GPU Memory Unified Memory Dramatically Lower Developer Effort Faster performance on SoC

17 TEGRA K1 FOR RENDERSCRIPT Kepler Architecture 192 CUDA Cores Cortex-A15 4-Plus-1 Shared Physical Memory 2D Engine / ISP

18 RENDERSCRIPT C99 based kernel language Easy programmability with host and device portability. Portable across wide range of devices, fastest on Tegra K1 GPU (+ CPU) and more Renderscript API 19 Support

19 RENDERSCRIPT ON THE SOC Acceleration of Renderscript Scripts over GPU ScriptC, not just ScriptIntrinsics Huge gains in performance and performance/watt Runtime capable of scheduling work across units CPU GPU 2D Engine/ISP Related Session: US S Efficient Parallel Computation on Android Jason Sams, Tim Murray

20 INTRODUCTION (REF S4885)

21 CONSTRAINTS #1, #2 (Ref S4885)

22 CONSTRAINTS #3 (Ref S4885)

23 GPU OR CPU? (REF S4885)

24 DESKTOP PERFORMANCE TODAY (REF S4885)

25 MOBILE PERFORMANCE TODAY (SHIPPING) (REF S4885)

26 ARCHITECTUAL DIVERSITY? (REF S4885)

27 GOAL OF RENDERSCRIPT (REF S4885)

28 WHAT IS RENDERSCRIPT? (REF S4885)

29 RENDERSCRIPT INTRINSICS (REF S4885)

30 TEGRA K1? (REF S4885)

31 TEGRA K1? (REF S4885)

32 SUMMARY Tegra K1 内蔵のKeplerはTesla/Quadro/GeForceとアーキテクチャを共通とするスケーラブルなGPU これによりTesla/Quadro/GeForceで熟成されたCUDA, OpenCL, OpenGL Shader Languageのソフト資産開発環境が使用可能さらにHPC WS PCとは対極にあるモバイル用のRenderscriptに関しても GPUを活用することにより優れた性能を発揮する

33 THANK YOU

INTEGRATING COMPUTER VISION SENSOR INNOVATIONS INTO MOBILE DEVICES. Eli Savransky Principal Architect - CTO Office Mobile BU NVIDIA corp.

INTEGRATING COMPUTER VISION SENSOR INNOVATIONS INTO MOBILE DEVICES Eli Savransky Principal Architect - CTO Office Mobile BU NVIDIA corp. Computer Vision in Mobile Tegra K1 It s time! AGENDA Use cases categories