Presentation is loading. Please wait.

Presentation is loading. Please wait.

Rendering pipeline: The hardware side

Similar presentations


Presentation on theme: "Rendering pipeline: The hardware side"— Presentation transcript:

1 Rendering pipeline: The hardware side
Zoltán Szandtner, college assistant lecturer of Department of Basic and Technical Sciences of DGC 3D Graphics Professional Days 2016.

2 I. Why are GPUs important?
Figure 1 – Intel CPU Trends [1] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

3 II. The graphics pipeline /1
Traditional pipeline …and its modern implementation Figure 2 – Graphics Pipeline [WikiMedia Commons] In layman’s terms Figure 3 – OpenGL infographic by Shamus Young [2] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

4 II. The graphics pipeline /2
Figure 3 – OpenGL infographic by Shamus Young [2] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

5 III. Brief history of Graphics Cards & GPUs /1
Figure 4 – IBM Monochrome Display Adapter [Wikimedia Commons] „1st generation”: Focus on 2D end of the pipeline improving resolution & color depth  SVGA standard On-board RAMDAC buffer  video memory precursor 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

6 III. Brief history of Graphics Cards & GPUs /2
Figure 5 – 3Dfx Voodoo [vgamuseum.ru] 2nd generation: 3D accelerators & 1st GPUs Graphics APIs: Glide, OpenGL, DirectX Hardware is a fixed-function pipeline, i.e. fixed, „built-in math operations” 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

7 III. Brief history of Graphics Cards & GPUs /3
Figure 6 – GeForce3 ti [vgamuseum.ru] 3rd generation: GPUs with programmable shader support First pixel, then vertex shaders are programmable Shader „languages”: GLSL (OpenGL), HLSL (DirectX) 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

8 III. Brief history of Graphics Cards & GPUs /4
Figure 6 – ATI Radeon HD 2000 [vgamuseum.ru] 4th generation: Modern GPUs with Unified Shader Architecture Figure 7 – Fixed-function vs Unified Shader Models [WikiMedia Commons] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

9 IV. A note on Intel integrated GPUs:
Figure 8 – GPU market shares Nvidia/ATI/Intel [WikiMedia Commons] „Intel is also doing a lot better with their integrated graphics parts, once the butt of jokes, but they've taken a couple of steps now which are fully competent parts” – John Carmack, 2011 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

10 V. Modern GPU architecture
Figure 9 – Typical CPU vs GPU circuit area ratios [3] Majority of a GPU wafer assigned to execution units Classic „brainiac” vs. „speedracer”  work smart vs. hard 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

11 Zoltán Szandtner - Rendering pipeline: the hardware side
VI. Why speedracer? Graphics processing (usually) embarrassingly parallel Simpler control is sufficient Workload is greatly predictable Logic optimization can focus on throughput vs latency Data shows very high locality Memory architecture can be likewise throughput focused GPU code (typically) compiled at runtime VLIW like static scheduling can be employed More area can be assigned to execution units 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

12 VII. NVidia SIMT (CUDA) Architecture in Detail /1
Figure 10 – CUDA blocks & memory model [3] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

13 VII. NVidia SIMT (CUDA) Architecture in Detail /2
Figure 11 – SMP architecture [4] Figure 12 – Maxwell GPU architecture [4] Grid  Entire GPU; Block  SMP; Thread  SP 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

14 VIII. How much faster are GPUs? /1 - Theory
Figure 13 – Theoretical difference [3] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

15 VIII. How much faster are GPUs? /2 - Reality
Figure 14 – Measured difference [5] 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side

16 Zoltán Szandtner - Rendering pipeline: the hardware side
Proceedings of the 37th Annual International Symposium on Computer Architecture IX. References [1] H. Sutter, „The Free Lunch Is Over: A Fundamental Turn Toward Concurrency in Software,” Dr. Dobb's Journal, web, vol. 30, no. 3, March 2005. [2] Young Shamus „ The Strange Evolution of OpenGL Part 2”, Twenty Sided Blog, web [3] Nvidia, „CUDA Programming Guide”, web, 2015, CUDA_ProgrammingGuide.pdf [4] Nicholas Wilt, „The CUDA Handbook”, Crawfordsville, Indiana, 2013 [5 ] V. W. Lee, C. Kim, J. Chhugani, M. Deisher, D. Kim, A. D. Nguyen, N. Satish, M. Smelyanskiy, S. Chennupaty, P. Hammarlund, R. Singhal and P. Dubey, „Debunking the 100X GPU vs. CPU Myth: An Evaluation of Throughput Computing on CPU and GPU,” SIGARCH Comput. Archit. News, vol. 1. no.3 , pg , 2010. 3D Graph. Prof. Days 2016. Zoltán Szandtner - Rendering pipeline: the hardware side


Download ppt "Rendering pipeline: The hardware side"

Similar presentations


Ads by Google