Transcript Document
Shader Performance Analysis on a Modern GPU Architecture Victor Moya, Carlos González, Jordi Roca, Agustín Fernández Department of Computer Architecture UPC Roger Espasa Intel DEG Barcelona 1 Introduction Shaders in GPUs evolving towards general programming Branches, generic loads, scatter New types of shaders: geometry in DX10 Current specialized shaders Area hungry Unbalancing leads to inefficiencies This paper: unify all shaders ~8% higher performance with less area & resources 2 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 3 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 4 ATTILA Our implementation of current GPUs Inspired in both NVIDIA and ATI Not exact to either pipeline Lack of detailed micro architecture information Educated guessing on our side Implemented Features 2D Homogeneous Recursive Rasterization Tiled Rasterization Hierarchical Z Texture compression Anisotropic filtering Depth compression, fast z/stencil and color clear 5 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 6 Attila Classic Vertex Fetch Vertex Shader Vertex Shader Vertex Shader Vertex Shader Primitive Assembly Clipping Specialized Shaders Triangle Setup Rasterization HierarchicalZ Fragment Shader Fragment Shader Fragment Shader Fragment Shader ROP ROP ROP ROP Memory Controller Memory Controller Memory Controller Memory Controller 7 Specialized Shader Issues Unbalancing In fragment shading limited scenarios (typical) up to 30% of the processing power remains idle (for a GPU with 8 vertex and 4 fragment shaders) In vertex shading limited scenarios up to 70% of the processing power remains idle. Dedicated Area 4 unused vertex shaders have the same processing power than one 1 fragment shader 4 vertex shaders require 66% the area of a fragment shader Different Designs Increases the complexity of the micro architecture Increases development and verification time 8 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 9 Attila Unified Vertex Fetch Shader Clipping Shader Shader Triangle Setup Distributor Scheduler Primitive Assembly Rasterization Shader HierarchicalZ Unified Shader Pool ROP Memory Controller ROP Memory Controller ROP Memory Controller ROP Memory Controller 10 Unified Shader Architecture Benefits Unified programming model DX10/SM4 and OpenGL/GLSlang are already pushing for it The same features for all the program targets Texturing, branching, outputs Not just vertex and fragment programs DX10 => geometry shader General Purpose GPU or Stream Processor Workload balance Shading resources allocated as required at any point of the rendering 11 Unified Shader Architecture Costs Scheduler Select which kind of workload must be processed next Partly implemented with multithreading in the fragment shader to hide texture access latency Larger instruction memory and constant bank Rerouting required All the paths cross the shader pool 12 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 13 ATTILA Framework OpenGL Interceptor tool OpenGL library for Attila GPU Driver for our Attila GPU Attila GPU simulator Signal Visualizer Tool 14 Collect Verify Simulate Analyze OpenGL Application GLInterceptor Trace GLPlayer Statistics Vendor OpenGL Driver Vendor OpenGL Driver ATTILA OpenGL Driver Signal Traffic ATI R520/NVidia G70 ATI R520/NVidia G70 ATTILA Simulator Framebuffer Framebuffer Framebuffer CHECK! Signal Visualizer CHECK! 15 Collect Verify Simulate Analyze OpenGL Application GLInterceptor •Capture a trace of OpenGL API alls from a real game GLInterceptor Trace GLPlayer Statistics Vendor OpenGL Driver Vendor OpenGL Driver ATTILA OpenGL Driver Signal Traffic ATI R520/NVidia G70 ATI R520/NVidia G70 ATTILA Simulator Framebuffer Framebuffer Framebuffer CHECK! Signal Visualizer CHECK! 16 Collect Verify Simulate Analyze OpenGL Application GLInterceptor GLPlayer •Reproduce the captured trace Trace GLPlayer Statistics Vendor OpenGL Driver Vendor OpenGL Driver ATTILA OpenGL Driver Signal Traffic ATI R520/NVidia G70 ATI R520/NVidia G70 ATTILA Simulator Framebuffer Framebuffer Framebuffer CHECK! Signal Visualizer CHECK! 17 Collect Verify Simulate Analyze OpenGL Library OpenGL Application - Transforms Fixed Function into Shader code - 200 API Calls supported - ARB Vertex and Fragment extensions - Alpha and Fog emulated via Shader code GLInterceptor Driver - Low level access - Attila memory management Trace GLPlayer Statistics Vendor OpenGL Driver Vendor OpenGL Driver ATTILA OpenGL Driver Signal Traffic ATI R520/NVidia G70 ATI R520/NVidia G70 ATTILA Simulator Framebuffer Framebuffer Framebuffer CHECK! Signal Visualizer CHECK! 18 Collect Verify Simulate Analyze ATTILA Simulator OpenGL Application - Detailed cycle-by-cycle simulation of all pipeline stages - 20 boxes, modeling a 100-deep pipeline - Execute@Execute: functionality embedded at each pipeline stage GLInterceptor Trace GLPlayer Statistics Vendor OpenGL Driver Vendor OpenGL Driver ATTILA OpenGL Driver Signal Traffic ATI R520/NVidia G70 ATI R520/NVidia G70 ATTILA Simulator Framebuffer Framebuffer Framebuffer CHECK! Signal Visualizer CHECK! 19 Find the differences NVIDIA GeForce FX 5900XT Attila 20 Outline Attila – our GPU architecture Attila-Classic: Non-unified shaders Attila-Unified: Unified Shaders Simulation Framework Results 21 Benchmark Unreal Tournament 2004 Fixed function OpenGL API Vertex and fragments shaders generated by our library 1024x768 resolution 8x Anisotropic Filtering 160 of 450 frames simulated 40 frames ~ 1 day simulation On a Xeon P4 @ 2.0Ghz 22 Baseline Configuration Four Vertex Shaders (only for Attila- Classic) Fragment and Unified shader configuration: 32 threads 4 fragments/vertices per thread 16 128-bit FP registers available for temporal storage per thread n SIMD ALUs 1 scalar ALU (optional) 1 Texture Unit per Shader Unit 16 KB texture cache Single cycle bilinear and two cycle trilinear AF up to 16x Geometry and Rasterization pipelines limited to 1 vertex and 1 triangle per cycle Two ROPs: 8 z and 8 color values written per cycle Four 64-bit DDR buses: peak bandwidth 64 bytes/cycle 23 “Classic” Performance 7% 3 relative performance 2,8 2,6 ~40% 8sh 2,4 2,2 2sh ~45% 6sh 2 1,8 6sh 4sh 8sh 1,6 ~75% 8% 1,4 1,2 4sh 2sh 1 1-way 1-way + scalar 2-way 4-way 8% improvement for 2-way Near linear improvement for 4 shaders Sublinear improvement for 6 and 8 shaders Limited by memory bandwidth and latency 24 Frame 330 – Detailed Zoom 1 0,9 Utilization 0,8 0,7 0,6 Vertex Shader 0,5 Fragment Shader 0,4 0,3 0,2 0,1 0 1 101 201 301 401 501 601 701 801 901 Time (10K cycles steps) Vertex shading limited Vertex shader and fragment shader workload for 4 vertex shader units and 2 fragment shader units 25 Unified Shader Performance relative performance 3 2,8 2,6 2,4 2,2 2 1,8 1,6 1,4 1,2 1 2sh 8sh uni2sh 6sh 4sh uni4sh 6sh 4sh uni6sh 8sh uni8sh 2sh 1-way 1-way + scalar 2-way 4-way Unified improvement ranges from 1% (2 shaders) to 8% (eight 1-way shaders) Fragment shading limited Vertex fetch limited Geometry pipeline limited 26 Area Estimation ATI R400 ATI RV400 Transistors (millions) 160 120 Vertex Shaders 6 4 Fragment Shaders 4 2 160 – 120 = 40 = 2 vertex shader * 2.5 + 2 fragments shader * 15 + 5 (other) Hardware Element Estimated Area Millions of Transistors Vertex Shader 2.5 Fragment Shader 15 Additional SIMD ALU +15% Additional scalar ALU +5% 27 Shader Scaling vs Transistors 170 150 2-way 130 fps 8sh uni 2-way 110 1-way 6sh uni 1-way 90 linear 4sh 70 2sh 50 30 80 130 180 MTransistors Linear for 4 shader units, sublinear for more than 4 shader units Up to 30% more efficient per area for the unified architecture (two 1way shaders) 28 Conclusion Attila Unified architecture has better performance than Attila Classic with less hardware Up to 8% better performance 8% to 25% less area required 10% to 30% better performance per area Up to 8% better performance for 2-way shader units 160% better performance from 2 to 8 fragment or unified shader units Memory bandwidth limited beyond 4 shaders 29 Questions 30 Performance of Attila Unified vs Classic Attila 1,09 relative performance 1,08 1,07 1,06 1-way 1,05 1-way + scalar 1,04 2-way 1,03 4-way 1,02 1,01 1 uni2sh uni4sh uni6sh uni8sh 31