1. 1 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac GPU  Precision, Power, Programmability –CPU: x60/decade, 6 GFLOPS, 6GB/sec –GPU: x1000/decade, 20 GFLOPs, 25GB/sec –Arithmetic heavy (read OR write): faster hardware –Parallelization –Multi-billion $ entertainment market drives innovation –32-bit Floating point –Programmable (graphics, physics, general purpose data-flow) –Can’t simply “port” CPU code to GPU David Luebke et al. GPGPU, SIGGRAPH 2004

2. 2 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac History of the 3D graphics industry  60s: –Line drawings, hidden lines, parametric surfaces (B-splines…) –Automated drafting & machining for car, airplane, and ships manufacturers  70’s: –Mainframes, Vector tubes (HP…) –Software: Solids, (CSG), Ray Tracing, Z-buffer for hidden lines  80s: –Graphics workstations ($50K-$1M): Frame buffers, rasterizers, GL, Phigs –VR: CAVEs and head-mounted displays –CAD/CAM & GIS: CATIA, SDRC, PTC –Sun, HP, IBM, SGI, E&S, DEC  90s: –PCs ($2K): Graphics boards, OpenGL, Java3D –CAD+Videogames+Animations: AutoCAD, SolidWorks…, Alias-Wavefront –Intel, many board vendors  00s: –Laptops, PDAs, Cell Phones: Parallel graphic chips –Everything will be graphics, 3D, animated, interactive –Nvidia, Sony, Nokia

3. 3 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac History of GPU  Pre-GPU Graphics Acceleration –SGI, Evans & Sutherland. Introduced concepts like vertex transformation and texture mapping. Very expensive!  First-Generation GPU (-1998) –Nvidia TNT2, ATI Rage, Voodoo3. Vertex transformation on CPU, limited set of math operations.  Second-Generation GPU (1999-2000) –GeForce 256, Geforce2, Radeon 7500, Savage3D. Transformation & Lighting. More configurable, still not programmable.  Third-Generation GPU (2001) –Geforce3, Geforce4 Ti, Xbox, Radeon 8500. Vertex Programmability, pixel-level configurability.  Fourth-Generation GPU (2002-) –Geforce FX series, Radeon 9700 and on. Vertex-level and pixel-level programmability.

4. 4 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Architecture Vertex Shader Rasterizer Fragment Shader Compositor Display Application transformed vertices, normals, colors fragments (surfels per pixel) pixel color, depth, stencil texture Geometry Shader

5. 5 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Buffers  Color: 8-bit index to color table, float/16-bit true color…  Depth: 24-bit or float (0 at back plane)  Back and front: display front, update back, swap  Stereo: Shutter glasses, HMD. Alternate frames  Auxiliary: off-screen working space. Helps reduce passes.  Stencil: 8 bits (left-over of depth buffer). … mask, ++  Accumulation: sum, scale (supersampling, blur)  P-buffer, superbuffers: Render to texture

6. 6 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Fragment operations  Depth tests:, <=, ==, Z  depth-interval  Stencil test: mask?, counter, parity.  Alpha tests: compare to reference alpha  Alpha blending: + max, min, replace, blend

7. 7 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Data Parallelism in GPUs  Data flow: vertices > fragments > pixels  Parallelism at each stage  No shared or static data (except textures)  ALU-heavy (multiple ALUs per stage in pipe)  Fight memory latency with more computation

8. 8 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac GPGPU  Stream: collection of records (pixels, vertices…) –Stored in Textures (a computational grid)  Kernel: Function applied to each element in stream –Transform, evolve (no dependency between records) Matrix algebra Image/volume processing Physical simulation Global illumination –Ray tracing –Photon mapping –Radiosity

9. 9 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Computational Resources  Programmable parallel processors –Vertex & Fragment pipelines  Rasterizer –Mostly useful for interpolating addresses (texture coordinates) and per- vertex constants  Texture unit –Read-only memory interface  Render to texture (or Copy to texture) –Write-only memory interface

10. 10 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Vertex Processor  Fully programmable (SIMD / MIMD)  Processes 4-vectors (RGBA / XYZW)  Capable of scatter but not gather (A[i,j]=x;) –Can change the location of current vertex –Cannot read info from other vertices –Can only read a small constant memory  Vertex Texture Fetch –Random access memory for vertices –Arguably still not gather

11. 11 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Fragment Processor  May be invoked at each pixel by drawing a full screen quad  Fully programmable (SIMD)  Processes 4-vectors (RGBA / XYZW)  Random access memory read (textures)  Capable of gather (x=A[i+1,j];) and some scatter –RAM read (texture), but no RAM write –Output address fixed to a specific pixel –But can change that address  Typically more useful than vertex processor –More fragment pipelines than vertex pipelines –Gather –Direct output (fragment processor is at end of pipeline)

12. 12 SIC / CoC / Georgia Tech MAGIC Lab http://www.gvu.gatech.edu/~jarekJarek Rossignac Branching  Not supported or expensive  Avoid, replace by math  Depth test  Stencil test  Occlusion query (conditional execution)  Pre-computation (region of interest, use to set stencil mask)