ShaderMark

Problems

If you have problems, please download the debug build and mail me the "debug.log" and the "HAL_featureset.txt" including a dxdiag log file !

Updates and Fixes - Download the Full Package from a mirror on bottom of the page and then copy the updated ShaderMark.exe into the ShaderMark v2.1 install folder:

build 130a

compiled with the latest Intel 9.0 Compiler
linked against the latest Microsoft DirectX SDK (Juni 2005)
fixed a shader bug (Combination Effect) on ATi R3xx / R4xx cards
required D3DX library build 26
build 130 was a wrong debug build with unfinished code in it!!!

build 129

compiled with the latest Intel 8.1 Compiler
linked against the latest Microsoft HLSL Compiler (October 2004)
added version 1.0 of the ShaderMark v2.1 Chart Maker , which requires the .NET Framewrok v1.1

build 128

fixes an normalization bug in the HLSL_envmap.fx therefore new MSE values are needed, if you want to use the picture quality MSE comparision test and the some benchmark results will change
fixes the GeForce FX crash bug

build 125

fixes the MRT detection bug of the GeForce FX - however the crash bug on the GeForce FX is still under investigation!!!
fixed a CTD (crash to desktop) on some systems
fixed a resource allocation problem with shader 25 and 26, which could affect the performance!!!
Picture Quality Mode bitmaps are now removed when deleting old ShaderMark results

IMPORTANT - Driver & GPU requirements

The minimum system spec. is a shader model 2_0 card with DirectX 9.0c installed!
ATi: You need the latest Catalyst 4.9 in order to run ShaderMark v2.1 - the Shadow Mapping (shader 20) and Bump Mapping (shader 19) issues will be fixed in one of the next Catalyst Releases!!!
NVIDIA: Please use the latest WHQL 66.31 Driver from Station-Drivers, the 61.77 won't work!!!

What's new or changed in v2.1?

it's now free for all, no limited shareware version anymore
reduced cpu + driver overhead through redundant render state filtering
GPU performance optimized resource allocation
support for shader model 2_b and 3_0
nearly all shaders undergone changes and fixes
testing of >2_0 shaders
dynamic pixel shader flow control testing
floating point texture filtering performance testing
multiple render target performance testing
multiple shader test
mean square error (MSE) calculation in the image quality test - hardware rasterizer vs. Microsoft reference rasterizer

Overview

ShaderMark 2.1 is a DirectX 9.0 pixel shader benchmark. All pixel and vertex shader code is written in Microsoft’s High Level Shading Language. ShaderMark provides the possibility to use different compiler targets and advanced options. It also features a picture quality comparison based on mean square error (MSE) values.

Motivation

Currently there is no DirectX 9.0 HLSL pixel shader benchmark on the market. Futuremark's 3DMark03 (www.futuremark.com) and Massive's AquaMark 3.0 (www.aquamark.com) are bases on hand written assembler shaders or partly HLSL shaders. HLSL is the future of shader development! The HLSL shader compiler and its different profiles have to be tested and this gap fills ShaderMark v2.1. Driver cheating is also an issue. With ShaderMark, it is easily possible to change the underlying HLSL shader code which makes it impossible to “optimize” a driver for a certain shader, instead of the whole shader pipeline. ShaderMark v2.1 is also the first pixel shader benchmark, which measures the performance of the new shader model 3.0

Picture Quality Mode

The picture quality mode allows comparison of the Microsoft reference rasterizer with the current hardware rasterizer. This works only for the default (1024x768x32 with No FSAA) settings! It will calculate an MSE for every shader. The MSE values give a rough overview of how close the hardware renderer matches the Microsoft reference rasterizer:

shader 2 mse: 0.56019783

shader 3 mse: 0.54107634

shader 4 mse: 0.52548536

shader 5 mse: 0.52206866

shader 6 mse: 0.53719330

shader 7 mse: 0.43692334

shader 8 mse: 0.71990108

shader 9 mse: 0.40015380

shader 10 mse: 1.38709068

shader 11 mse: 0.64171155

shader 12 mse: 0.60582860

shader 13 mse: 0.66471100

shader 14 mse: 0.33196863

shader 15 mse: 1.29484272

shader 16 mse: 0.89861202

shader 17 mse: 0.30872790

shader 19 mse: 0.85641289

shader 20 mse: 0.56255150

shader 21 mse: 0.56256898

shader 22 mse: 7.91011810

shader 23 mse: 8.11475976

shader 24 mse: 8.11018531

shader 25 mse: 0.53177166

shader 26 mse: 0.53177166

The smaller the values, the better. Shader 18 isn’t included, because the hair animation would invalidate the results.

Community and Overall Points

Sorry, no community and no overall points. You should play games, not benchmarks ;).

If you want just one result, take shader 19 (Combination Effect)!

Technical Features

HLSL shaders only
pixel/vertex shader targets: ps_2_0, ps_2_a, ps_2_b, ps_3_0 / vs_2_0, vs_2_a, vs_3_0 with it’s advanced features
multiple render targets
sRGB textures
support for floating point textures and render targets (A16B16G16R16F, A32B32G32R32Fand R32F)
picture quality mode with Mean Square Error calculation based on the picture quality of the Microsoft Reference Rasterizer
gamma correction through DAC or frame buffer write (FBW)

low quality DAC

high quality FBW

most shaders uses an IBL (image based lighting) texture
auto generated mipmaps
command line mode for batch runs
width range of display options can be changed
shader execution is based on the hardware’s feature set, if specific features are not supported, the specific shader(s) will not work, for example the floating point texture shaders won’t work on NVIDIA’s GeForceFX 5xxx / S3’s DeltaChrome and XGI’s Volari cards

Basic Shaders

Features required:

· vs_2_0 / ps_2_0 shaders

· auto generated mipmaps

· floating point textures

(shader 1) ps1.1, ps1.4, ps2_0 and ps3.0 precision test (exponent + mantissa bits)
(shader 2) directional diffuse lightning

(shader 3) directional phong lightning

(shader 4) point phong lightning

(shader 5) spot phong lightning

(shader 6) directional anisotropic lighting

(shader 7) fresnel reflections

(shader 8) car paint shader (multiple layers)

(shader 9) environment mapping

(shader 10) bump environment mapping

(shader 11) bump mapping with phong lighting

(shader 12) self shadowing bump mapping with phong lighting

(shader 13) procedural stone shader

(shader 14) procedural wood shader

(shader 15) procedural tile shader

(shader 16) refraction and reflection shader with phong lighting

(shader 17) BRDF-phong/anisotropic lighting

(shader 18) fur shader (shells+fins)

(shader 19) combination effect

Advanced Shaders

· vs_2_0 / ps_2_0 or vs_2_x / ps_2_x or vs_3_0 / ps_3_0 shaders

· multiple render targets

· floating point textures

· floating point texture filtering

(shader 20) dual layer shadow map with 8x8 jittered percentage closer filter without flow control - (2_x shaders required)

(shader 21) dual layer shadow map with 8x8 jittered percentage closer filter with flow control - (3_0 shaders required)

(shader 22) HDR shader - low quality version without filtering - (2_0 shaders required)

(shader 23) HDR shader - high quality version with floating point filtering - (2_0 shaders required + floating point filtering)

(shader 24) HDR shader - high quality version without floating point - filtering(2_x shaders required)

(shader 25) NPR non photorealistic rendering using 1 render target and two passes - (2_0 shaders required)

(shader 26) NPR non photorealistic rendering using 2 render target and one pass(2_0 shaders required + multiple render targets)

Shader Details

performance impact of heavy alpha blending

shader 18 (Fur Shader With Anisotropic Lighting) makes heavy use of alpha blending

performance impact of multiple shaders switches

shader 19 (Combination Effect) uses many shaders, therefore it can be used to examine the overall performance when many shaders are used

performance impact of dynamic flow control

shader 20 (Dual Layer 8x8 PCF Shadow Mapping without Flow Control)
shader 21 (Dual Layer 8x8 PCF Shadow Mapping with Flow Control)
both shaders produce the same output, but shader 21 will be faster, if dynamic pixel shader flow control is efficient enough

performance impact of floating point filtering

shader 22 (High Dynamic Range Shader - low quality version without filtering) uses floating point or integer filtering only in the last stage and a 13x13 gaussian filter kernel to fit into 2_0 shaders

shader 23 (High Dynamic Range Shader - high quality with fp filtering) uses floating point filtering to reduce the gaussian filter kernel to 9x9

shader 24 (High Dynamic Range Shader - high quality without fp filtering) uses no floating point filtering and therefore requires a 25x25 gaussian filter kernel to achieve the 100% same output as shader 23

performance impact of multiple render targets

shader 25 (Per Pixel Edge Detection And Hatching Shader using 1 RT and 2 Passes)
shader 26 (Per Pixel Edge Detection And Hatching Shader using 2 RTs and 1 Pass)
with these two shaders, you can compare multi pass 1RT vs. single pass 2RTs rendering performance and therefore see how efficient MRTs (multiple render targets) are handled

Download Mirrors - Full Package