GPU PerfStudio

Overview

GPU PerfStudio is a real-time performance analysis tool which has been designed to help tune the graphics performance of your DirectX 9, DirectX 10, and OpenGL applications. GPU PerfStudio displays real-time API, driver and hardware data which can be visualized using extremely flexible plotting and bar chart mechanisms. The application being profiled maybe executed locally or remotely over the network. GPU PerfStudio allows the developer to override key rendering states in real-time for rapid bottleneck detection. An auto-analysis window can be used for identifying performance issues at various stages of the graphics pipeline. No special drivers or code modifications are needed to use GPU PerfStudio.

» Versions
» Counter Descriptions
» State Overrides
» Screenshots
» Downloads
» Setup/Installation

AMD GPU PerfStudio Versions

What’s New in Version 1.2?

Client and server support for OpenGL on XP and Vista

What’s New in Version 1.1?

Client and server support for DirectX® 10/Vista/Radeon™ HD 2000 family

Automatic bottleneck detection

Cell formatting

Enhanced filtering

Separate render target and back buffer state overrides

Drag and drop for plots and bar charts

Selectable UI antialiasing

Flexible bar charts

Counter Descriptions

GPU PerfStudio provides statistics on every D3D call executed per frame in addition to the hardware and driver level counters described below. Real time graphs and bar charts can easily be created for all numeric data. Not all counters are available for all cards.

Hardware Computer	Description
% Hardware Utilization	Percent time GPU is busy
% Vertex Wait for Pixel	Percent time vertex processing is waiting for pixel processing to finish (can indicate slow pixel shaders)
% Pixel Wait for Vertex	Percent time pixel processing is waiting for vertex processing to finish
Pre-clip Primitives	Primitive count before clipping
Post-clip Primitives	Primitive count after clipping
% Blended Pixels	Percent of total pixels drawn with blending enabled
ALU to Texture Instruction Ratio	Ratio between pixel shader ALU and texture instructions
% Pixels Passed Z-test	Percent of pixels which passed the Z-test
Overdraw	Total number of pixels drawn divided by the Overdraw counter resolution. This counter can also be representative of the number of render targets in use.
Texture Cache Miss Rate	Texture cache miss rate in bytes per pixel
Post HiZ Sample Count	Number of samples after HyperZ
Post TopZ Pixel Count	Pixels after early Z culling has taken place
Post Shader Pixel Count	Pixels after shading and alpha test have taken place
TopZ Reject Rate	Rate of pixel rejection due to early Z test
Back to top

Driver Data Counter

Description

Framerate

Frames per second

LocalTextureMem

Local Texture Memory used

AGPTextureMem

AGP texture memory used

LocalVBIBMem

Local Vertex buffers and index buffer memory used

AGPVBIBMem

AGP Vertex buffers and index buffers memory used

TextureUpload

Texture data uploaded

VBIBUpload

Vertex buffers and index buffers data uploaded

PrimsPerRSChange

Primitives rendered per render state change

PrimsPerTSChange

Primitives rendered per texture state change

PrimsPerVSChange

Primitives rendered per vertex shader change

PrimsPerPSChange

Primitives rendered per pixel shader change

PrimsPerVSCChange

Primitives rendered per vertex shade constant change

PrimsPerPSCChange

Primitives rendered per pixel shader constant change

FlipStall

Stalls on frame buffer flip

VBStall

Stalls on vertex buffer

GeometryBufferAllocatedDefault

GeometryBufferAllocatedImmutable

GeometryBufferAllocatedDynamic

GeometryBufferAllocatedStaging

GeometryBufferAllocated

Allocated memory for vertex and index buffers and stream output

GeometryBufferUsedPercentage

Percentage of allocated geometry buffer memory used

ConstantBufferAllocatedDefault

ConstantBufferAllocatedImmutable

ConstantBufferAllocatedDynamic

ConstantBufferAllocatedStaging

ConstantBufferAllocated

Allocated memory for constant buffers

ConstantBufferUsedPercentage

Percentage of allocated constant buffer memory used

RenderTargetAllocatedDefault

RenderTargetAllocatedImmutable

RenderTargetAllocatedDynamic

RenderTargetAllocatedStaging

RenderTargetAllocated

Allocated memory for render targets

RenderTargetUsedPercentage

Percentage of allocated render target memory used

TextureDepthStencilShaderAllocatedDefault

TextureDepthStencilShaderAllocatedImmutable

TextureDepthStencilShaderAllocatedDynamic

TextureDepthStencilShaderAllocatedStaging

TextureDepthStencilShaderAllocated

Allocated memory for ShaderResources, DepthStencil buffers and Textures

TextureDepthStencilShaderUsedPercentage

Percentage of allocated TextureDepthStencilShader memory used

PrimsPerDepthStencilStateChange

Primitives rendered per depth stencil state change

PrimsPerBlendStateChange

Primitives rendered per blend state change

PrimsPerGeometryShaderChange

Primitives rendered per geometry shader change

PrimsPerPSSamplerStateChange

Primitives rendered per pixel shader sampler state change

PrimsPerVSSamplerStateChange

Primitives rendered per vertex shader sampler state change

PrimsPerGSSamplerStateChange

Primitives rendered per geometry shader sampler state change

State Overrides

Override	Description/Possible Bottleneck
Force 2x2 Textures	Is texture bandwidth (large textures) affecting performance?
Force Disable Texture Filtering	Are expensive texture filtering modes affecting performance?
Force 1x1 Scissor Region	Identifies vertex processing bottlenecks (by removing most pixel processing)
Force Simple Pixel Shaders	Identifies expensive pixel shaders
Force Skip Draw*Prim Calls	Identifies non-GPU bottlenecks (by removing most 3D graphics work)
Force Z Test Enable	Identifies z-order performance issues
Force Z Write Enable	Identifies z-order performance issues
Force Alpha Blend Enable	Identifies alpha-blending performance issues
Force Alpha Test Enable	Can identify problems related to early Z test
Force Cull Mode	Can show culling efficiency
Force Fill Mode	Used for debugging and identifying vertex density
Back to top

GPU PerfStudio Vista Screenshots

Figure 1: Click to enlarge.

Important notes:

DX9 applications on Vista are only supported in API mode; no hardware data or driver data is available

Applications to be profiled cannot be started with batch files or “launcher” applications (an exe which starts another exe). You can use shortcuts (links) for command line arguments.

Catalyst 7.10 or later is not supported for DX9 applications on systems running XP with HD 2000 and HD 3000 family graphics cards; instead use Catalyst 7.9.

Downloads

» GPU PerfStudio 1.2 installer: GPUPerfStudio-v1.2.msi
» GPU PerfStudio 1.2 release notes: GPUPerfStudio-ReleaseNotes.txt
» GPU PerfStudio 1.2 documentation: GPUPerfStudioHelp.pdf

Contact Us

Please send any feedback, questions or suggestions to gputools.support@amd.com.

Setup/Installation

GPU PerfStudio requirements:

Windows XP SP2 / Vista (32 bit)

Radeon 9500 or better (for hardware data, server machine; CrossFire not supported)

CATALYST 7.12 (for hardware data, server machine)

Before installing GPU PerfStudio, uninstall all previous versions of GPU PerfStudio and PerfDash. GPU PerfStudio works best when run remotely. You must run the GPU PerfStudio installer on both machines – the “client” machine running the GPU PerfStudio application and the “server” (target) machine running the application to be profiled. If you select a custom install, you can install only the server or client on an individual machine.