Intel® Graphics Performance Analyzers (Intel® GPA) Quick Start Guide

Submit New Article

Last Modified On :   October 30, 2009 1:52 PM PDT
Rate
 


NOTE: even though the screen shots depicted in this article were created with GPA Version 2.0, this Quick Start Guide shows how to use some of the key features of the product, and is therefore still usable with versions 2.1 and 2.2 of GPA.

INTRODUCTION

Welcome to the Intel® Graphics Performance Analyzers (Intel® GPA) Quick Start Guide. In this guide, we'll take you through the installation, setup, and running of the key components of the GPA product.

We recommend running GPA in a client/server configuration utilizing two computer systems. Here, the game to be analyzed runs on the target graphics system (or "server"), and the GPA analysis tools run on the client system. The target system should be a fully supported graphics system, such as a system utilizing the Intel® G45 Express Chipset.

For optimal performance, both the client and server systems should be located on the same subnet of your LAN, and both should be using a wired connection. In order to ensure accurate metrics, you'll also want to disable as many other non-essential processes as possible on both systems (including screen savers), and be sure to run the game in full-screen mode.

For detailed information on the GPA tools, refer to the help file that is installed with the product. If you have any issues using GPA, or require additional detailed information on the product, please refer to the "Release Notes" or the GPA Home Page for technical Forums, Knowledge Base articles, and other training materials.

GPA Installation

First of all, download the GPA installation package and the "Release Notes" from the Intel installation site, and then copy the installation package onto both the client and server systems. Be sure to read the "Release Notes" for any last-minute information on installing, configuring, and running the GPA product.

On the client system, I've double-clicked the installation package, which starts the installation process (see screen shot below). If any required dependent software is missing from your system, the installation process will first install these other packages.

On the server system, the installation of the GPA package proceeds pretty much the same way as on the client. However, as we'll be running our game on this system, we'll also need to install the latest version of the Intel graphics driver on this system (these drivers have been updated to include updates to the instrumentation modules required by GPA). To do this, using Microsoft Internet Explorer* you can browse to the driver update page, and follow the instructions to update the drivers to the correct version. If for some reason the driver install process does not complete successfully, you'll need to refer to the "Release Notes" for detailed manual download instructions for the graphics drivers. Here's what the driver install screen looks like:

GPA System Analyzer

The GPA System Analyzer application provides access to system-wide metrics for your game, including the CPU, GPU, Microsoft DirectX*, and the graphics driver. This tool allows you to diagnose at a high level whether your game's performance bottlenecks are concentrated within one or more of these areas, and helps you determine whether additional fine-tuning of your application using Frame Analyzer (if GPU-bound) or other Intel performance optimization products would be helpful. Though we won't be discussing specific optimization techniques here, we'll provide an overview of the basic capabilities of the Frame Analyzer application.

For GPA versions 2.0 and 2.1, the first step is to start the GPA Remote Server on the target machine (by the way, always startup the GPA Remote Server, even if the target and client applications are running on the same system). We'll run it by navigating to from the Start Menu (you'll find it in the "Intel® Graphics Performance Analyzers" folder). Note the IP address (192.168.1.103 in this example), as we'll need the IP address of the target system for the next step. For GPA version 2.2, the key difference is that the GPA Remote Server has been renamed to the GPA Monitor, and you can specify that the GPA Monitor process automatically runs all the time (so that you never have to worry about starting it up manually).

Now we'll launch the GPA System Analyzer on the client machine from the Start menu, and enter the IP address of the target machine and the full pathname to the game on the target system (note that you cannot use the "browse" button across the network for this). For this example, we'll use one of the Windows DirectX* samples (HDRCubeMap), but obviously you would substitute the pathname of your game on the target machine. This is what our screen looks like on the client before selecting "OK" to start the analysis task:

If all goes well, you'll see the HDRCubeMap sample running on the target machine (we'll ensure that it's running in full-screen mode in order to eliminate the overhead of the Windows* Window Manager).

On the client machine you'll see the GPA System Analyzer starting to display metrics gathered from the application:

If all did not go as planned, check the following:

  • Make sure the pathname to the game on the target machine was entered correctly (check and re-enter the correct path). Note that many games have a "launcher" application or script that runs the "real" game, so you'll need to provide the path to the "real" game, not the launcher.
  • Make sure you have the latest graphics driver installed. See the "Release Notes" for driver requirements.
  • For other issues, check the "Release Notes" and/or refer to the GPA site for technical support Forums and Knowledge Base articles)
Now we'll perform some basic GPA System Analyzer operations (as highlighted in the screen shot below):
  1. Drag and drop a metric from the metrics list area to the chart area, replacing the existing chart with the chart for this new metric.
  2. Change the highlighted metric in a chart containing multiple metrics (affecting both the visual attributes of that metric on the chart, as well as the minimum, maximum, and average values that are displayed for that metric).
  3. Select a state override mode to see the effect of this mode on overall rendering time and its impact upon chart metrics.

GPA Frame Analyzer

OK, now let's grab a frame capture of the current view, for processing by the GPA Frame Analyzer application.

  • Press the "capture frame" button in the GPA System Analyzer (located in the bottom left of the System Analyzer window):


  • When the dialog box appears, select the "Launch Frame Analyzer" checkbox, and press OK:


On the target machine, the game will exit after capturing the frame, to allow the GPA Frame Analyzer to utilize the target machine's graphics. On the client machine, the GPA Frame Analyzer will be displayed.

 

Whereas the GPA System Analyzer examines overall system performance characteristics, the GPA Frame Analyzer allows you to drill down into a frame to analyze bottlenecks and try experiments to maximize performance. Once the GPA Frame Analyzer starts, you'll see an image similar to the following displayed on the client system (by the way, you won't see anything happening on the target system, since GPA Frame Analyzer renders to an off-screen window). As shown here, the GUI includes four major user interface areas: the "work item visualization panel" at the top, the "scene overview panel" and the "render target viewer" on the lower half of the window on the left, and the "tabs panel" on the lower right. In this image, I've also selected the work items for the rendering of the teapot.

The "work item visualization panel" at the top displays a bar chart representation of your frame, with the x-axis representing the sequencing of the work items (with the leftmost being the "earliest") and the y-axis representing the actual time spent performing each work item. This way you can quickly see which work items are the most "expensive". To select an item, just click on it, or drag across multiple items to select a contiguous range. Moving to the "scene overview panel", you'll see the work items represented in a hierarchical tree view form, grouped logically by the frame level, region level, and draw call level. Selecting items in this panel is accomplished by clicking on the box on the left-hand side of one or more items. You can also sort the items in this panel, making it easier to then select the topmost items for whatever criteria is most important to you. The "render target viewer" provides visual feedback as to what items have been selected. Finally, the "tabs panel" is used to view various details of the contents of the individual items within the graphics frame, and perform "what if" modifications to the current items that have been selected.

Now's a good time to mention a key point of the GPA Frame Analyzer GUI, namely the current "selection set". When you select one or more items in either the work item visualization panel or the scene overview panel, the results of that selection are immediately synchronized across all four panels of the interface. Furthermore, if you have selected a subset of the work items, performing experiments or other modifications in the "tabs panel" affects only those items in the selection set.

The "tabs panel" is the primary "work-horse" of the application. Here, you can view (and set) all components of your selection set, include the textures, DX state, and shaders associated with these work items. For example, if you select the "state" tab, you can try setting different texture filtering modes to see not only whether the changes result in faster rendering times, but also whether the visual effect of a less costly filtering mode is acceptable. Other similar experiments can be performed on you selection set, such as finding an optimal set of textures that will fit within the GPU, and modifying your shader code to optimize rendering time without sacrificing visual quality. We've also heard from customers that the GPA Frame Analyzer is a great debugging tool for their game prior to even optimizing performance, as the tool can help you examine the sequence of DX calls within each frame.

So a typical work flow might be something like the following. First of all, when you bring up the GPA Frame Analyzer, use the "work item visualization panel" to see where you are spending most of your total rendering time (on a per work item basis), and/or use the "scene overview panel" to examine logical portions of your scene for how long each one takes. Then when you have selected a subset of the work items that you believe may be key targets for optimization, move to the "tabs panel" to perform various "what-if" experiments on your selection set. You can then see whether the changes have made a positive impact by examining the net effects in the rendering time in the "scene overview panel" and/or the "work item visualization panel". In this screen shot we've decided to examine the rendering of the teapot by selecting the "experiments" tab, then selecting "simple pixel shader". We see that the result is an increase in performance of ~44% (which means that the pixel shader portion of the rendering process is taking nearly half of the total rendering time):

 

Conclusion

We hope this Quick Start Guide has you up and running with the tool in a short period of time.

Please read through the online help for details on the rest of the features of the Intel® Graphics Performance Analyzers, and go to the GPA Home Page for support, technical notes, and other items of interest regarding GPA.

 





This article applies to: Game Development,   Intel® Graphics Performance Analyzers (GPA),   Visual Computing,   Intel® Graphics Performance Analyzers Knowledge Base