Skip to main content

Part 2: 2D, Acceleration, And Windows: Aren't All Graphics Cards Equal?

Tom2D: Our Simple 2D GDI Benchmark

At this point we like to present our simple 2D benchmark program to our readers. It creates GDI commands as output, enabling the measurements discussed in the following pages.

Important note: the results we are about to present do not correspond to the results from Part 1. This is especially true for drawing lines, because we’ve made various changes to the benchmark since we first wrote about it.

User interface for our small test program (devoid of results)

System requirements:

  • Windows 2000 or better
  • 2MB disk space available
  • Screen resolution of 1024 x 768 or higher

This program doesn't need to be installed and is highly portable from system to system.All measurements are extremely CPU-dependent! Also, running other programs and services can alter these results. Measurements will also fluctuate slightly from run to run.

Test Setup

For this round of testing, we completely cleaned up a desktop machine, then created a new user account. We installed no gadgets, and we avoided opening any additional graphics windows on the test desktop. Once started, we associated the Tom2D.exe program with a single CPU core. We also disabled or deactivated all unnecessary background programs and services as well.

Testing Modes

We selected either direct drawing to the display device or DIB buffering  as shown in the next screenshot (for more information on these topics, consult the two preceding sections in this story).

Changing program modes for Tom2D.exe

Test Coverage

The Tom2D.exe program tests various GDI drawing commands one at a time, so it can accumulate data acros all tests. By inspecting the results and comparing them to others, you can readily see where bottlenecks may exist.

Running individual tests

How-to: Complete Test Run

Click the “Run complete benchmark” button to run all tests automatically and calculate a cumulative result. You can also use the “Copy to clipboard” button to grab and save test results in text form, if you're looking to import them into your favorite number-crunching tool.

In the pages that follow, we’ll present each of the tests individually and explain how they work. We’ll also share our own test results for each item along the way as well.

Our Test Systems
ProcessorIntel Core 2 Quad Q6600, 2.4 GHz @ 3.2 GHz, G0 Stepping, 8MB L2 Cache, LGA 775
RAM8GB DDR2-1066 CL5
MotherboardDFI LANParty DK X48 T2RS
Operating SystemWindows 7 Ultimate x64
Graphics CardsRadeon HD 5870, GeForce GTX 285
Graphics DriversCatalyst 9.12, ForceWare 195.62
Other Graphics CardsRadeon HD 5750, Radeon HD 4870, Radeon HD 3650GeForce GTX 285, GeForce 8800 GTS 512MB, GeForce 8400 GS, GeForce 6800 GT 256MB
Onboard GraphicsGeForce 7050 (nForce 610i), GeForce 8100 (nForce 730i) Intel G45 (+Intel Pentium E5200, 4GB), Intel GMA X4500HD (+Core 2 Duo P8700 (2.53 GHz), 4GB), Intel GMA 950 (+Intel Atom, Windows XP Professional)
Retro PCsPentium MMX (P55C) 233 MHzTyan Tomcat IVD S1564D(Intel 430HX)  224MB EDO DRAM3dfx Voodoo4 4500, 166MHz GPU (VSA100) 32MB PCI (Windows 98SE)

In addition to the platforms listed in the preceding table, our readers tested lots of other combinations with Windows 98, Vista, and XP. All values collected will be presented in a comprehensive summary results table later in this story. To those who note that results for the 780G/785G on-board graphics chipsets are missing, we’re still waiting for new drivers for those items to appear, and will be sure to add those results once they become available. We don’t want to combine our results with those from other sources to help us maintain our objectivity.

Many Thanks to One and All!

At this point, it's also appropriate to thank the many readers who ran our benchmark on their systems and shared those results with us. In the meantime, all of these values (we got over 800 sets of individual results in all) have been ranked, accumulated, and compared to our own results to create the best possible ranking scheme we could devise. Most of these reports match our own observations fairly closely, with only a few exceptions here and there. This helps improve our confidence level when is comes to judging the fairness of our test regime.