Pixels for Pennies: SimHQ’s Spring ’08 Midrange GPU Roundup

by John Reynolds



High-end graphics boards have become one of the most sought after components inside a PC enthusiast’s gaming machine, but also one of the most expensive. The GTX, Ultra, or XTX nomenclatures of the graphics market go after the halo effect of the performance crown, which is the general perception that any part belonging to the fastest product line is, by proxy, superior to whatever the competition may be offering. Yet high-end GPUs are generally for those with deep pockets, costing upwards of $400-600 for a single board, and double that for those interested in multi-GPU setups based on them. Fortunately for those of us who haven’t won a lottery or were born into inherited wealth, midrange GPUs costing $150-200 are available, and these cards, while not as fast as performance crown parts, are often as feature-packed and provide a far better bang for the buck.

SimHQ’s GPU roundup today pits two of AMD’s 3000-series boards, the Radeon 3850 and Radeon 3870, against NVIDIA’s GeForce 9600 GT. All three boards now cost under $200, putting them well within the midrange bracket in terms of pricing. All three boards have also been reviewed by numerous media outlets months ago, so we’ll skip any long detailing of the technical specifications and simply use the table below to list the more pertinent details.

Radeon 3850
Radeon 3870
GeForce 9600 GT
Core Speed:
Shader Units:
Shader Units Speed:
Onboard Memory:
Memory Speed:
Memory Interface:
Memory Bandwidth:
ROPs (render back end):

While there are some similarities between the three boards — such as the memory interface width, ROP units and support for multi GPU setups (SLI and CrossFire) — the GeForce 9600 GT, unlike the two Radeons, has a clock domain design, allowing the shader units to run at a much higher frequency compared to the rest of the G94 chip. In contrast, both Radeon 3000 parts have a much higher shader unit count, a whopping 320 compared to 64, though it’s worth noting that these units are not identical in terms of math capabilities so this number can be misleading. The GeForce 9600 GT also has a more powerful back-end of the rasterization pipeline, with more texture filtering throughput and ROPs that retain the traditional capability of resolving color samples for anti-aliasing (this task is performed by the shader units in the Radeon 3000 series). In contrast, the Radeon 3870 also has a fairly significant memory bandwidth advantage over the GeForce 9600 GT, so we’ll see if that impacts the benchmark scores below. Last, all three are DirectX10 GPUs, with the two AMD parts being the first to market to support DX10.1.

Test System Details

All testing was conducted on a clean install of Windows XP Professional SP3. The test system used the following components:

  • Intel E8500 (3.16GHz dual core) CPU
  • ASUS Maximus Formula X38-chipset motherboard
  • 2GB Corsair DDR2-800 memory
  • Creative SoundBlaster X-Fi sound card
  • Western Digital Raptor 150GB hard drive
  • Plextor PX-712S DVD drive
  • Dell 3007WFP 30” LCD

Video drivers used were the Catalyst 8.4s for the 3850 & 3870, and NVIDIA 174.74s for the 9600 GT.

The latest BIOS and chipset and add-in component drivers were installed and all test applications patched. The control panels for both manufacturers’ GPUs were set to application-controlled whenever the test application allowed for control over those settings; otherwise, they were forced via the control panel. All testing was conducted with 4x anti-aliasing and 8x anisotropic filtering enabled. High quality filtering was used throughout all testing. FRAPS 2.9.4 was used to record frame rates for applications that do not produce their own performance results. And the following list of synthetic applications and game titles comprised SimHQ’s benchmark suite for this article.

  • 3Dmark06
  • Crysis
  • Call of Duty 4
  • Rainbow Six: Vegas 2
  • DiRT
  • World in Conflict

Details per application can be found in the test results section below. Readers will notice the lack of hard simulations used for this roundup, a strange scenario for a web site with the domain name of SimHQ. Unfortunately, due to the current market reality that saw damn few simulations — flight or otherwise — released in recent years we’re forced to make use of games we feel are still of some interest to our readers. Titles such as FSX are simply too CPU-bound in their performance for graphics board testing or, like IL-2: 1946, based on an aging graphics engine that simply isn’t challenging to newer GPUs.

SimHQ used the following resolutions for our testing: 1024×768, 1280×1024, 1680×1050, and 1920×1200. This gives a nice range while offering a mix of display ratios, from 4:3 to 16:10. And, again, 4x anti-aliasing and 8x anisotropic filtering were enabled for all tests.

Go To Page 2

Powered by WordPress. Designed by WooThemes