The Big GPU Shootout: Part I, Upgrade Now or Wait?

Test Configuration

We are testing at 1650×1050 and 1920×1600 resolutions. These are probably the most common resolutions that most gamers would be upgrading their upper-midrange and top cards from the last generation or so for. We decided to test 2900XT, which is largely unsurpassed by 3800 series – and is also the near-equivalent of the 8800-GTS series – as representative of upper-midrange cards of that generation. We also picked the venerable 8800-GTX, which has been a top choice standard for almost 2 years and mostly unsurpassed by Nvidia’s offerings, except for the higher-clocked Ultra, until Tesla architecture launched a few months ago. We took some of the very latest games and we benchmark them for you at the most demanding in-game settings, with Vista 32 and in DX-10 whenever possible with 4xAA/16xAF. These settings are probably the most commonly chosen and difficult settings for new games that a video card will run. The games we are testing include, Crysis, Call of Juarez, F.E.A.R., STALKER, Half-Life2, Enemy Territory: Quake Wars (ET:QW), Unreal Tournament 3 (UT3), Lost Planet, GRID, FarCry2, PT Boats DX10 demo, as well as the synthetic benchmarks, 3DMark06, Vantage and Fur.

We’d like to give our opinion here about benchmarking for this article. We prefer to use built-in benchmarks whenever possible. As a second choice, we use custom recorded time-demos and FRAPS. We find that ‘real world’ testing has too many variables to truly paint an accurate picture of differences between closely competing cards, and especially when performance is sometimes only a frame rate or two different between 2 closely competing video cards. We also found that non-repeatable real world testing may skew conclusions. We will let you know when we do this kind of testing and exactly how we set it up. Of course we feel that the video card manufacturers will not ignore the built-in benchmarks or established recorded time-demos and will of course make sure that they run decently on their respective hardware. So you may find that these benchmarks run slightly better than your own in-game experience might. Take this into consideration, but also remember that the other cards will run well on these benchmarks also; you will get a very good comparison of relative performance for each game tested. Finally, we also use synthetic benchmarks, such as Futuremark’s 3-D Mark 06 and Vantage, as well as the OpenGL Fur benchmark. For the later Futuremark benches, pay particular attention to the two mini-games that are benchmarked within these synthetic benchmarks – where the actual frame rates can be compared. If you take all of these benchmarks together, then you will get a pretty good and accurate picture of the current state of the performance of each of the video cards as they relate to each other. Each card has been tested for weeks and you will also get impressions of relative performance beyond benchmarking to actual game play – which is subjective and will be clearly stated as personal opinion or preference.

We will also be testing crossfire combinations which include crossfire X-3 = 4870×2 + 4870-512MB vs. crossfire X-3 = 4870×2 + 4870-1GB, to see what penalty – if any – Crossfire X3 (CFx3) experiences when limited to 512MB. We will also be testing our Gigabyte crossfire P-35 motherboard’s performance – 1.0 PCIe specification and 16x+4x PCIe crossfire slots – vs. ASUS p35e deluxe x48 Mother board performance with its 2.0 PCIe specification and 16x+16x PCIe crossfire slots. We also compare Intel’s core2duo, e8600 at 3.33Ghz and also at 4 GHz, with each of our top cards, motherboard and crossfire combinations to look a little bit at CPU “bottlenecking”.

So let’s begin with our first round of tests. We are using Catalyst 8-8 and GeForce 177.41 – final certified drivers are used for our testing all through this first part of this review. Identical 250GB hard drives are set up with the latest version of Vista32-SP1; each with identical programs, updates and patches – the only difference are the video cards. We used a Gigabyte P35 Crossfire MB – 1.0 PCIe spec and CF 16x + 4x for this Part One. The testing hardware is detailed in the chart:

Test Configuration- Hardware

• Intel Core 2 Duo E8600 (reference 3.33 GHz ).
• Gigabyte P35-DS3P (Intel P35 chipset, latest BIOS).
• 4 GB DDR2-800 RAM (4×1 GB, dual-channel).
• Nvidia GeForce GTX280 (896 MB, nVidia reference clocks) by BFGTech
• Nvidia 8800-GTX (768MB, reference clocks)
• ATi Radeon 4750 (512 MB, reference clocks) by Sapphire
• ATi Radeon 4750×2 (2 GB, reference clocks) by VisionTek
• HD2900xt-512MB, (reference clocks) by VisionTek
• Onboard RealTek audio
• 2 Seagate Barracuda 7200.10 Hard Drives [setup identically, except for the graphics cards]
• 850 watt OCZ power supply

Test Configuration – Software

• ATi Catalyst 8.8, highest quality mip-mapping set in the driver.
• nVidia Geforce 177.41, high quality driver setting, all optimizations off, LOD clamp enabled.
• Windows Vista 32-bit, SP1; very latest updates
• DirectX August 2008.
• All games patched to their latest versions.

Test Configuration – Settings

• as noted, vsync off in the driver to “application decide” and never in game.
• AA only enabled in-game; all settings at maximum16xAF applied in game or in CP [except as noted]
• All results show average, minimum and maximum framerates
• Highest quality sound (stereo) used in all games.
• Vista32, all DX10 titles were run under DX10 render paths.