In this set of results, we expand our graphics card charts with a supplementary comparison. This addition covers more mainstream graphics cards and compares how they fare when the graphics-quality slider is set a bit lower in a number of popular games.
The previous charts, which cover the latest and greatest GPUs with the graphics-quality settings set as high as possible, also get some new entries. With this update, our charts allow you to compare 68 different graphics chipsets, retail cards, and reference cards online.
You can now peruse precise performance results based on over 30 gaming benchmarks. The current basis for comparison will be updated monthly to include new graphics card reference models across all categories. In this new version, we cover additional benchmarks, results for when the graphics-quality slider is set at low and high in games, and more actual products.
In order to offer the best possible overview, we first measure performance for all graphics chips at standard clock rates. These charts are not, however, limited just to reference models from ATI and Nvidia. You can also compare results for off-the-shelf retail products, including standard offerings that match reference specs, plus cards with extra RAM, overclocked memory and GPUs, additional display connectivity, and other unique features.
This comprehensive approach lets us track the market more closely, because many vendors alter performance, cooling, and card designs in their commercial offerings.
Here are the links to our graphics charts, so you can investigate our findings or compare graphics cards to one another:
The benchmark suite that we're using draws from a broad palette of different games and 3D engines. Our goal is to create a workable mix of real-time strategy (RTS) games, simulations, role-playing games (RPGs), and 3D shooters, and to make sure we cover all of the most important 3D engines in our tests, including the Gamebryo, Source, and Unreal 3 engines. For OpenGL benchmarks, we were only able to find a new game from id Software after we had already completed our tests.
| Benchmark Suite For 2009 | API | 3D Engine |
|---|---|---|
Fallout 3 | DX9 | Gamebryo (Oblivion) |
Far Cry 2 | DX10 | Dunia Engine |
F.E.A.R. 2 | DX10 | LithTech Jupiter Extended (EX) |
Left 4 Dead | DX9 | Source Engine (Half-Life 2) |
The Last Remnant | DX10 | Unreal 3 |
Tom Clancy's EndWar | DX10 | Unreal 3.1 |
Tom Clancy's H.A.W.X | DX10 | Ubisoft (Blazing Angels) |
3DMark06 v1.1.0 | DX9 | Futuremark |
We divide graphics cards into two quality categories: High and Low Settings (highest and lowest graphics quality). Cards fall into one or the other of these categories depending on their overall performance characteristics. This new breakdown improves the comparability of graphics chips by design. The High Settings category is for powerful graphics chipsets. Because these are always improving and getting faster, everything remains open to new challengers at the top of our rankings.
More mainstream cards suffer when graphics sliders or resolutions are set too high in games, as these offers are quickly overwhelmed, resulting in unplayable frame rates. Because it’s simply impractical to play games at the demanding settings we use on the faster boards, we extended our charts to include lower resolutions and lower graphics quality (in the Low Settings category).
By and large, older graphics chipsets fall into the low-end category (for instance, the GeForce 7-, Radeon X1000-, or HD 2000-series cards). They typically don’t support DirectX 10 and may not work at all or be too painfully slow to use with high settings. Model numbers for graphics cards are often helpful when assigning them to a category. Anything with a number less than x600 (for example, GeForce 8400 GS, GeForce 9500 GT, Radeon HD 4550, or Radeon HD 4350) won’t perform well enough to support higher graphics quality and larger resolutions.
High Settings usually offer the upper range of graphics quality that a game supports. We ran our tests for these charts with 4x anti-aliasing (AA) and 8x anisotropic filtering (AF). If a game permitted it, we used 8x AA and 16x AF. These settings target ever-higher graphics performance, which keeps increasing over time thanks to faster GPUs, plus SLI and CrossFire configurations. That’s why our scale remains “open at the top.”
By contrast, our Low Settings category targets weaker graphics boards. Graphics quality settings will be set at or near their lowest levels, and we use the DirectX 9 API as a baseline. That’s about as low as gaming graphics can go. For some time now, modern games have accommodated lower-end cards, so that even slow graphics cards produced reasonable frame rates. Those who want higher performance must reduce display resolution or tweak graphics driver settings. Our highest level of testing here is 4x AA and 8x AF. We also tested with AA and AF turned off.
| High Settings | Low Settings |
|---|---|
Fallout 3
Far Cry 2
F.E.A.R. 2
Left 4 Dead
The Last Remnant
Tom Clancy’s EndWar
Tom Clancy’s H.A.W.X
3DMark06
Cumulative Frame Rate High Settings
1680x1050 4x AA High Settings
1920x1200 No AA High Settings
1920x1200 4x AA High Settings
1920x1200 8x AA High Settings
| Fallout 3
Far Cry 2
F.E.A.R. 2
Left 4 Dead
The Last Remnant
Tom Clancy’s EndWar
Tom Clancy’s H.A.W.X
3DMark06
Cumulative Frame Rate Low Settings
1280x1024 No AA Low Settings
1280x1024 4x AA Low Settings
1680x1050 No AA Low Settings
1680x1050 4x AA Low Settings
|
Notes about CPU performance: CPUs nearly always slow high-end graphics cards down. The lower the display resolution and the lower the graphics quality, the more impact a CPU has on maximum frame rates. To minimize this effect, an overclocked CPU, a high resolution, or very high graphics quality settings make good sense. For cards in our High Settings tests, the lowest resolution was 1680x1050 with AA turned on. This means that graphics cards must perform well just to work okay. At a 1920x1200 resolution with AA disabled, games with DirectX 9 graphics engines will be more limited by CPU performance rather than actual graphics performance.
For the Low Settings category, an overclocked CPU is a must-have because lower graphics quality and resolution don’t do much for better graphics chipsets, and maximum frame rates are therefore highly dependent on CPU performance. In any case, this kind of test only makes sense for those who seek to measure unadulterated graphics card performance. In practice this is illogical, because nobody’s going to equip an overclocked $1,000 PC with a $50 graphics card to play 3D games.
To enable comparisons for the High and Low Settings tests, we tested a batch of graphics cards for both categories. We chose the Radeon HD 3850 and HD 4670, along with the GeForce 8600 GTS and 8800 GTS 320 models for this head-to-head. While the Radeon HD 4670 works well with DirectX 10 and our higher-end settings, the GeForce 8800 GTS proves something of a problem child. With a game's graphics quality slider set to high, this card’s 320 MB of RAM is improperly optimized, very slow, and unusable at 8x AA. When Low Settings are in effect, this card is much faster, if not under-utilized in light of significant CPU limitations.
Our small test table shows total results in frames per second (FPS) and should shed some light on differences and distinctions in performance. The GeForce 8600 GTS works well at lower graphics quality settings, but when asked to deliver higher graphics quality in a DirectX 10 environment, it just barely limps along. This also applies to the Radeon HD 3850, which is designed more for use with DirectX 9 and somewhat lower graphics quality settings. The Radeon HD 4670 functions well in both categories and is noticeably stronger and more capable as graphics quality increases. In fact, it’s surpassed only in the Low Settings tests by the GeForce 8800 GTS 320 MB.
| Overall Performance Results | Low Settings | High Settings |
|---|---|---|
GeForce 8800 GTS (320 MB) | 3,247.8 FPS | 533.1 FPS |
Radeon HD 4670 (512 MB) | 2,458.6 FPS | 608.5 FPS |
Radeon HD 3850 (512 MB) | 2,281.1 FPS | 454.7 FPS |
GeForce 8600 GTS (512 MB) | 1,737.8 FPS | 384.9 FPS |
In the real world, you’d probably boost the graphics quality slider somewhat for the GeForce 8800 GTS 320 MB. For our charts, we were forced to use a setting with medium graphics quality, which didn’t really pay off because it makes the basis for comparison with other chipsets questionable. We also include 3DMark06 tests in our suite, which are independent of graphics chipset speed because we always use the default graphics quality at a 1280x1024 resolution. This too provides a direct basis for comparison across all tests.
Here, we provide a precise description of the individual benchmarks and settings for the different graphics chipsets, including vendor tweaks and information about SLI or CrossFire compatibility.
Fallout 3
This game is a mix of an RPG and FPS title, with visuals that come from the improved Oblivion graphics engine. Fallout 3 supports HDR rendering using Shader 3.0. If you want to activate surface textures and AA at the same time, you must use a high-end graphics card, because the older GeForce 7-series cards can only do one of these at a time. This graphics engine renders outdoor scenes very fluidly and interiors with short horizons pose no difficulties for capable graphics chips.
Those who want to bring their graphics cards to their knees can integrate add-ons for improved texture handling from mod Web sites (such as Fallout 3 Nexus). These will boost graphics quality to nearly photorealistic levels.
To maximize game compatibility, we ran our tests using the standard version of Fallout 3. We used a scene from Tenpenny Tower to measure frame rates with FRAPS during an outdoors scene, with a distant horizon and numerous objects and ruins in the visible landscape.
The Oblivion 3D engine supports SLI and CrossFire well. For our High Settings tests, we set the graphics quality slider to Very High (the maximum usable setting) where 8x AA was the highest value available for AA. For our Low Settings tests, we set the graphics quality to Low and deactivated HDR rendering, which enabled older graphics chips to use AA without difficulty.
Far Cry 2
Far Cry 2 is a 3D shooter in the grand tradition of Crysis. Its Dunia 3D engine displays very nice DirectX 10 effects, particularly when rendering fire, shadows, water, and background vegetation, and streaming sunlight through dust, fog, and so forth. We set graphics quality settings at Very High for the high-end tests, because the maximum Ultra bogs even the best of graphics cards down too much, which is why we elected not to use it. For our Low Settings tests, we set the graphics to Low in DirectX 9 mode, with HDR rendering disabled.
We ran the Ranch Small benchmark sequence for our tests, which shows people, a sweeping plain, burning grass, and various huts in great profusion. SLI worked well, but CrossFire ran even better, which made more graphics RAM and higher AA at 1920x1200 resolutions pretty important. Our highest test setting for AA was 8x.
F.E.A.R. 2
This game is a horror-themed 3D shooter. Once Warner took over this label, the game was polished with professional Hollywood shock effects of all kinds. The interiors and short sequences run very smoothly and produce very high frame rates. SLI and CrossFire both work superbly, while the performance boost from adding a second card is stupendous. Overall, graphics quality is superior and dream or vision scenes benefit from beautiful shader effects on screen. We used the Maximum graphics quality setting for our High Settings tests. For the Low Settings tests, all options were set to Minimum and the graphics enhancements and HDR rendering were turned off inside the game. We used the elevator scene in the Mission Ruin for our FRAPS measurements.
Left 4 Dead
In Left 4 Dead’s 3D shooter game world, protagonists bite, claw, and blast their way through hordes of zombies. The game runs on an enhanced version of the Source Engine from Half-Life 2. Owing to very good support for multi-core processors and a modest appetite for 3D effects, gamers are virtually guaranteed completely fluid frame rates.
For our high-end tests, graphics quality was set to Very High, which guarantees the best possible graphics quality. The highest test setting for AA was 8x AA. SLI and CrossFire enjoy superb support, where CrossFire delivers higher frame rates for Half-Life 2 but falls off slightly for this game. We used a variety of timedemo items for our frame rate tests, where the group conducts a running street battle with countless zombies.
The Last Remnant
This role-playing game follows a more leisurely tempo because its battles are conducted in predictable rounds, much like those in Final Fantasy. Visuals come from the Unreal 3 engine, which keeps gaining more DirectX 10 content throughout. Because you can’t turn AF on for this game manually and the Nvidia drivers can’t deliver higher settings at all, we used the standard settings, which are locked at 4x AF. We used a real battle for the test sequence that we measured with FRAPS. Because game action and encounters varied, we averaged two different runs for our readings. Without AA turned on, mainstream graphics cards deliver decent frame rates, while higher-end GPUs deliver absolutely fluid action throughout.
When set to High, graphics quality settings are maxed out for the game, so that’s what we used for our High Settings tests. The graphics engine works very well with SLI, and CrossFire has finally been optimized to deliver three times more performance. This turns 20 into 60 FPS, and 15 into 45 FPS. Performance for the Radeon HD 4870 has also been improved in Catalyst 9.6. Alas, the Radeon HD 4670 is still hampered, so we’re waiting on another new driver version. It’s not unusual for games that use the Unreal 3 engine to require stepwise optimization, while in our experience, it takes one or two driver versions before they get things right .
Tom Clancy’s EndWar
EndWar presents the RTS game World in Conflict. It uses an enhanced Unreal 3 engine that looks very good on the screen. This game really isn’t ideal for benchmarking, because its frame rate is capped at 30 FPS by a software limiter. This is typical for most recent RTS games, for which settings options for benchmarking are limited to a narrow range.
Nevertheless, we observed that it was possible to decrease frame rates below 30 FPS in the Replay Kopenhagen scene. We could only use the 1920x1200 resolution without AA, while the 3D engine and the fastest graphics cards all had enough headroom to hit the 30 FPS limit, which produced identical results for all contenders.
Borderline cards in this category include the GeForce 9800 GTS+ and the Radeon HD 4870, both of which achieved frame rates of 29.5 FPS (or 30 FPS when rounded up). Any faster cards were clipped to 30 FPS, although they probably could have delivered at least a few frames per second more.
When AF was turned on our measurements worked better, because our replay could even tax the most powerful graphics cards more heavily. The top-end graphics chip classes all hit the 30 FPS ceiling more often. At the bottom of the range, results are less ambiguous. If a graphics card lacked sufficient power, a difference of 10 FPS meant a 30% decrease in performance. The High setting was as high as we could use for graphics quality; SLI was well-supported; and CrossFire appeared to deliver no advantage.
Tom Clancy’s H.A.W.X
H.A.W.X. will never challenge Flight Simulator X for realism, but this new flight simulator delivers very pretty DirectX 10 graphics and hectic dogfights. Thanks to automatic image stabilization, daredevil flight maneuvers are rendered perfectly and contribute mightily to the excitement of the game. In our tests, DirectX 10 crashed at 8x AA and the game and screen went black. Switch to DirectX 9 instead, and the game works at 8x AA, with frame rates up to 50% higher.
The graphics differences are huge; the sunlight effects are sharply reduced and the haze over landscapes and cities is missing. For our tests, we used DirectX 10 and the High setting to achieve maximum graphics quality. In our new Low Settings tests, we switched to DirectX 9 mode and turned off HDR rendering. This produced good results even from otherwise slow graphics cards.
We used the test sequence Mission: Glass Hammer over Rio to measure frame rate. Frame rates were good overall, but AA reduces 3D performance by as much as 50%. It’s not unusual for this game to stutter or hiccup when running on a single graphics card (or GPU), even though frame rates consistently show over 35 FPS. When AA is enabled, this effect is particularly noticeable with ATI graphics cards. The performance boost from adding a second graphics card, either in SLI or CrossFire, is astonishing.
3DMark06
We continue to use this synthetic DirectX 9 benchmark, primarily because 3DMark Vantage doesn’t always finish error-free and because it works only with DirectX 10 graphics cards. Thus, 3DMark06 remains relevant to our 3D charts and different generations of graphics cards, particularly older ones. In the meantime, it has become more of a diagnostic tool that we can use to compare CPU performance and 3D graphics values, to help us track down issues or problems with overclocking, SLI, or CrossFire. That said, its results when comparing different graphics cards to each other aren’t always meaningful, because 3D performance in real games is often much lower than what 3DMark06 reports.
