Nvidia and ATI have really shaken up the 3D market in the past couple of months. While the GeForce GTX 260 and GTX 280 hold a number of speed records (simultaneously making the GeForce 9-series look geriatric), AMD’s Radeon HD 4850 and HD 4870 provide good performance at aggressive prices. We have been seeing prices steadily sliding in the past few weeks. The GTX 280 started at $649, and it can now be purchased for just $419.
To find the best performing and least expensive combination, we put the fastest graphics cards through a series of tests consisting of 60 benchmarks, numerous side-by-side comparisons, and a price analysis.
This story seeks to answer a number of questions. First, is it worth changing from a Radeon HD 3850, HD 3870, or GeForce 8800 GTS 512 to one of these new GPUs? Second, how well do the old GeForce 8800 GTX and 8800 GTS 512 cards hold up? Third, which of the current cards is fastest? Fourth, are the drivers for CrossFire and SLI already well-optimized? Fifth, where can you expect increases in performance? And finally, how fast is the card you want at 1280x1024, 1680x1050, and 1920x1200?
If you’re not sure about the right power supply to drive your graphics setup, we have recommendations for power supply sizes. These should be considered guidelines, as PC configurations differ considerably.
The CPU in our test system reaches its limits here—the Core 2 Extreme X6800 at 2.93 GHz is barely capable of keeping up with current dual-card graphics setups. These twin-configs often show little or no improvement in overall 3D performance. Your resolution of choice often determines whether it is worth purchasing a pair of your favorite cards or if an individual board would be better.
The highlights of our comparison are the factory-overclocked models from MSI. As a result of the faster clock speeds, there is additional performance guaranteed by the manufacturer. Special drivers or tools are unnecessary, as the improved specifications are automatically set in each board’s BIOS. To ensure a fair comparison, each graphics card runs at its normal rate, with a detailed analysis of the overclocking given as an additional feature. All overclocked variations are tested at 1920x1200 pixels with anti-aliasing, and the performance differences over standard clock speeds can be compared directly using the frame rates with percent evaluation.
- Taxing Modern CPUs With Powerful Graphics
- Comparing The GPUs And Test Setup
- Radeon HD 4850
- CrossFire With Radeon HD 4850
- Radeon HD 4870 OC
- CrossFire With Radeon HD 4870 OC
- GeForce GTX 260 OC
- SLI With GeForce GTX 260 OC
- GeForce GTX 280 Superclocked
- SLI With GeForce GTX 280 Superclocked
- Assassin’s Creed v1.02
- Call of Duty 4 v1.6
- Crysis v1.21 High Quality
- Crysis v1.21 Very High Quality
- Enemy Territory: Quake Wars v1.4
- Half Life 2: Episode 2
- Mass Effect
- Microsoft Flight Simulator X SP2
- World in Conflict v1.05
- 3DMark06 1280x1024 v1.1.0
- How Overclocking Affected The MSI Cards
- Overall Performance
- Price/Performance Comparison
- How About Graphics Image Quality?
- Power Consumption, Noise, And Temperature
- Frames-Per-Watt For The GTX 200-Series And HD 4800-Series
- GTX 200-Series And HD 4800-Series At 1280x1024
- GTX 200-Series and HD 4800-Series At 1680x1050
- GTX 200-Series And HD 4800-Series at 1920x1200
- All Cards Compared At 1280x1024
- All Cards Compared At 1680x1050
- All Cards Compared At 1920x1200
- Is The Upgrade Worthwhile?
- Swapping Old Chips For New
- Evaluation Of The New Generation
- Conclusions – Radeon HD 4850 Is The Winner

My one complaint? Why use that CPU when you know that the test cards are going to max it out? Why not a quad core OC'ed to 4GHz? It'd give far more meaning to the SLI results. We don't want results that we can duplicate at home, we want results that show what these cards can do. Its a GPU card comparason, not a complain about not having a powerful enough CPU story.
Oh? And please get a native english speaker to give it the once over for spelling and grammar errors, although this one had far less then many articles posted lately.
Remember, the more you know.
but yes the article was off to a great start, maybe throw some vantage in there as well?
Precisely; several other websites tested with 8.7 and 8.8 long before this article was published. Why couldn't you? Look at the 8.6 release notes; it doesn't even mention the HD4000 series cards as supported devices.
Brilliant guys.
This is another reason why the results are tanked, in XP you get 15% more performance compared to these values
After having the Mythbusters appear, you would think this would be the most comprehensive, "scientific," factual, and update article meeting Tom's usual standards.... I didn't finish reading this.
(82 temperature in 2D 69 in 3D with no fanfix)
Also, I can understand why TH didn't have time to use 8.8 since it was released publicly on August 20, 2008 (Although ATI would have gladly released a beta version to TH for testing purposes).
However, AMD publicly released stable Catalyst 8.7(internal version 8.512) on July 21, 2008. That's more than a month ago. It has numerous improvements (for example, CF performance increase, improved stability and performance under Vista). To be honest, most of the improvements range from 4% to 15%. (In CF case, up to 1.7 X scaling)
TH has rarely been unfair and/or inaccurate and they always owned up to their mistakes before, and I trust them to re-test ATI products with at least 8.7 if not 8.8 to continue to uphold their values and integrity.
Now on to my criticism.
I can understand how you want to keep the results homogeneous with previous results but if you already know that a stock QX6800 will bottleneck the system, be proactive in fixing it. At the very least you should have done a small segment of the review showing the newer cards with a quad core overclocked to 4.0Ghz.
Also, if you have ever read any of the older Toms articles, you would know that you can still minimise the bottleneck from a slow GPU bye raising the resolution. Perhaps you should test the fastest cards at the highest resolutions?
I can also understand why you did not use the latest nVidia drivers. It takes time to create a review of this scale and the GF8/9 series drivers have been stable for some time. As the GT 200 series brings no new features to the table, they would needed little optimisation for their newer cards allowing the slightly dated drivers to perform nicely.
What I can not understand is why you would use ATI's 8.6 drivers??
The 8.7 drivers have been out for more than a month bringing quite a few fixes/optimisations with it. I understand it probably took more than 9 days to complete all of these benchmarks (today is the 29th, the 8.8 drivers were officially released on the 20th) but you should have called ATI and asked for their latest drivers. The 8.8 drivers were leaked at least a week before the official release which means, if you could nurture a relationship with the people you review, they could/probably would have provided them to you. There is still no excuse I can see for testing with the old 8.6 drivers. Seriously, it does not even have official support for the 48X0 cards...
From the title of the article,"The Fastest 3D Cards Go Head-To-Head", I would have assumed that you would have been testing the Fastest 3D cards? What happened to your 4870x2? As you have already attempted to review it, we know you have your hands on one. How can you claim to review the "Fastest 3D Cards" and still leave out the fastest card?
In summation, I liked many things from this article. The layout was nice and a little more technical than we have been seeing as of late. I enjoyed the comparison charts at the end and I think you should adopt a similar method for the CPU and GPU charts. I would have thought this was an excellent and well thought out article if it had not been for the glaring and obvious deficiencies in reason. I give you credit for stepping Toms in the right direction. With a little more unbiased comparison, critical thinking and common sense I could come to see reviews such as this in a very positive light.