Test Setup
We have two "similar" systems that we used to test AC, one using ATI graphics and one using NVIDIA graphics. It's important to note that these are not identical systems, as the hardware we had on hand is limited. Specifically, SLI support requires an NVIDIA chipset and CrossFire support requires an Intel or AMD chipset. Our NVIDIA testbed comes courtesy of Dell, their midrange XPS 630 that uses the nForce 650i chipset. Our ATI testbed is the same X38 platform we have used in previous gaming articles. It includes more memory (which doesn't affect performance) rated at DDR2-800 and Windows Vista 64-bit. Thus, it is worth noting that we are not comparing apples-to-apples... or at best, we're comparing Granny Smiths with Fuji apples.
| Custom X38 Test System | |
| Processor | Core 2 Quad Q6600 (2.40GHz 2x4MB
cache) Overclocked to 3.00GHz (QX6850) Overclocked to 3.42GHz (1520FSB) |
| Motherboard | Gigabyte GA-X38-DQ6 |
| Memory | 2x2048MB OCZ DDR2-800 Running at DDR2-800 4-4-4-12 |
| Graphics | 2 x AMD Radeon HD 3870 (CrossFire) |
| Hard Drive | Samsung F1 750GB (7200RPM 32MB) |
| Operating System | Windows Vista Ultimate 64-bit |
| . | |
| Dell XPS 630 Test System | |
| Processor | Core 2 Quad Q6600 (2.40GHz 2x4MB
cache) Overclocked to 3.00GHz (QX6850) |
| Motherboard | Dell nForce 650i |
| Memory | 2x1024MB DDR2-667 Running at DDR2-667 5-5-5-15 |
| Graphics | 2 x GeForce 8800 GT 512MB (SLI) |
| Hard Drive | Seagate Barracude 7200.10 500GB (7200RPM 16MB) |
| Operating System | Windows Vista Home Premium 32-bit |
| . | |
In the processor department, the two systems are identical, sporting Intel's Q6600 revision G0 quad-core processor. Both systems also support overclocking, and we will investigate how that affects performance. The Dell system could only run reliably at around 3.0GHz, so we chose that frequency as a comparison point, simulating a QX6850 CPU (9x333MHz on a 1333FSB). Finally, we tested in both single and dual-GPU configurations on both systems. We're not using this to come to a strict conclusion on whether ATI or NVIDIA graphics are better for running AC, but rather to get a general idea of what sort of hardware is required to run the game well.
Before we get to the actual benchmark results, we want to define some of the settings we'll be using. For benchmarking purposes, we tested at Medium, High, and 4xAA. Note that we also left Level of Detail at maximum for the benchmarks, and the crowd density was likewise set to maximum. Medium drops the Shadows and Graphic Detail settings down one notch. Additional testing of lower detail settings can be found further in the article.
Note: if you're not interested in performance testing, you may want to skip to the conclusion.
Facebook
Twitter
RSS
32 Comments
View All Comments
bill3 - Monday, June 2, 2008 - link
Actually it's terrible, I cant read the graphs AT ALL.seriously my eyes just glazed over those terrible charts..completely unreadable. I still, have no idea what I'm looking at. Is ATI supposed to be faster in this game? Why did they test with version 1.00 on ATI and 1.2 on Nvidia? I dont know because the graphs are totally useless.
Nihility - Monday, June 2, 2008 - link
I second that. The graphs are terrible. Maybe bar graphs would have been better?Sometimes when you're the one making the graph it's hard to imagine what other people are seeing when they look at them. I suggest having another pair of eyes check the graphs out for readability.
Besides that, I loved the review. Especially the performance part and the 10.1 controversy.
JarredWalton - Tuesday, June 3, 2008 - link
Charts are colored with similar colors used either for ATI vs. NVIDIA, 1.00 vs. 1.02, or dual-GPU vs. single-GPU. I could have generated four times as many graphs to show the same data, but I figure most people are capable of reading the labels on a chart and figuring out what they mean. Here's a hint: when you can't see the difference between two lines because they overlap, it's a tie.If you want to give specific examples and recommendations on what would look better and still convey the same amount of information, I'm all ears. However, simply stating that "the graphs are terrible" does little to help. Tell me what graph specifically is terrible, and tell me why it's terrible.
As an example of why I used these graphs, page 9 has two charts showing 40 total data points. You can get a clear idea of how performance scales with single or dual GPUs at the various detail settings looking at a single chart. Green is NVIDIA, Red is ATI. That makes a lot of sense to me. Creating ten different bar charts with four lines in each to show the same data makes it more difficult to compare how Medium graphics compares to High graphics performance, and it takes up five times as much space to tell the same "story".
Page 6 is the same thing, but with green used for dual-GPUs (light and dark for 1.00 and 1.02) and red for single GPUs. 24 data points in two charts instead of using six charts. Having established that 1.00 doesn't perform any different than 1.02 on NVIDIA hardware, I skipped the 1.00 NVIDIA numbers to make those charts easier to read on page 7. Then I put in the four standard test system (0xAA and 4xAA, ATI and NVIDIA) on 1.02, with 1.00 4xAA ATI in blue as a reference.
Lastly, on page 8 I have two clock speeds on NVIDIA, three on ATI, with different base colors for single and dual GPUs. ATI and NVIDIA are in separate charts, and brighter colors are for a higher overclock.
There's method to my graphing madness. Are the charts immediately clear to a casual glance? No, but then that's really difficult to do while still conveying all of the information. I spent a lot of time trying to make comprehensible charts, and settled on these as the best option I could come up with. Again, if they're so bad, it must be easy to generate something clearly better - have at it, and I'll be happy to use any sensible suggestions. However, if the only complaint is that you actually have to look at the charts and think for a minute before you understand, I'm not likely to be very sympathetic. I think our readers are smart enough to digest these graphs.
mpjesse - Monday, June 2, 2008 - link
While I appreciate the detailed review, isn't it a little irrelevant now? I mean, the game's been out for nearly 2 months now and it's been reviewed everywhere. The only thing new about this review are the performance benchmarks, in which case I would have have made the review solely about performance instead of gameplay.Just my 2 cents.
ImmortalZ - Monday, June 2, 2008 - link
Its sad that the companies with money always manage to suppress innovation.I hope this article by AT will raise some ruckus in the collective Interwebs and cause something. But I doubt it.
ViRGE - Monday, June 2, 2008 - link
For what it's worth, another forum I read had some screenshots comparing DX10 and DX10.1. The problems the poster had managed to find involved trees; there was some kind of post-processing rendering going on with trees that wasn't occurring with DX10.1, which made them look weird.Not fixing 10.1 may be an NVIDIA thing, but there was definitely a problem with it as-is.
tuteja1986 - Monday, June 2, 2008 - link
Well why where the hell is nvidia dx10.1 support if dx10.1 actually brings some kind of performance improvement in AA.Why aren't GT200 series have DX10.1 ?
I thought PC gaming was all about being the cutting edge on all technology front...
Anyways , this is not the 1st time Ubisoft or Nvidia have done this.
wyemarn - Monday, June 2, 2008 - link
Maybe because Nvidia GPUs cant support AA through shaders. So no use supporting dx 10.1. ATI GPUs have 320 stream processors so it can utilize for shaders and etc. Nvidia cards have less SPs but more ROPs, TMUs which translates to more brute power if games dont use shaders or SPs much. Technology wise, I think ATI is ahead but NVIDIA GPUs have game developer support and more raw horsepower so performance wise NVIDIA is ahead and I think this trend will continue with GTX200 series. I choosed G92 over RV670 because the raw performance is much better even though on paper HD 3800 series look great.SteelSix - Monday, June 2, 2008 - link
Worthy of a thread in Video. I just started one..Gannon - Monday, June 2, 2008 - link
The original halo had performance issues but they weren't alarming, halo was actually not too bad port compared to many other console to PC disasters. Halo 1 got 'better with hardware' advancing. Halo 2 on the other hand is just all around atrocious. Halo 2 was just not a very well made game, period, despite the addition of cutscenes, etc. Halo 1 had a much better feel and better vehicle design IMHO, I hated how the warthog looked in Halo 2, it annoyed me to no end.