Six budget graphics cards fail in DX10 shoot-out

10 Dec 2007 | 15:31 GMT

By Theo Valich

Graphical Group Test ATI Radeon HD 3850, HIS IceQ 2600 XT, PowerColor 2600 XT, XFX 8600 GTS, Zotac 8600 GT, 8500 GT, 8400 GS

IF YOU ARE contemplating what to buy for Christmas and have 100-150 USD/EUR to spend, on a graphics card, should you really take the plunge and buy anything less than than a Radeon HD3850 or 8800GT 256MB?

We took six graphics cards from this low-end of the spectrum and put them up against the new mainstream king, the ATI Radeon HD 3850.

The middle and low market segments of graphics are where the money is made. So, we have an army that buys graphic cards just to play movies and browse the web and sometimes finds the time to play some games. Usually these users don't strive to keep up with the latest trends, but they'll hear from someone that a stunning new Crysis or a new UT is out, and they try to check it out. Should they even try? This is a text for those who do not hear the call of geekdom.

So, do cheap GPUs really provide what an average user would expect from them?

Whenever you go in a shop,you'll be bombarded with ads for these cheapola cards with a great price, while shiny stickers will proclaim their prowess in the world of "HD" (HD without inputs...). As spinners will tell in off-the-record conversations, this boards are created to give good video experience, and give a just a "taster" of DirectX 10.

The contestants

We tried to acquire 2400Pro and XT, but sadly we could not find the product available for testing. Expect even lower performance than this one, somewhere in the range between 8400GS and 8500GT.

We were also unable to get the 8800GT 256MB board, so the 3850 could have some competition. However, we were told by partners that they aren't sampling 8800GT 256MB for reviews, because there is a shortage of ASIC and every chip they get is going into the channel. While we have no doubt in performance of the part, it remains untested here, thus we can't reccommend it.

These are the cards we tested:

ATI Radeon HD 3850
HIS IceQ 2600 XT
PowerColor 2600 XT
XFX 8600 GTS
Zotac 8600 GT
Zotac 8500 GT
Zotac 8400 GS

We updated our test setup with a Gigabyte P35 motherboard after our sixth Bad Axe 2 kicked the bucket. We had the misfortune of getting all six D975XBX2 motherboards from the bad batch. Here's the rig:

Intel Core 2 Extreme QX6800 @2.93 GHz
Zalman CNPS7700-Cu Cooler
Gigabyte P35-DQ6 Motherboard
2GB Corsair Dominator PC2-9136C5D
250GB Seagate Barracuda 7200.10 16MB cache
Pioneer DVR-108 DVD Burner
Be quiet! Dark Power Pro 850W Power Supply
Samsung SyncMaster 931-BF 2ms Display
Logitech MX510 rodent

The main goal was to have a system that would give the graphics card free reign to deliver maximum performance. Of course, in less powerful systems, these boards might encounter a weaker part of the chain (be CPU-bound, for instance).

Benchmarks were run on Windows XP (DirectX 9) and Vista (DirectX 10) to see the scope of these cards and what one can expect from them in new titles. UT3 and Hellgate: London were benchmarked with the Fraps utility due to lack of proper built-in tools for testing. In the future, we plan to have completely manual runs in order to see how the cards perform in real-world, rather than time demos that have nothing to do with the game itself. We note that the Crysis fly-by demo does not have a lot to do with real game, especially after you venture in the frozen world, so we're going to use two saves: tropical island and frozen island in the future, and bring an average of both.

3DMark06
Our all time favorite synthetic grinder of pixels is a good way to start this story. Now, you shouldn’t pay too much attention at 3DMark on these cards, since there are no increments of thousands. Why? Because a really faster card would lead by by thousands, not hundreds.

In this synthetic test, it does not take long to see the difference between 3850 and 2600XT.

It is clear the 8500GT and 8400GS are poor here, since they scored as low as some notebook chips. On the other side, ATI 2600XT is falling behind XFX 8600GTS which shows a notable performance with its higher clocked parts. However, accolades go to Radeon HD3850, which just shatters the competition.

Crysis
The game everybody wants to hear about and play, Crysis will suck all the juice from your card and CPU and ask for more. It's set on a complete island with mountains, thick dense forest (and yes, we do mean dense), destructible vegetation and structures, high-def textures, amazing water effects, motion blur, full physics, the works.

HD3850 is showing what 320 shader units can do against 32 or 120, depending on marchitecture...

As far as DX10 is concerned, almost all contenders were a painful slideshow at 1024x768, measured in fpm (frames per minute) rather than fps (frames per second). DirectX 9 on the other hand, while still looking good, is playable on the ATI 2600XT, although with some problems with flickering ground textures - but this is something that ATI has to fix with Catalyst drivers, not the GPU error itself. XFX scored first again and you can almost call it “normal”. Almost… we wouldn’t recommend playing it with any of the cards if you want the maximum experience and, frankly, you do! One more specific problem here is that in order to get decent frame rates you have to use 1024x768 resolution on your LCD display, which, coupled with vast vegetation everywhere, gives you a garbled screen of green mash most of the time. Not very pleasant when searching for even a tiniest movement in the woods. For this one, buy some monster if you wanna play or simply do what you did with FarCry... waited until the 6600GT came out and forgot about playing on 5600/5700/5800/5900/5950 boards.

Call of Juarez
This is a flyby benchmark of the DX10 version of the game, and by our opinion is not a very good example. It runs slow as Crysis but is visually far behind it. Thus it is only good for comparing card performance, and not an indicator for current games.

Call of Juarez is an oldie, but a 20fps difference between 3850 and 2600?

Benchmark can be run on Vista only of course, but the game itself works quite well in DX9 actually. During testing, we compared results to our previous tests of 8800GT, 8800GTS and Radeon 2900XT - it seems to us that this benchmark is either CPU-bound or there is something seriously wrong here. The difference between mainstream and high-end hardware is just too low and too close to low-end and mainstream parts. Having said that, we're tossing out CoJ benchmark from our test suite.

Unreal Tournament 3
The king of frenetic shooting is with us again and looking prettier than ever. Wizards of Epic outdid themselves once again and gave us yet another brilliant game. What is more interesting is that UT3 is a DX9 engine, just like many other surprisingly good titles that have spawned recently. Some brilliantly designed CTF maps will make "Is this a movie or a game?" bubble up in your head. Yeah, it is that good.

The best game for lower-end hardware is none other than the almighty UT3

That is why the scores are almost identical. For the first time in this whole roundup, we could get playable framerates at 1280x1024, and that wasn't all. Resolution was followed by setting in-game quality settings to the max, which means 16x anisotropic filtering with textures on full quality. Not standard, but the cards can take it. Surprisingly, all boards perform nicely, which is refreshing, and only 8500GT and 8400GS fail the task. Thumbs up for UT3 and its scalability - best buy for Christmas if you have this kind of hardware!

Hellgate: London
In the time of testing we only had Hellgate DX9 demo, but it is still a good showing of what you can expect, especially with these cards. In the full game there will be more detail and more enemies in DX10 mode - and only the strongest will survive. We're introducing full version of the game with our next graphics card review, currently under way.

Yes, 3850 is that good. Shocking scores show the real difference between mobile GPUs and a real desktop charger

Again, all cards except the 8500GT and 8400GS work well, with 8600GT falling behind a little. We had some problems in Vista though, manifested as stuttering for no apparent reason. This too will definitely be a title to look for - and if you own 3850, no performance problems here. You can take this card and play the game at 1920x1200, frame-rate will be above 50fps, or the scores other cards achieved in this resolution.

World in Conflict
Strategies are often regraded as titles that work on every setup and something you can play even at work. Well, not this one. WiC needs a solid graphics card to run, and because of many details on the screen lowering the image quality is not really an option. Here DX10 is taxing the cards hard, halving the frame-rate but giving you more detail and effects.

Again, we were rewarded with non-playable framerates at 1280x1024, and had to default in lowest resolution...

The 2600XTs are looking almost on par with XFX , but we don’t recommend DX10 on any of the cards. Zotac's 8600GT is gold middle for the 100-dollar market. It is important to mention that playing the game goes much faster as the benchmark is doing a very low fly-by through explosions, something you’ll never do. 8500 and 8400 GPUs don't have the horse power and are just plain useless here. This game also shows that 320 shader units has its limits when it comes to World in Conflict. This is one game where neither brilliantly-performing Radeon nor GeForce chips can shine.

Company of Heroes
Here's another RTS that strangles your PC. This one is well known for it, more precisely its’ DX10 part. Old DirectX mode is fine and playable, but the new one can make you cry. Sadly, there is not much to look for visually, so you can easily turn to DX9 and save yourself, and your eyes, the trouble of stutter-play.

Radeon HD3850 is the first card in this price range that can give playable performance in DirectX 10 mode...

The game literally killed everything but one when running DX10, as if we were running Crysis or something. But on DX9 it flew on 2600XTs and XFX 8600GTS. 8600 showed relative mid-level performance, and even the 8500GT was playable… sort of…

When it comes to results in 1024x768, even though the average was quite high in 1280x1024, all boards dipped in sub-10fps zone as an average, and the goal of this roundup was to show performance in playable resolutions. Performance delta on RV670 GPU just goes to show that there are some severe issues with DX10 API or the way that API is being implemented. 132 fps versus 40 fps on the same machinery... worse of all, game does not look brilliantly different, but rather looks like subtle tweaks all around. However, is that worth sacrificing 70% of performance?

In Short
In the end, after struggling to find playable settings for these cards, and considering what a monster rig we used for testing - the only card we can recommend to acquire for Yuletide is the Radeon HD 3850. The only two other boards that have enough horsepower for DX10 gaming are XFX 8600GTS and HIS IceQ 2600XT.

The cards will run all the older DX9 titles with no fuss, even with a standard setup they are likely to be found in - a dual core Intel/AMD with 1+ GB of RAM. The bigger problem would be 1024x768 resolution on flat panels. Garbled screen is not so appealing, and all cards have 256MB onboard that struggles with 1280x1024, undoubtedly the most popular resolution out there. The slowest ones, 8500GT and 8400GS, are no match even for some older games and are recommended only for desktop performance and movies. Since these boards come with HDMI adapter (unique feature from Zotac), we would feel much more comfortable if the boards were named VideoForce and focus on video features (PureVideo inside), rather that being in the same family with 8800 series.

Bear in mind that our Bartender's Report only considers these cards as gaming cards. Their purpose might be really good in other segments, but if you want to play some of tested games during the Yuletide shopping season and break, this is how they compare.

Our message is: for DirectX 10 you need a faster graphics card, simple as that. µ