Choosing the right parts for a performance PC has never been easy, and one of the biggest things that has added to this problem recently is multiple graphics card solutions - also known as SLI from Nvidia or Crossfire from ATI.

Given that ATI's latest range of HD 4800 series graphics cards seem to be the most popular choice of cards for gamers, and Intel's Crossfire-capable chipsets (X38, P45 and X48) are also quite prevalent, now's a good time to put the cards on the table and see just what benefit there is, if any, of going for two GPUs instead of one.

There's been a lot of arguments about the benefit of Crossfire, especially at the "low" resolution of 1680x1050, so today we're going to compare a number of single and dual-GPU solutions at both 1680x1050 and 1920x1200 in a variety of games and at different quality settings just to see exactly where the performance advantages lie. This will be good not just for those contemplating what GPU setup to get, but will also help those who have one card and are looking to upgrade.

Testing methods

Today we're going to test three different cards, the ATI Radeon HD 4830, 4850 and 4870. We'll do all three cards by themselves, then doubled up for Crossfire, and to spice things up we're also going to throw in a 4830 + 4850 CrossfireX combo, just to see if you can save a few bob without losing too much performance. This promises to get interesting...

Specifications

Test System

CPU: Core 2 Duo E8500 @ 3600MHz
RAM: 2x2GB G.Skill DDR2-1000 CL5
Mobo: Asus P5Q Pro
O/S: Vista Home Premium 64bit SP1
Drivers: ATI Catalyst 9.1

3DMark06

As usual, our first port of call is 3DMark06 just to get a relative view of where the cards sit against each other. As we go up in resolution, we also go up in anti-aliasing (AA) strength just to stress the cards more and try and sort the sheep from the lambs.

Dang - that's some nice scaling right there. Notice how all the cards get a massive points boost by adding another card even at a measly 1280x1024 resolution. What's most impressive is how the Crossfired 4870s barely twitch at the increasing resolution and AA settings, with a large lead over the other setups at 1920x1200 with 4xAA. Confusingly, the 4830+4850 doesn't fare as well as the 4830 Crossfire does, possibly due to the extra overhead of making two different cards work together. Let's see how that pans out in the "real" world...

Crysis Warhead

A recent addition to our testing lineup, this is the sequel slash stand-alone addon for the orginal Crysis game. It's also a brutal torture test for modern hardware, and has been known to largely ignore any potential benefit from Crossfire setups, so let's see what we can squeeze from it.

Well that's reassuring - the single cards aren't playable on "Gamer" settings (the second highest settings that the game offers) at 1920x1200, but all the Crossfire setups offer acceptable framerates (I draw the line at 30fps as playable for Crysis games). What's also good to see is that the 4830+4850 combo actually manages to nudge ahead of the 4830 Crossfire cards, only just though, and then in turn the 4850 Crossfire cards beat it again. The 4870 simply shines here, basically matching what the lower cards can do in Crossfire.

GRID

I've said it before and I'll say it again - this is a sexy, sexy game. This is the game to show off the latest ATI cards with, but which ones exactly?

Well, you definitely don't want a 4830 in the mix if you want to crank the AA up to 8x - most likely due to its slower memory bandwidth bottlenecking the rest of the card. Again the single 4870 muscles it right up there with the Crossfired cards at 4xAA, but bafflingly loses to the 4850 at 8xAA. Never mind, the take home point here is that GRID absolutely, positively loves Crossfire - even at 1680x1050.

Trackmania Nations Forever

Another game that's been known to shun Crossfire setups, but is also very CPU intensive so this could explain why. Nevertheless, let's give it a crack...

Ouch, some pretty underwhelming results there. The 4830 and 4850 Crossfire setups yield less than 20% performance increase at 1920x1200, and only about 10% at 1680x1050. The 4870 does however crack the 60fps barrier at 1920x1200 in Crossfire, so that's some good news at least.

Call of Duty 4: Modern Warfare

Look, I'll buy Call of Duty: World at War and test with that one day, I swear. Unless Modern Warfare 2 is out by then. Which it probably will be. Anyway...

Bam! There is some more awesome scaling for you. Although it is disappointing to again see that adding a 4850 to a 4830 yields no benefit as opposed to having two 4830s. Oh well, with 90fps at 1920x1200 with 4xAA, it seems like a fairly academic complaint.

Fallout 3

Another recent addition to our lineup, Fallout 3 is the latest post-apocalyptic vision of the future in a game, and a mighty addictive game at that. Even if first-person role playing games aren't your style, it's worth checking this game out simply for the mind-blowingly huge outdoor arenas it features.

Good Crossfire scaling here as well, more so at 1920x1200 than 1680x1050. All the single cards pull acceptable results at the lower resolution, but you'll be wanting Crossfire for high-res and high-AA. Good to see the 4830+4850 CrossfireX combo offering a few more frames than the 4830 Crossfire setup at high resolutions too. NB: See how all the Crossfire setups hit a wall at ~66fps at 1680x1050? That's called a CPU bottleneck. So now ya know.

Average FPS

Now for the hard part - maths! Lucky for you, I've done all the comptabulations and found the average frames per second each setup offers over all the games, separated into the two different resolution/quality settings. This way we can get an overall picture of how much performance increase you're getting from adding another card.

So basically, if you take a single 4830, adding another one will give you a 42% boost at 1680x1050, and 53% at 1920x1200. Adding a 4850 instead will give you a 43% increase at 1680x1050, and 57% at 1920x1200. Kinda hard to justify the extra expense of adding a 4850 instead of another 4830 when you look at it like that.

Now if you start with a 4850 and add another 4850, then you'll receive 33% more performance at 1680x1050, and 57% at 1920x1200. Adding a 4830 instead will give you a measly 31% more at 1680x1050, and 40% more at 1920x1200.

Finally, adding a 4870 to another 4870 will give you an uninspiring 14% boost at 1680x1050, but a rather more positive 36% boost at 1920x1200. Given that you're literally doubling the cost to do this, is it worth it? Well, there's one more way to look at it from a cost perspective...

NZ Dollars per FPS

Unique to us here at Gameplanet is how we break the costs down to show you exactly how much you're paying for each frame per second in the titles we've tested here today.

Remember, you have to take these figures on board and combine them with the average frames per second figures from above to get a clear picture about what is offering the best price/performance ratio.

Conclusion

My personal recommendation is this - if you want to stay above the 60fps waterline, then at 1680x1050 you have two options: The single 4870 offers an average of 72.8fps at a reasonably low $6.66 (!) per frame per second, and the 4830 Crossfire setup coughs up 76.6fps for $7.02 per fps. These solutions are the best at this resolution in my opinion.

At 1920x1200 it's a slightly different story. Here I think the obvious answer is 4850 Crossfire, offering an average of 65.1fps for $10.41 per fps. Certainly not the cheapest solution, but the only other solution that outperforms it is the 4870 Crossfire setup which offers a 73.5fps average for a whopping $13.19 per fps.

Well, there you have it. All the info you need to justify that second graphics card. Go forth and multiply.

---

Our thanks to PlayTech for providing the review samples.