Many gamers upgrade their systems over time, and one of the most cost-effective ways to add performance to an existing rig is to add a second video card. AMD's Crossfire technology allows users not only to add a second identical card, but to add any card within the same series.
We've previously tested HD 7870 Crossfire, which you can read about here, but what type of performance increase can one expect from using an HD 7870 and HD 7850 in a mixed Crossfire, or "FrankenFire," arrangement? And what are the effects of AMD's new frame-pacing driver on such mixed Crossfire setups? We aim to find out!
Our test system for this article concisted of an Intel Core i7-3770K@4.4GHz, an Asus Maximus V Gene motherboard, an XFX Radeon HD 7870 2GB, a Sapphire Radeon HD 7870 2GB, an MSI Radeon HD 7850 2GB, 16GB of Samsung DDR3@1866, Windows 7, Catalyst Version 13.2 beta 7, and Catalyst Version 13.8 beta 2.
In Part I of this article, we'll be providing gaming benchmarks using a mixed Crossfire setup of an HD 7870 and HD 7850, to show you exactly how performance scales in Crossfire versus a matched HD 7870 set. In Part II of this article, we'll compare the results of the mixed Crossfire set using Catalyst version 13.2 and version 13.8, the latter which incorporates AMD's new frame-pacing technology. To provide the greatest insight into how mixed Crossfire works, we ran all cards at their reference speeds, then ran the HD 7850 at HD 7870 speeds, above HD 7870 speeds, and even ran the HD 7870 underclocked to approximate the speed of an overclocked HD 7850. That generated a lot of data to review, and as always, the results may surprise you!
Part I: Performance Scaling of Mixed HD 7850/HD 7870 Crossfire versus Matched HD 7870 Crossfire on Catalyst 13.2
For our tests, we had one HD 7850 and two HD 7870 cards on hand, so we'll have to extrapolate from the 7870 duo how two 7850s would perform. The focus of our analysis, therefore, will be on how mixed Crossfire performs versus two of the faster cars. To figure this out, we ran three benchmarks, all at a resolution of 1920x1080, as follows:
- 3DMark Fire Strike Graphics Score - Standard Preset
- Battlefield 3 (Swordbreaker Single-Player Level) - Ultra Settings/4xAA
- Hitman: Absolution (Built-in Benchmark) - Ultra Settings/4xAA
Each of these benchmarks focuses on different aspects of performance. 3DMark's Graphics Score is the purest test of the video processing capabilities of the system, and excludes CPU power almost entirely. Battlefield 3 uses an intense graphics engine to stress the video cards, and because this is a real-world test, it isn't scripted and thus requires some CPU power. Finally, Hitman: Absolution's built-in benchmark has a very demanding graphics engine, but also requires a lot of CPU power to simulate a large crowd of people in the scene. As you'll see, each reacts differently to additional video processing horsepower.
3DMark Fire Strike
We've packed a lot of information in this graph, so study it carefully. What we have on the left axis is the Graphics Score, which is represented by the wide bars on the graph. On the right, we have the scaling results, represented by the two lines on the graph, orange for scaling versus an HD 7850, and light blue for scaling versus an HD 7870.
Right off the bat, you'll see that Crossfire scaling isn't very good in 3DMark. Even looking at the HD 7870CFX set on the far right, it only achieves 1.65x scaling versus an HD 7870, and can barely double the performance of a lone HD 7850. 3DMark Fire Strike's Graphics Score is based on the frames per second generated over two separate and very different scenes, and for whatever reason, the second of these two scenes barely utilizes the second video card. Our assumption is that it basically comes down to more advanced rendering technologies being employed that aren't being handled correctly by the Crossfire drivers. As a preview of what you'll see on the next page, Catalyst version 13.8 does not fix this, and in fact may make it worse.
But back to our main event - how does mixing two cards compare to just one card, or a pure dual HD 7870 setup? Well, the mixed HD 7850/HD 7870 actually performed exactly as we would expect two HD 7850s to perform, scaling at 65 percent, identical to the scaling of our two HD 7870s over a single HD 7870. Throughout our benchmarks, we'll want to determine whether adding a more powerful card to an existing card offers any benefit over just adding a second slower card - in other words, is the faster card simply held back to the slower card's speed?
To make clear what happens to the faster card in a mixed Crossfire set, have a look at the MSI Afterburner screenshots to the left, which indicate the GPU usage of our two cards when the HD 7850 is clocked at 860/1200, and then when it's clocked at 1000/1200. Clearly, the slower card in a Crossfire set will be working harder, which is why the HD 7850 in both tests is pegged at 99 percent usage. The HD 7870, on the other hand, is just barely breaking a sweat when paired with an HD 7850@860, trotting along at 82 percent usage. It starts to come alive when the HD 7850 is at 1000MHz, hitting around 91 percent usage, but as you can see, it's still not able to give its best effort, because the HD 7850 is always holding it back. The HD 7870 performs, not surprisingly, exactly like an HD7850@1000MHz. How do we know this? From the 3DMark results above, you can see that the overclocked HD 7850 is 11 percent faster than the stock card, which happens to be the same delta as going from 82 percent usage to 91 percent usage.
A few other things to mention here - first, an HD 7870 is approximately 21 percent faster than an HD 7850 when both are run at stock speeds, and is about 10 percent faster when both are run at 1000MHz Core.Unfortunately, our HD 7850 sample simply couldn't clock high enough to reach the speed of an HD 7870, which would require a core clock of around 1125MHz, so the best we could do was test at 1050MHz. This setup performed quite well, nearly matching two HD 7870s, but what was very ironic in this test was that underclocking the HD 7870 to 925MHz in order to emulate an HD 7850@1050MHz actually improved the score.
This graph gets a bit more complicated, as we're also plotting average frames per second along with minimum frames per second. But a few things are evident from these graphs. First of all, Battlefield 3 scales far better than 3DMark. The HD 7870 pair hits 1.94x scaling over a single HD 7870, so that should be our target for the HD 7850. But when that 7850 is paired up with an HD 7870, scaling is actually 2x, which is better than two HD 7850s could do. So unlike in 3DMark, having mismatched cards is arguably better than having two of the lower-tiered cards. In fact, when the HD 7850 is overclocked to 1050MHz and paired with an HD 7870, the result is 111fps, versus the pure HD 7870 tandem's 112fps. It really couldn't get much closer. And again distinguishing the results from 3DMark, slowing down the HD 7870 to match the HD 7850's overclocked processing power does not help - in fact, minimums drop way down.
In this built-in benchmark, the HD 7870 Crossfire set scales at about 1.84x versus a single HD 7870. That's actually quite good, and probably about what you should expect generally of Crossfire. The HD 7850 paired with an HD 7870 scales a bit better, at 1.93x, and we can extrapolate that this is a better result than two HD 7850s would do. Note, however, that the mixed pair can't perform miracles - even when the HD 7850 is overclocked to 1050MHz, the matched HD 7870 pair still beats the mixed pair by 4fps. Turning down the core clock on the HD 7870 to "match" the overclocked HD 7850 doesn't help at all, unlike in 3DMark.
A Word on Smoothness
Now, of course everyone is wondering what the actual perceived performance of the mixed Crossfire pair is. Well, we'll say more on the next page, but our opinion is that the concerns about Crossfire are generally overstated when there's performance to spare, i.e., the frames per second are comfortably above 60fps. In our run-throughs of Battlefield 3, we didn't notice any hitching or other performance degradation running Crossfire. It was definitely better than a single card on its own. Hitman's built-in benchmark wasn't entirely smooth in Crossfire, although still better than a single card, which struggled at under 40fps. Note that the minimums in this game were almost always below 60fps even in Crossfire, which could play a role in the perceived problems. And 3DMark demonstrated considerable hitching - in fact, in 3DMark's second scene, where the GPU usage dropped way down on the faster card, the visuals were terrible, worse even than a single card. So does Catalyst version 13.8 beta 2 fix these problems? Turn to the next page to find out!