by Alexey Stepin , Yaroslav Lyssenko
03/29/2009 | 09:39 AM
We used to say that multi-GPU solutions, especially discrete ones, were limited to expensive and luxurious computers of enthusiasts who did not care about money when it came to ensuring maximum performance. Even if not worse in terms of sheer speed, entry-level multi-GPU solutions used to be inferior to single-GPU cards in such consumer properties as reliability, compatibility, ease of use, etc.
Still, multi-GPU technologies have been evolving and getting rid of their downsides. And we can now say that they have matured for real, especially in the hands of AMD/ATI specialists. Their Radeon HD 4870 X2, based on two rather simple and inexpensive RV770 GPUs, has enjoyed a long period of being the fastest single-PCB graphics solution, beating the best single-chip products Nvidia could offer.
CrossFireX technology has not been that successful in the lower market sectors, though. For example, the Radeon HD 4850 X2, a less advanced counterpart of ATI’s flagship model, has not been recognized by ATI’s manufacturing partners and is so far represented with only one product on the market (it is the Sapphire HD 4850 X2 2G/1G GDDR3 and you can read about it in our earlier review). Users’ demands seem to be fully satisfied with classic single-chip graphics cards in the lower price segments although there is a niche for multi-GPU technologies there, too. As we showed in our earlier article, a pair of Radeon HD 4830 cards cost less than one Radeon HD 4870 but delivered higher performance (and offered an inexpensive way of upgrading the graphics subsystem for people who had one Radeon HD 4830).
Considering the recent reduction of prices on AMD/ATI’s produces, we are interested to learn how appealing an even cheaper CrossFireX subsystem, based of two Radeon HD 4670 cards, may be. This simple affordable RV730-based graphics card is a perfect choice for HTPCs but is no good for gamers due to its low performance in modern games – it has only 8 raster back-ends and a 128-bit memory bus. The new recommended price of the Radeon HD 4870 is only $149 (for the version with 512 megabytes of memory) and buying two Radeon HD 4670 cards at once won’t be much of a saving. But is there an option of cheap upgrade if you’ve already got one such card? Let’s see how effective this anti-crisis solution is from a technical point of view.
Like its senior RV770-based mates, the Radeon HD 4670 offers full support for CrossFireX technology with up to four graphics cards working together in one subsystem. Compared with single cards, the Radeon HD 4670 CrossFireX tandem looks like follows:
It is clear that the two-card configuration boasts solid technical parameters: 128 superscalar shader processors (with a total of 640 ALUs), 64 texture processors and 16 raster back-ends. Considering the high core frequency (750MHz), this tandem is theoretically comparable to the Radeon HD 4870 in terms of performance and even better than it in terms of texture-mapping speed.
The two Radeon HD 4670 working in CrossFireX mode consume a total of 94 watts, which is lower than the power consumption of the Radeon HD 4850/4870 or the GeForce GTX 260 Core 216 based on the 55nm G92 core but higher than that of the GeForce 9800 GTX+/GeForce GTS 250.
So, a Radeon HD 4670 CrossFireX configuration looks like a viable solution, but technical characteristics do not guarantee a success in real-life applications. So, let’s move on to practical tests.
We are going to investigate the gaming performance of the budget ATI Radeon HD 4670 CrossFireX tandem in the following testbed:
The graphics card drivers were configured in the same way as before: to provide the highest possible quality of texture filtering and to minimize the effect of default software optimizations. We enabled transparent texture filtering, with multisampling mode used for both architectures, because ATI solutions do not support supersampling for this function. As a result, our ATI and Nvidia driver settings looked as follows:
We used the following synthetic benchmarks and games:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the ordinary user doesn’t have to know how to do it. We made a few exceptions for selected games if that was necessary. We are going to specifically dwell on each exception like that later on in our article.
Besides Radeon HD 4670 CrossFireX, we have also included the following graphics accelerators to participate in our test session:
We ran our tests in the following resolutions: 1280x1024, 1680x1050 and 1920x1200. Everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.
Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 2.9.8. In the latter case we ran the test three times and took the average of the three for the performance charts.
The entry-level CrossFireX tandem has an excellent start, boasting fantastic scalability of 90-95%. That’s close to the theoretical maximum! It cannot compete with the premier-league solutions but easily beats the Radeon HD 4850 and GeForce 9800 GTX+. Unfortunately, its bottom speed is not high enough for playing the game at 1920x1200.
The Radeon HD 4670 CrossFireX is only inferior to the Radeon HD 4870 1GB and GeForce GTX 260 Core 216 in this test, and the gap from the latter is a mere 10% at the first two resolutions and about 15% at 1920x1200. That’s an impressive result for a tandem of two cards each of which costs far below $100. On the other hand, the frame rate is still too low for comfortable play.
We disabled the integrated frame rate limiter in the game console for the sake of comparing the cards. The game’s built-in benchmarking options do not provide information about the bottom speed, so there is no such info in the diagrams.
The tested CrossFireX platform performs well in this game, too. The scalability of the system relative to the single Radeon HD 4670 is 66% at 1280x1024, which is somewhat worse than in the two previous tests.
Despite the good average performance, the bottom speed of the Radeon HD 4670 CrossFireX is not high enough for playing this game at resolutions higher than 1280x1024. Yes, the tandem is head above the mainstream graphics cards but is no match to the performance-mainstream products.
In other words, you won’t be able to play Far Cry 2 at high resolutions if you add a second Radeon HD 4670 to the one you already own. On the other hand, such an upgrade is still more appealing than replacing your Radeon HD 4670 with a Radeon HD 4850 because the latter has problems with bottom speed even at 1280x1024.
Far Cry 2 is the only game in which we had some problems with the smoothness of control. The game seemed to react more slowly to mouse movements, yet not to the extent of total unplayability as had been the case with the Radeon HD 4830 CrossFireX.
Nvidia’s GPUs feel better in this game, delivering higher bottom speeds, so it is no wonder that the Radeon HD 4670 CrossFireX is unable to beat the GeForce 9800 GTX+. Of course, installing a second 4670 allows playing at 1680x1050 but you can achieve the same with a single Radeon HD 4850.
The game runs on the Source engine and has an integrated benchmark, but the latter does not report the bottom speed information.
The Radeon HD 4670 pair turns in a splendid performance in Left 4 Dead, being ahead of the GeForce GTX 260 Core 216 at 1280x1024 and only 3-4% behind the latter in the higher display modes. It cannot catch up with the Radeon HD 4870 1GB, but that’s not a problem since its average frame rate is over 60fps even at 1920x1200.
To achieve a playable speed in this game we disabled FSAA and such resource-consuming options as Sun rays, Wet surfaces and Volumetric Smoke. We use the Enhanced full dynamic lighting (DX10) mode for our test and additionally enabled the DirectX 10.1 mode for the ATI cards.
Things are not so bright for the Radeon HD 4670 CrossFireX tandem in this test. Notwithstanding good scalability, it cannot deliver a playable frame rate even at 1280x1024. The tandem seems to have enough graphics memory, memory bandwidth and computing resources, so we guess the reason is that the 2x8 RBE design is less efficient than the Radeon HD 4850’s 16 RBEs due to CrossFireX overhead or the rendering techniques employed by the game engine.
The problem we observed in our Crisis Management CrossFireX review is not present here. CrossFireX mode is enabled normally, without your having to enter the vid_restart command in the game console. This problem must have been corrected in the new release of the Catalyst driver.
Being ahead of the GeForce 9800 GTX+ at first, the Radeon HD 4670 CrossFireX falls behind it as the display resolution grows up. The gap is as wide as 12% at 1920x1200. However, it is in this test that the CrossFireX tandem shows outstanding scalability. The average frame rate grows up by almost 120% at 1280x1024. The tandem looks appealing considering the minimum cost of a second Radeon HD 4670.
Although such a high growth of speed looks suspicious, our brief check of image quality revealed no difference between a single Radeon HD 4670 and two such cards working in CrossFireX mode.
Like in the previous test, the effect from joining two Radeon HD 4670 cards into a single CrossFireX subsystem is very high: the tandem can be 125% faster than the single card in terms of average performance. The bottom speed doesn’t scale up so well and is no higher than that of the Radeon HD 4850. Therefore the resolutions above 1280x1024 are still unplayable on the CrossFireX configuration.
Like in Devil May Cry 4, there is no difference between the single card and the Radeon HD 4670 CrossFireX in terms of visuals, so we can’t say that ATI uses some optimizations to increase the performance of its multi-GPU solutions at the expense of quality.
Here, purchasing another Radeon HD 4670 for your system is a simple and inexpensive way of opening up higher resolutions for play, including 1920x1200. Of course, you need a CrossFireX-compatible mainboard for such an upgrade. But if you’ve got a Radeon HD 4850 or GeForce 9800 GTX+, you should not care about replacing it with a pair of Radeon HD 4670 cards because you won’t get any practical benefits – the bottom speeds of these solutions are almost the same.
Mass Effect is yet another test where CrossFireX technology boasts 100% and higher efficiency. And there is a nice growth of bottom speed too, especially at 1920x1200. Alas, the overall level of performance is not high enough for comfortable play. The GeForce GTX 260 Core 216 is unrivalled.
Like in the previous cases, we could not spot any degradation in image quality on the CrossFireX platform.
The Radeon HD 4670 CrossFireX configuration performs well in this test, closing the gap from the GeForce GTX 260 Core 216 to zero as the resolution grows up. Moreover, it is a mere 8% behind the leading Radeon HD 4870 1GB. So, we have to correct our opinion about the usefulness of entry-level multi-GPU solutions, although they only make sense if cost much less than GeForce GTX 260 Core 216 and Radeon HD 4870 1GB cards.
The Radeon HD 4670 CrossFireX is only inferior to the Radeon HD 4870 1GB at high resolutions (the gap is a mere 6-8%) and equals the GeForce GTX 260 Core 216 at 1280x1024. The scalability varies from 40 to 60%, which is somewhat lower than in the previous games.
The game has a built-in frame rate limiter set at 30fps.
This is the only game in this review where the performance growth of the CrossFireX platform is less than 25% relative to the single Radeon HD 4670. The bottom speed grows up considerably, yet does not reach the playable minimum.
The Radeon HD 4670 CrossFireX is comparable to the Radeon HD 4850 at every resolution, but the latter doesn’t need two PCI Express x16 slots and a CrossFireX-compatible mainboard and costs less than the total cost of two Radeon HD 4670 cards.
Like in most gaming tests the Radeon HD 4670 CrossFireX offers excellent scalability that approaches the theoretical maximum. It is ahead of the Radeon HD 4850 and (in the SM3.0/HDR tests) comes close to the Radeon HD 4870 1GB due to the high frequency of its cores (750MHz).
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all display resolutions to provide a full picture.
The CrossFireX tandem is just as efficient in 3DMark Vantage as in 3DMark06, but inferior to the other tested solutions including the Radeon HD 4850 and GeForce 9800 GTX+.
The Radeon HD 4670 CrossFireX feels better in the second test, successfully competing with the GeForce 9800 GTX+ and Radeon HD 4850. It loses to these cards at 1920x1200, though.
Now we will try to sum up the test data and make our final verdict about the capabilities of an ATI CrossFireX tandem consisting of two Radeon HD 4670 graphics cards. Can you improve your computer’s graphics performance by purchasing a second 4670 card if you’ve already got one? Will this upgrade be worth the trouble? Let’s discuss each resolution starting from 1280x1024.
The CrossFireX tandem boasts superb scalability: the average performance gain over the single Radeon HD 4670 is 85%. The frame rate even occasionally grows up by 100% and more, like in Devil May Cry 4, Prince of Persia and Mass Effect – without any degradation in image quality! The only exception is Red Alert 3 where the performance gain is only 15%.
Comparing it with more advanced single-chip graphics cards, the Radeon HD 4670 CrossFireX enjoys an average advantage of 7.7% over the Radeon HD 4850 and 14.5% over the GeForce GTX 9800+. It loses but few tests to them and by no more than 9%, excepting Red Alert 3 where the CrossFireX tandem has serious problems with scalability. On the other hand, it is not as much faster than the mentioned single cards as to justify your replacing them with two Radeon HD 4670 cards, especially as the new recommended price of the Radeon HD 4850 is a mere $129. However, purchasing a second Radeon HD 4670 for your graphics subsystem is a good and inexpensive way of increasing your computer’s graphics performance if your mainboard supports CrossFireX technology.
As opposed to the Radeon HD 4830 CrossFireX tandem described in our earlier report, the Radeon HD 4670 CrossFireX cannot beat the faster single-chip cards such as GeForce GTX 260 Core 216 and Radeon HD 4870 1GB. It is slower by 11% and 16%, respectively. The occasional victories (like the 12% lead over the GeForce GTX 260 Core 216 in X3: Terran Conflict) should not be viewed seriously: the Radeon HD 4670 CrossFireX is no match to the mentioned single-chip products from ATI and Nvidia.
The CrossFireX platform behaves in the same way at 1680x1050, offering an average 82.5% performance growth over the single card, but its advantage over the Radeon HD 4850 has shrunk from 7.7% to 4.6% whereas its gap from the Radeon HD 4870 1GB and GeForce 9800 GTX+ has grown wider to 20% and 14.3%, respectively.
So, we can say it once again: purchasing two Radeon HD 4670 cards for building a CrossFireX configuration does not make sense. It is better to buy a single Radeon HD 4850 instead. Such a configuration is only reasonable when you buy a second Radeon HD 4670 in addition to the one you already have.
Oddly enough, it is at the resolution of 1920x1200 that the Radeon HD 4670 CrossFireX enjoys the highest success. It beats the Radeon HD 4850 and GeForce 9800 GTX+ by 8% and 18.5%, respectively. However, it is still 21% and 14% behind the Radeon HD 4870 1GB and GeForce GTX 260 Core 216, respectively. The CrossFireX tandem cannot close that gap.
The efficiency of CrossFireX technology is as high as at the lower resolutions: the two cards enjoy an average 85% advantage over a single Radeon HD 4670. And like in the previous cases, the overall positive picture is somewhat marred by poor scalability in Red Alert 3.
Comparing the Radeon HD 4830 CrossFireX and Radeon HD 4670 CrossFireX tandems, we should confess that the latter has vaguer perspectives. Although its performance is higher than that of Radeon HD 4850 and GeForce 9800 GTX+/GeForce GTS 250 cards, the difference is not as big as to make the multi-GPU tandem more appealing than a single, more expensive, card that occupies only one PCI Express x16 slot. Moreover, this difference can be rather easily eliminated by means of overclocking. In terms of power consumption the tandem is more economical than a Radeon HD 4850 (94 against 110 watts) but less economical than a GeForce GTX 9800+/GeForce GTS 250 that uses a 55nm G92 core and consumes only 80W.
So again, such a subsystem can only be interesting for people who already have one Radeon HD 4670 and a CrossFireX-compliant mainboard. Considering that the Radeon HD 4670 is an entry-level solution, there is a low chance of the user having such a mainboard. Entry-level systems are usually based on mainboards with only one PCI Express x16 slot. Moreover, the owner of a Radeon HD 4670 CrossFireX will face all the problems associated with modern homogeneous multi-GPU solutions, the most important of which is the dependence on driver optimizations as is indicated by Red Alert 3.
We could not spot any problems with the image quality or stability of the multi-GPU solution except for higher control lags in Far Cry 2. We even performed an image quality check in suspicious cases but could not see any difference. Thus, ATI has done a good job on its driver and the situation has improved greatly since our Crisis Management CrossFireX report.