Let's analyze results of each test.
3DMark benchmarkly responds to any effective (in terms of performance in games) upgrade in hardware, so both a faster processor (in the perfect order of CPU ranks in the AMD product line) and an upgraded graphics card increase the rating. This test did not appreciate benefits of 1GB video memory in return for extra MHz in case of Radeon HD4850. Games share this opinion, they prefer an overclocked graphics card with less video memory.
In case of the cheaper graphics card, an upgrade of the weakest processor to the fastest one yields only 1.3-fold gain. The same upgrade with Radeon HD4830 yields 1.6-fold gain already. The top card will enjoy a twofold gain in this situation. However, ratings in this test do not answer which processor is an optimal choice for each graphics card (games will tell us that). However, judging by the absolute value and taking forced AA and AF into account, we can say that scores above 3000 correspond to requirements of popular games, and even Radeon HD4670 with an average processor earns this score. The other cards score above 4000 even with Athlon X2 4800.
So, we can draw conclusions only by exception: even a dual-core Sempron (which is a better choice in its class than Celerons) is too weak for games, even if it's combined with Radeon HD4670, to say nothing of more powerful cards. Small wonder. But it was still worth it to test the bottom, as even weak Pentium E2160 and Athlon X2 4000 worked well with far from the weakest graphics cards. That's probably the effect of growing GPU performance, as not long ago it would have been strange to talk about choosing a proper processor for a graphics card within $80-100. We just wrote that such cards were not designed for gamers.
Although we still use the first version of the game, 1680x1050 resolution, full dynamic lights, and forced anisotropic filtering and antialiasing should generate enough load even for a modern graphics card.
However, as we can see, it's a perfect match for all cards. And we can speak of overcapacity starting from Radeon HD4830. However, this game, at least our demo, helps evaluate a minimal frame rate -- unfortunately, minimal instant fps measured in most tests is of no help, resettability of results is very low. Results of configurations with two processors from the beginning of the list are practically completely limited by a processor (9-12 fps for Sempron and 14-19 fps for Athlon X2 4850e), no matter how fast a graphics card is. It can only accelerate the average frame rate.
But in configurations with other processors, minimal fps grows proportionally to the average fps. For example, in case of the HD 4830, it grows from 20 (Athlon X2 6000) to 24 (Phenom X4 9850) and from 29 to 35 for the HD4870. In other words, neither a processor nor a graphics card are sole limiting factors. You can expect performance gains after you upgrade any of these components. It's apparently up to a user to decide upon the golden mean proceeding from his/her budget and performance in other games.
Call of Juarez
This game does not have any surprises in store. As we found out in our previous articles, compromises in game settings are necessary even in not very high resolutions. As we used forced antialiasing and filtering in all our tests today, only the fastest card managed to provide a sufficient frame rate to play this game. In other cases you will have to reduce graphics quality and even use a lower resolution for cheaper cards.
What concerns processors, only our Sempron is not fast enough for the game. The majority of other processors have to wait for the graphics card to do its part most of the time.
Company of Heroes: Opposing Fronts
In this case not only graphics settings, but also the game itself, were updated to the second part, which generates much more load on a graphics card in the first place. And processors also take some rest only in the second part of the list: the three fastest graphics cards in any combinations with three processors provide similar results. When we take a look at a minimal frame rate, we can see a distinct bondage of this parameter with video memory size, even though the measurement error is big here. Radeon HD4850 with 1GB of memory and Radeon HD4870 in combination with three top processors demonstrate the identical result (about 24 fps). This parameter varies significantly: even the overclocked HD4850 slows down to 10 fps, performance of weaker cards drops to 5-7 fps. So an apparent conclusion here is that minimal fps slumps down during active data exchange with memory, and 512MB is not enough.
We did not expect to see anything new in this game, because this test is limited by GPU performance in the first place. And as we found out, even Radeon HD4870 is not fast enough to play this game with high graphics quality settings in this resolution. To say nothing of antialiasing and filtering, which can slump any card's performance. However, it did not happen, at least the three best cards lose only several fps to forced antialiasing and filtering. What concerns the comparison of processors, this game seems to favor multi-core CPUs, especially in combination with powerful cards.
World in Conflict
Just like in the previous game, the three most powerful graphics cards do not differ much here, if we take absolute values. But there is some performance gain in percents. Despite a serious graphics load, strategies have traditionally high CPU requirements -- we can clearly see this here. So it's easy to determine which processors are better. By the way, the triple-core Phenom demonstrates very good results not only in this game, and it's a very inexpensive processor.
Devil May Cry 4
We have chosen the average frame rate in the second scene with maximum load on a graphics card (the lowest frame rate) -- all options set to maximum and even MSAA x8 enabled in the game -- it apparently runs well even on inexpensive cards. And it appears that a processor is not too busy in this game, so game performance is determined by a graphics card practically in all PC configurations. Even a Sempron can maintain a comfortable frame rate. And if you install at least Athlon X2 4800, there will be no visual slow-downs in the minimal frame rate. You may think that such simple tests shouldn't be used to benchmark modern hardware, but this console game does a good job of representing hardware requirements of similar titles in this genre. In other words, modern popular games still run well on mediocre hardware.
Unreal Tournament 3
We tested this game in two modes: flyby and botmatch. As we already mentioned, the game will load mostly a graphics card in the first case. In the second case, it's a CPU test. However, botmatch actually loads a graphics card as well, because it imitates combat operations accompanied with proper fireworks. What concerns flyby, it's hard to say why it favors fast processors, but it's a fact.
Even though we raised graphics quality settings and resolution, Radeon HD4830 and Athlon X2 6000 are still sufficient for comfortable gameplay. By the way, performance gains from hardware upgrades are proportional in flyby, but botmatch is less predictable. Along with a wide spread of results and a high measurement error in this mode, this game requires a balanced hardware configuration (a more powerful graphics card needs a more powerful processor and vice versa). For example, if the above-mentioned combo yields good results, the same processor with any HD4850 card instead of the HD4830 suffers from a performance drop, which cannot be possibly written off to a measurement error, because we've run several series of this test for each card. However, HD4850 + Phenom X3 8750 expectedly demonstrate a performance gain.
It's easy to answer the questions we asked in the beginning of the article with these test results. Radeon HD4830 is certainly a very good addition to the segment of inexpensive graphics cards. It copes with very difficult tests in combination with a Mid-End processor: 1680x1050, anisotropic filtering and full-screen antialiasing. But the HD4670 will evidently require lower resolutions. Even though our modification was a little overclocked at the factory, it's clearly a budget card designed for users, who play games only occasionally.
It was very interesting to compare the overclocked HD4850 with the same card operating at the nominal frequencies, but equipped with twice as much memory. Higher frequencies win in most classic games. On the other hand, the latest games often require more than just 512MB, especially when we choose a native resolution for modern LCD monitors. If we take a minimal frame rate into account, not just an average frame rate, the number of games that benefit from higher memory capacity grows. It's hard to draw a conclusion here. To be more exact, there are two contrary but true comments: all HD4850 cards are powerful enough to run modern games. You will have to find a reasonable balance of performance and graphics quality in hightech games for any modification, so there are no fundamental differences between them, just take prices into account. This conclusion always holds true. Rollouts of new GPU generations are always accompanied by releases of hightech games, where none of the cards allows to choose top quality settings and still get high fps. It may be possible for CrossFire and SLI tandems with well-tuned drivers, which will take some time. So the second conclusion concerns those users, who still want to get reasonably top performance. For example, overclockers may take a closer look at cards with more memory, nominal frequencies, and a good cooler -- they can try and overclock them on their own. However, memory rarely overclocks much. It depends on memory chips, so Radeon HD4870 with 1GB of GDDR5 memory surely looks like the most promising choice, although the most expensive one.
Write a comment below. No registration needed!