Let's analyze results of each test.
SYSMark 2007 Preview
This time we've replaced PCMark with SYSMark for general performance benchmarking, which is notable for using tests based on real applications: you can see the list on the official web site. This is the reason why it's a very whimsical benchmark from the point of view of testers -- scripts that imitate user activities can easily "go wrong". So a test may just fail, for example, if you run it in non-clean Windows. However, if you need a relatively quick performance comparison of several configurations in office and popular professional software, this benchmark is the most adequate choice. It's developed in cooperation with an open consortium of PC component manufacturers. So biased optimizations are practically out of the question.
At the same time, the "real" approach leads to a strong influence of other components besides a processor. For example, memory with stricter timings and a fast hard drive may add up to 10 additional points to the total score, which exceeds the effect of upgrading a processor one step up. For this reason it's incorrect to tie obtained results to a processor. If you compare processors, you must make sure the other PC components are identical (which is not always possible in practice; for example, it's hard to control what timings are set by BIOS, because the automatically generated report does not include all minor details).
Keeping this peculiarity in mind, it was very interesting to see what happens when a different graphics card is installed. Strictly speaking, if you have a look at the list of programs above, you will see that there is nowhere this dependency can come from. But as the total score is affected by memory timings and other minor issues, which are not noticeable in practice, we should check it up anyway. However, we found nothing unusual. In our case, test results to within a measurement error were tied to a processor. The highest score was reached by the most expensive dual-core processor among our contenders, the second place was taken by the Q6600 quad-core processor. Performance difference between the outsider (E2180) and the leader (E8400) reaches 1.5 times. However, if we compare it with the price difference between these processors, it turns out a Pyrrhic victory -- the E8400 is almost three times as expensive! Upgrading Pentium E2180 to the E4700 or Athlon X2 6000+ yields a performance gain proportional to the costs (1.3 times).
3DMark Vantage
Just like in 3DMark 06, scores in the new 3DMark depend primarily on GPU performance, but the effect of a CPU has also grown. To be more exact, the test was optimized for quad-core processors. So the only quad-core processor in our tests demonstrates the best result with a permanent graphics card.
However, results with cards ranging up to the 9600 GT are still limited by GPU performance even in combination with the slowest CPU. Keep in mind that a synthetic test is not a game, so it's not correct to speak of sufficient/insufficient performance here. On the contrary, a sensitive test must ideally respond even to such configuration differences, which are not apparent in modern games, but which may eventually manifest themselves. We cannot say for sure that it shall happen. We can only establish a fact that, according to 3DMark Vantage, a quad-core processor makes sense even with GeForce 8800 GT cards or the like.
But let's estimate how justified this or that upgrade is by comparing performance and price differences. For the cheapest graphics cards, up to the 9600 GT, it makes no sense to evaluate performance gains from CPU upgrades. You will have to pay three times as much for the maximum gain of 16% (between the cheapest and the most expensive processor in combination with the 9600 GT). What concerns the 9800 GTX, this difference grows to 24%. It cannot be compared with the price difference, of course, but it hints that 3DMark Vantage developers do not recommend gamers to ignore CPUs completely. Let's see how this recommendation holds in real games.
Crysis
We'll start with the most resource-intensive game in our test procedure and probably among all modern games. What have we got here? Performance is apparently limited by a graphics card in all cases. We can formally note some signs of CPU dependence with the 8800 GT and 9800 GTX, but it's quite clear that in order to obtain comfortable frame rates, you must have a more powerful graphics card in the first place. Only then we may speak of possible CPU dependences. Cost-proportional performance gains are demonstrated only for GPU upgrades up to the 8800 GT inclusive. And the 9800 GTX does not look like a justified deal in this game, where performance grows out of proportion to the performance difference.
Unreal Tournament 3
We've added this game to our test procedure for graphics cards at our readers' request. We run this test in two modes: flyby and botmatch. Both modes load a processor and a graphics card. But botmatch formally focuses on a CPU, and flyby -- on a graphics card. Our conclusion about this game is apparent -- GeForce 8600 GTS or any faster card in combination with any processor will do to play this game in both resolutions, because all of them manage to provide a comfortable average frame rate.
However, if you don't want the minimal frame rate to drop below the comfortable level, it makes sense to choose a CPU from the middle of our list. Pay attention to a wide spread of results in this test (measurement error). For example, some processors demonstrate best results with a weaker graphics card, which just cannot be true. Taking into account this and the fact that the game is of little interest for future tests (all graphics cards demonstrate sufficient performance here even now), we'll most likely replace it with a more interesting game in our next tests.
S.T.A.L.K.E.R.
S.T.A.L.K.E.R. generates a much heavier load. This game looks appropriate in our test procedure, symbolizing average-load shooters, the most popular games now. In prospects, the next part of this game will most likely be included into our updated test procedure.
As resolution grows, the 8800 GS becomes the minimal graphics card in this game. Upgrades to more powerful graphics cards (up to the 8800GT) are highly desirable. But it's not expedient to replace the 8800GT with the 9800 GTX -- additional 5 fps are hardly worth the money for this upgrade.
The formal leader among processors is the Q6600. However, the spread of CPU results is minimal even with GeForce 9800 GTX, so you can save on this component here. Within due limits, of course. As in case of Unreal Tournament and the like, you'd better stick to "mid-line" processors, if you want to avoid lags (frame rate slumps, which do not affect the average frame rate, registered in tests, but quite nasty in practice). Such phenomena appear, for example, if a computer has to process a lot of characters, including those off the screen (that is those characters who are present in a game, but currently out of sight).
Call of Juarez
This game turned out to be even a harder ordeal for the graphics cards, taking part in this article, than Crysis. Logically, we cannot speak of any CPU dependences here. There can be only one conclusion: you need a more powerful graphics card (than our present contenders), only then we may have a chance to speak of CPU dependences.
Vista Mark
As weak cards disappear from our list of contenders, performance results in graphics shell and games reach the maximum level of 5.9. However, our processors still fail to sate the test. Regardless of our changes in the line of dual-core processors, Microsoft OS still prefers the quad-core processor. Funny observation -- to all appearances, NVIDIA nForce 730a disk controller demonstrates better skills with our hard drive than the Intel P45+ICH10R controller.
Optimal configurations
As our objective is to determine which CPU/graphics card combo for about $300 (if you want to assemble a computer for $650--700) demonstrates the most attractive results in most tests, we published such combinations in a separate table (each graphics card is presented with three processors) and kept only natural tests (no synthetics).
The only configuration that fits the $300 budget is still with the 9600 GT card, if we choose a processor from the middle of the list (recommended for the above-mentioned reasons). However, as we added such games as Crysis and Call of Juarez, it becomes apparent that GeForce 8800 GT still demonstrates higher performance. If you choose among NVIDIA cards only, you should stick to this very graphics card. What concerns the CPU choice, Core 2 Duo E4700 is an apparently non-optimal solution as far as its price-performance ratio is concerned (it's more expensive than the other two processors and it's at least not faster). Athlon X2 6000+ and Core 2 Duo E7200 provide the same deal of comfort: the E7200 has an advantage only in those games, where it makes no sense to strive for a higher frame rate (however, if we have a look at SYSMark results, it must demonstrate higher office performance). In its turn, Athlon outperforms its competitor in Crysis, even if only formally. Besides, taking into account the platform in general, we must note that a motherboard for Socket AM2+ is cheaper than a motherboard for LGA775 with the same functionality (or worse). Besides, the AMD platform will have better upgrade prospects, because future processors will be compatible with Socket AM2+ for at least a year. While Intel may initiate the transition to a new socket for future processors, incompatible with LGA775, already this autumn.
A small note about Athlon X2 6000+: if you want a perfect solution, you may want to choose a processor with TDP of 89 W, which designation starts with ADA, for example, ADA6000IAA6C. This processor is supported by most existing motherboards. You may also come across an early version of the 6000+ processors with TDP=125 W (marked ADX) offering the same performance level. But before you buy this processor, you should make sure that your motherboard supports processors with this TDP.
Conclusions
Despite significant summer changes in Mid- and High-End graphics card segments, our test results prove that if you want to stay within your modest budget (about $300 for a processor + graphics card), your choice of graphics cards hasn't changed since spring. You may want to read our test results obtained in the research devoted to AMD GPUs. As for processors, no products have appeared to radically change the alignment of forces as well (that is with a new price/performance ratio).
Now good news, everyone: an inexpensive PC configuration does not quail at the 1680x1050 resolution in games with typical hardware requirements. But in order to play higher-technology games at high graphics quality, you will have to pay more.
Write a comment below. No registration needed!
