by Ilya Gavrichenkov
11/09/2005 | 06:04 PM
The end of the year is approaching and so are Christmas holidays. This is the season when many users consider buying new hardware. It has already become a tradition: the end of November-December time frame are the hottest sales season in most computer stores. So, the hardware manufacturers as well as hardware sellers fight for the user’s money and prepare new hot products alongside with exciting marketing promotions and campaigns for this time of the year. Unfortunately, it turns out more and more difficult to make the right choice especially under informational and marketing pressure from the vendors and retailers. Those users who prefer to go shopping for their computer upgrades during the last few weeks of the year have twice as high risk of becoming the victims of aggressive marketing and promotion campaigns. In this situation we here at X-bit labs cannot leave you, guys, up for grabs. In order to help you find your way among all the hardware appearing in stores we prepare large articles covering comparative tests of a wide range of products.
This article tends to be your ultimate guide for the today’s processor market, where the cut-throat competition between two major rivals, Intel and AMD, has never come to an end. This year has brought in serious changes to this market: we all know that this is the year when dual-core processors arrived. Therefore, besides the budget solutions, each of the manufacturers offers three other types of products: dual-core CPUs, CPUs for hardware enthusiasts and extreme users, and “regular” mainstream processors. In the CPU section of our site you may find detailed reviews about the individual products from each category, therefore, we decided to apply a completely new approach to writing large comparative articles, like the one you are about to read today.
And we got most help here from the game developers, no matter how strange it may sound to you. The thing is that Christmas fever also spreads into the gaming market. New gaming applications start selling better when the New Year’s Eve is approaching, that is why most software developers and publishers try to schedule the release of new gaming titles closer to the Christmas season, too. Therefore, there appeared a few fresh new games, which require much more from the hardware platforms than their predecessors. Well, we decided to test contemporary processors using these particular games. This benchmarking session will allow us to make our recommendations about the best gaming CPUs, as the gamers are actually the most active buyers of expensive hardware. These are the games we are going to use as our major benchmarking titles today:
But before we pass over to the actual tests we have to make a few comments about the way the obtained performance results should be interpreted. We have to stress that contemporary games require a powerful graphics accelerator in the first place, and the CPU is only the second most important hardware system component. So, it may seem not really necessary to test CPUs in games at first glance. In order to evaluate how fast the tested CPUs are in games, had to use the settings that lower the image quality during gameplay. In most cases the tests were run with disabled anisotropic filtering and full-screen anti-aliasing. This is done in order for the actual CPU gaming performance to move forward. Otherwise, the system performance will be limited by the power of the graphics card and will not that greatly depend on the CPU. Well, these are the today’s realities: we can say that the CPUs have left the graphics accelerators far behind from the gaming performance point of view. However, we can always look at it from a different prospective and say that the today’s game developers got too much involved into developing the exterior looks of the game, and have almost completely forgotten that very often a lot of CPU capacities simply get wasted, while they could be involved into the calculation of gaming physics environment or into the improvement of the enemies AI.
So, when you are assembling a gaming system, you should first of all choose the proper graphics accelerator and then think about a CPU. Although, if your graphics card is very fast and the CPU is pretty slow, it can certainly prevent you from taking advantage of the performance potential of the video subsystem. The thing is that while the graphics card deals with building and displaying the scene, the central processor is responsible for preparing the data for the graphics accelerator. So, the simple rule everyone usually sticks to is as follows: the more powerful the graphics card is, the more powerful should be the CPU.
To test the contemporary CPUs in the latest graphics applications we used a testbed built with one of the fastest graphics cards of today: NVIDIA GeForce 7800 GT with 256MB of graphics memory onboard. If we used a slower graphics card the performance of the today’s fastest CPUs would be limited by the graphics card potential and we wouldn’t be able to test them the right way.
Also, in order to eliminate all potential bottlenecks of the test platform we selected the mainboards that have already stood the test of time and have proven to be the best solutions for enthusiast systems. They are: Socket 939 mainboard from DFI based on NVIDIA nForce 4 Ultra chipset and ASUS mainboard based on i955X Express chipset.
Another peculiarity of our test platforms worth mentioning is the 2GB of RAM we installed. Contemporary games are very sensitive to the amount system memory, and most of them already need more than 1GB for the swap file to work properly. As a result, we assembled a few testbeds, which consisted of the following components:
The tests were carried out with the BIOS Setup of the mainboards used adjusted for maximum performance.
Battlefield 2 shooter, a continuation of one of the most PC popular military strategies played on the Net today, cannot be considered a new-generation game. It was released in summer and doesn’t really demand too much for the graphics subsystem. Nevertheless, we decided to include it into our today’s testing session for multiple reasons. Its incredible popularity is actually one of them. Another important reason the determined our decision in favor of this game is the fact that unlike many other shooters it sets pretty high requirements to the system processor, rather than graphics accelerator.
For example, if we look at the graphics card reviews on our site, we will notice that in low resolutions the graphics cards perform more or less equally in Battlefield 2, even if anisotropic filtering and full-screen anti-aliasing are enabled. This is a clear indication that the graphics processor cannot be the bottleneck of the system.
In order to evaluate the performance of different CPUs in this game, we decided to refrain from using heavy graphics modes. All benchmarks were carried out in 1024x768 with Medium gaming quality setting. Here I would like to stress that Battlefield 2 is one of those games where the maximum fps rate is locked at 90 fps. To make our analysis more correct we disabled this feature.
The diagram below represents the obtained results:
The performance results you have just seen seem to be quite logical. AMD CPUs have always been better off in gaming applications, and Battlefield 2 is no exception here. However, we would still like to draw your attention to the advantage of AMD Athlon 64 processors over their rivals from the Pentium 4 family. The thing is that the gap between them is simply huge, so that the tope Pentium 4 CPUs appear running neck and neck with the slowest Athlon 64 models. This way, it will hardly be a smart move to go for an Intel processor if you are a Battlefield 2 fan.
If we take a closer look at the obtained results, we will be able to make some additional conclusions. Just like most contemporary games, Battlefield 2 doesn’t support dual-core processors. If the system is equipped with a CPU like that, the game will only use one of the processor cores. As a result, the fps rate will not be as high as expected, because most today’s dual-core CPUs are still working at lower frequencies than their single-core fellows.
As for the amount of available cache-memory, this parameter affects the system performance in Battlefield 2 only by CPUs with K8 architecture. The performance difference between the systems built on a Pentium 4 5xx and Pentium 4 6xx processors is pretty small: no higher than 5%. The situation with Athlon CPUs is slightly different. The CPUs with 1MB L2 cache memory are faster than the CPUs working at the same clock speed but featuring half as big L2 cache and the performance difference is about 10%-15%. So, the Socket 939 platform owners should pay special attention to the L2 cache memory size of their CPUs.
Besides the actual influence of the CPU performance on the gaming experience in Battlefield 2, we decided to also take a look at the performance dependence on the amount of system memory installed. To conduct this investigation we compared the performance of two identical systems differing only by the amount of RAM: 1GB or 2GB. This test was carried out in an Athlon 64 FX-57 based system. It order to avoid the influence of the memory timings on the gaming performance, we used a pair of DDR400 DIMM modules with 2-3-2-10 timings in both cases.
We can see some performance difference, although it is really small: about 1%. However, it is just a performance difference we managed to detect during a prerecorded demo playback. In a real game you will feel the difference much better, of course, as the game will be addressing the disk subsystem. I wouldn’t state that you cannot enjoy Battlefield 2 to the full extent with only 1GB of memory installed, however, having 2GB of memory will definitely make your gaming experience much more pleasing.
The second game we selected for our today’ test session is the 3D shooter announced a few weeks ago by the Monolith Company – F.E.A.R. This game has already managed to win many gamers’ hearts and awards due to its unbelievable gaming environment and high-quality implementation. From the technical prospective, F.E.A.R. is a great measuring tool for the system performance, as it uses the system resources to the full extent.
Although the game developer recommends using a 3GHz CPU, 1GB RAM and a RADEON X800 or GeForce 6600 graphics accelerator for pleasant gaming experience in F.E.A.R., the game will not actually be that much fun on a system with this configuration. At least, if you enable full-screen anti-aliasing and anisotropic filtering together with SoftShadows, the fps rate will drop below acceptable gaming performance even with the today’s fastest graphics card installed.
According to the game developers, even though F.E.A.R. loads the graphics subsystem heavily enough, the CPU power is also very important for this shooter. To check what they were actually talking about we carried out a very simple test: we looked how greatly the gaming performance would decrease for game modes with different quality settings when the CPU frequency reduces. Note that we used not only NVIDIA GeForce 7800 GT, but also an ATI RADEON X800 PRO graphics card.
Well, these are hardly very illustrative results, I should say. Of course, you can notice a slight performance drop when the processor frequency reduces (this is, however, true for all graphics settings other than the maximal). In other words, if your graphics accelerator doesn’t work at the utmost of its power, you will see some difference in the fps rate. But one you set the graphics quality to high or maximum, the CPU speed will not affect the gaming performance at all, or will have very little influence on it. So, you shouldn’t really be that much concerned about buying a proper CPU for playing F.E.A.R.
As for our tests, we had to set the graphics quality to Medium (no anti-aliasing and no tri-linear filtering), in order to eliminate the influence of the graphics subsystem on the obtained results.
Well, we don’t see any overwhelming victory of the AMD processors here, unlike the situation in Battlefield 2. We can even state certain parity between the rivals with a slight advantage of the Athlon 64 processors. For example, the top Athlon 64 with 4000+ performance rating is only 2% faster than Pentium 4 670. We can see the same situation if we compare the performance results of the slower CPU models. Athlon 64 CPUs are just a few frames per second faster than Pentium 4 6xx processors working at the clock sped corresponding to their rating. In fact, the only reason why you see so many green bars at the top of the diagram above, is because AMD is currently offering a lot of expensive Athlon 64 FX processors working at high frequencies of 2.6-2.8GHz.
Despite the fact that F.E.A.R. is an absolutely new game, it cannot boast efficient support of dual-core processors. The major influence on the gaming performance is imposed by the L2 cache size and CPU clock frequency.
Besides the CPU speed and graphics subsystem performance, the gaming performance in F.E.A.R. is also affected by the amount of system memory installed in your platform. The formal tests carried out with an Athlon 64 FX-57 processor (the same platform and settings as in Battlefield 2) reveals a difference of about 1-2%:
However, in order to avoid constant turning to the swap file for data during the actual gaming process, F.E.A.R. you may want to go for more than just 1GB of RAM. In conclusion, I would like to stress that F.E.A.R. doesn’t really need a powerful CPU. Our tests showed that even a Pentium 4 3GHz or Athlon 64 3000+ are fast enough to make this game run at over 100fps. Although these results were obtained at the Medium graphics quality settings and the performance will definitely be lower with maximum settings, it will not go down because of the CPU. So, if you are going to spend a lot of time playing F.E.A.R., them it would be best to invest into the powerful graphics subsystem in the first place.
Although the Serious Sam 2 shooter released recently has been receiving pretty controversial feedback from the gaming community, we decided to include it into our today’s test session. On the one hand, Serious Sam 2 appeared very sensitive to the power of hardware resources, and on the other hand, it uses a lot of contemporary algorithms for physics models and AI calculations, i.e. uses a lot of CPU power. Unlike many other shooters Serious Sam 2 manages to load with plenty of work even the today’s fastest processors.
The minimal system requirements for the Serious Sam 2 game provided by the developer are very similar to those set for F.E.A.R. Although, Serious Sam 2 uses these resources in a completely different way. The calculations of gaming physics and enemies’ behavior may eat up to 30% of the time during frames rendering. This happens because Serious Sam 2 is rich in all sorts of monsters and diverse objects. As a result, the fps rate remains dependent on the CPU speed even when the graphics settings are close to the absolute maximum.
Just like in the previous case, we carried out our performance tests with Medium graphics quality settings, in order to make the dependence of the fps rate on the CPU speed more evident. However, we would like to stress that in this particular game the Medium settings imply the use of 2x FSAA and 2x anisotropic filtering.
What we see here, is a pleasing sight for AMD fans again. CPUs with K8 architecture are significantly faster than their competitors from Intel. It looks like during image rendering Serious Sam 2 uses a lot of computational algorithms, which have always been AMD’s trump. Athlon 64 processors with the performance rating over 3200+ outperform any of the Pentium 4 solutions available in the today’s market.
However, the major intrigue is not about the victory of Athlon 64 over Pentium 4. As you can see from the diagram, CPUs with two computational cores run faster in Serious Sam 2 than single-core processors working at the same frequency. It means that this game uses parallel computational threads actively enough, and it is a very big achievement, because there hasn’t been a single game yet that could boast anything like that.
Here is a CPU utilization graph for the dual-core processor during Serious Sam 2 gameplay:
This graph was taken from a system with the dual-core Athlon 64 X2 4800+ processor. As we see, the overall CPU utilization stays around 60%. It means that Serious Sam 2 also involves the second computational core. However, it would be not quite fair to praise only Croteam guys for the implementation of dual-core support. I have to point out NVIDIA drivers also started supporting dual-core architectures. ForceWare version 81.xx allows enjoying the advantages of dual-core technology in DirectX as well as in OpenGL.
Dual-core Athlon 64 X2 processors perform outstanding in Serious Sam 2. Athlon 64 X2 4800+ working at the actual 2.4GHz frequency outperforms even Athlon 64 FX-57 working at 2.8GHz nominal clock rate. Unfortunately, dual-core Intel processors cannot boast anything like that. Even though they are faster than their single-core counterparts working at the same clock frequency, this performance advantage is really tiny. Intel’s dual-core architecture is not as efficient for the fps rate as AMD’s one for some reason.
Quake 4 is one of the most impatiently anticipated games of this year. The thing is that this shooter is a continuation of the world’s famous gaming sequel. Although Quake 4 is based on a modified Doom III gaming engine, it went quite far away from its forefather. You can tell it not only by the image quality and some other details in the game itself, but also by the system requirements set by the developers. In fact, this is the “heaviest” contemporary game of all judging by the resources consumption.
Quake 4 boasts very realistic gaming physics and pretty advanced artificial intelligence of the enemies and allies. I would also like to stress that Quake 4 has the best-balanced model of hardware resources utilization of all contemporary games (although I have to admit that event his model is far from ideal). As a result, the gaming performance in this 3D shooter is affected not only by the graphics subsystem but also by the system CPU. You can even see this influence in high resolutions and maximum graphics quality settings. In particular, we managed to test the CPUs in this game even with the Ultra Quality graphics settings: even in this case we could really see how the fps rate depends on the processor speed. However, to better illustrate the scalability of the obtained results, we still disabled full-screen anti-aliasing during our tests.
All in all, the processor performance numbers in Quake 4 are quite typical. Only the dual-core CPUs kind of stand out here, however we are going to talk about them later in this article. As for the performance results of the regular single-core processors, the situation is very similar to what we see in Serious Sam 2: the performance of Athlon 64 and Pentium 4 correlates just fine with the performance rating of the former. In other words, Pentium 4 processors yield slightly to the Athlon 64 with the performance rating corresponding to the frequency of their rival. In this respect we can state with all certainty that extreme gaming CPUs, AMD Athlon 64 FX, do not have any competitors in the Intel’s camp, because they are confidently ahead of all of them including Pentium 4 Extreme Edition.
However, the most interesting and unexpected results during Quake 4 test session appeared to be those of the dual-core processors. The graph below shows how the two CPU cores get loaded during Quake 4 gameplay. The graph was taken from the Athlon 64 X2 4800+ based platform.
As we see, even though Quake 4 originated from Doom III, where dual-core technology wasn’t supported at all, we can see a truly high-quality optimization of the gaming engine for the multi-core architectures. Looks like the software developers didn’t waste their time and modified the engine according to the latest innovations in the hardware world.
As a result, the maximum performance in Quake 4 was demonstrated by dual-core processors from AMD, which turned out even faster than even Athlon 64 FX. Maybe AMD engineers should start thinking about transferring their high-end solutions for wealthy hardware enthusiasts to dual-core architectures? Of course, the number of games benefiting from the two computational cores in a system will keep growing. And as we have seen in Quake 4, multi-core architectures can be quite efficient in games.
Even though the fps rate in Quake 4 depends a lot on the CPU speed, we have to admit that even the not very fast (according to today’s standards) Athlon 64 3000+ and Pentium 4 3.0GHz can offer quite acceptable performance numbers in this game. If the graphics subsystem is powerful enough, of course.
Besides different CPUs, we have also checked how the doubling of the system memory can affect the performance in Quake 4. The results obtained on a platform with Athlon 64 FX-57 were the following:
Formally, we cannot state that the gaming performance speeds up once the amount of system memory doubles. However, we have to mention that you can feel the gaming comfort improve noticeably as the game doesn’t request data from the hard disk drive as often any more, which is exactly what we have already pointed out during analogous experiments in other games.
In conclusion I would like to say that since Quake 4 is based on a modified Doom III engine, these games have a lot in common. However, Quake 4 leaves the impression of a better-planned shooter, so that it works slightly faster than its predecessor on the same test platforms and with the same settings. However, this is my subjective opinion, that may not be the universal truth :)
The last game we are going to look at today is Call of Duty 2. even though it is based on Quake 3 engine, it looks very up-to-date and moreover, has very strict system requirements. Although, this has mostly to do with the graphics subsystem, of course. For some reason, the game developers freed the CPU of its usual routine tasks: the enemies behavior models are managed by scripts and not by the game AI, besides the physics is not calculated as thoroughly as in many other contemporary shooters. The results are obvious: the gaming performance in Call of Duty 2 is primarily limited by the graphics card and hardly depends on the CPU power.
Our tests prove this statement with all certainty. We carried out the tests in 1024x768 with disabled full-screen anti-aliasing and medium level of detail.
I would like to draw your attention to the fact that the overall fps rate is quite average in this game in general. As we see, it is not the slow processors that we should blame for that, as the processors performance doesn’t scale down that much in Call of Duty 2. The game must be very demanding on the graphics subsystem. Therefore, the processors do not have that much influence on the fps rate as in other games we have already discussed.
However, we can still observe certain tendencies. First of all, I would like to stress pretty good results of the Pentium 4 processors in Call of Duty 2. We can actually state that they are running on equal terms with Athlon 64. And this is a true wonder for the gaming applications. Pentium 4 3.8GHz CPU appears faster than Athlon 64 4000+ and Athlon 64 3800+, yielding only to Athlon 64 FX solutions.
The second surprise is good dual-core architecture support. Just like in Quake 4, Athlon 64 X2 and Pentium D outperform single-core CPUs working at the same clock speed. The additional core in Athlon 64 X2 allows them to catch up with the single-core K8 processors working at about 200MHz higher clock speed. If we try to make the same analogy for the Pentium D processor, the “cost of the second core” will get as high as additional 600MHz.
Here I would like to stress that Pentium Extreme Edition 840 performs pretty weakly in Call of Duty 2, as well as in Quake 4. Each physical core of this CPU consists of two virtual cores managed by the Hyper-Threading technology, however, the processor is still defeated by the dual-core Pentium D 840 with no virtual cores at all. The modified Quake 3 and Doom III gaming engines seem to be optimized for multi-threaded calculations performing them in two parallel streams. However, the task manager cannot distribute these streams to actual physical cores of the Pentium Extreme Edition 840. These streams seem to be sent to the virtual cores, which are evidently considerably slower.
Well, it’s time we made some conclusions. The processor tests in new gaming applications once again proved that AMD CPUs boast the best architectures for gaming needs. Almost in all contemporary shooters K8 CPUs provided better results. And sometimes the performance difference between AMD solutions and the rivals reaches unbelievably high rates. The maximum CPUs with NetBurst architecture from Intel can boast is a slight lag behind AMD processors in a few selected games, while in other games the performance rates may be dramatically low. So, the major statement remains the same: gamers should go AMD.
However, the mere fact that there are a lot of CPU types in the today’s market, even within the same price group, can be quite confusing for the PC users. While it is not that hard to decided between the Athlon 64 and Pentium 4 processors offered at the same price point, the dual-core solutions may appear a stumbling stone. Some time ago we could simply mark them as “not for gamers” and we could be absolutely right about it. Their working frequencies are lower than those of their single-core fellows, and the actual advantages of having a second core haven’t been used by the gaming applications at all. Now the situation has changed a lot. Graphics processor developers released new driver versions that can support multi-threading, and game developers began to modify their engines so that they could take good advantage of dual-core architectures. As we see, new shooters based on gaming engines from id Software can already use the power of two processor cores. Since these gaming engines are among the most popular nowadays, Quake 4 and call of Duty 2 may very soon become far not the only ones using dual-core. Moreover, it wouldn’t be surprising at all if some other gaming engines also acquire dual-core support: progress keeps going forward.
Therefore, the question about the best gaming CPU starts to seem pretty hard to answer turning from a practical matter into something from the applied philosophy area. However, let’s not give in and try to find an answer anyway.
Having analyzed the obtained results we can state that any of the contemporary CPUs will be good for games. Let’s take a look at the results once again. Even though all our tests were carried out in real applications are still somewhat synthetic. In order to reveal the performance dependence on the CPU speed we had to set lower resolution, disable anti-aliasing, reduce the textures quality, etc. In these testing conditions all CPUs provided more or less acceptable fps rate. Some processors were faster, some were slower, however, in real gameplay with real graphics quality settings any gamer would use all this advantage will disappear. This is because the graphics quality and other gaming settings are usually determined by the graphics card potential. By increasing the quality settings, the fps rate will drop down to 40-60 fps, which is ok for normal gaming experience. And you know, any Pentium 4 CPU with the actual working frequency of 3.0GHz and up and any Athlon 64 with the performance rating of 3000+ and up can process that number of frames per second, as we have already shown in our tests. In other words, in real gaming conditions the performance will still be limited by the graphics processor, and not by the CPU.
I have to stress that we arrived at this conclusion having one of the today’s fastest and most powerful graphics cards in our system: NVIDIA GeForce 7800 GT. And if this powerful graphics card doesn’t require a super-fast CPU, then what can we say about the mainstream graphics solutions? It means that gamers with the mainstream or slower video subsystem shouldn’t even think of getting a powerful CPU: it will be just a waste of money.
So, we can state that games are not the applications you should look at when shopping for the new CPU. There should be some other applications involved, and each of you should define the set of tasks for himself.