by Ilya Gavrichenkov
08/24/2004 | 11:16 PM
The new game from id Software made us all reexamine and revise our hardware. A really powerful computer is needed for you to enjoy fully the superb graphics of Doom 3. For example, the recommended system configuration for this game lists a 3GHz CPU (or with a rating of 3000+), at least 512MB of RAM and a graphics card no worse than NVIDIA GeForce 5900/ATI RADEON 9700.
It’s no surprise then that many devoted gamers have faced the problem of upgrade yet another time now. We have already published two reports about the speed and graphics quality as provided by modern graphics cards in this game. You can read them by the following links:
However, we have hitherto omitted the problem of the central processor in Doom 3. Of course, it is the graphics card that affects the speed of this game most of all, but still, Doom 3 is a CPU-dependent game as you can see from the test results of graphics cards in simple graphics modes.
So, in this review we will examine the influence of the CPU and of the amount of system memory on the speed of Doom 3.
Before showing you the results of our tests, let’s first discuss the uses the developers of the game had put the CPU to. So, from the numerous interviews of John Carmack, id Software’s Technical Director, at various Internet sites it seems that skinning and shadows are the CPU’s main responsibility in Doom 3. That is, the CPU calculates the positions of the characters’ “skins” as they move between certain key points. This approach helps to animate the characters as well as render their facial expressions.
Evidently, the realization of dynamic shadows is thus better implemented using the CPU resources, too. That’s why Doom 3 requires an assiduous CPU to run smoothly, as besides skinning and shadow generation the CPU has to solve the traditional problems of any game like monsters’ AI, the physics of the game scene and tangent space transformations, necessary for a correct depiction of the game events on the screen.
It’s rather easy to determine which system component – the graphics card or the CPU – is the bottleneck as concerns running Doom 3. There’s the r_skipRenderContext variable among the game engine options. Set to 1, it changes all calls of OpenGL functions to calls of zero functions. In other words, it “disables” all the game graphics. Comparing the speed of a demo record with and without graphics we would understand what the system’s brake is in each particular case. But we are going talk about our benchmarking methods in the next section of this report.
First of all, we gave up testing with disabled graphics. Although the entire load would lie on the CPU in this case, we wouldn’t be able to claim the superiority of this or that processor in real game conditions. The OpenGL driver also uses CPU resources, so although testing with r_skipRenderContext=1 is an excellent synthetic benchmark, it doesn’t produce a clear picture of what we really have when playing Doom 3. So, we ran all the tests with the game graphics enabled.
To minimize the effect of the graphics subsystem on the end result, we used one of the fastest graphics cards of today, based on the GeForce 6800 GT GPU. According to our previous reports, graphics cards on the newest GPUs from NVIDIA run Doom 3 faster than their rivals from ATI.
The settings of the game itself don’t practically affect the results of CPU tests – the main thing is having dynamic shadows enabled. So all the tests were run at the Medium Quality settings in 1024x768 resolution.
We had Intel’s processors with frequencies of 3GHz and higher, and AMD’s processors with ratings of 3000+ and higher for our tests. Of course, we tested Intel’s CPUs on the two available platforms, i875P and i925X Express. Thus we will also be able to evaluate the innovations implemented in the i925X Express chipset like DDR2 SDRAM and the new graphics bus PCI Express x16.
So, we used the following hardware in our test systems:
We performed our tests in Microsoft Windows XP with Service Pack 2 and DirectX 9.0c installed.
The standard shipping version of Doom 3 comes with a prerecorded demo demo1, which a majority of users and reviewers employ for benchmarking purposes. That’s why we began to test our CPUs in Doom 3 with this record. The demo is launched with the console command timedemo demo1 (to open the console, press Ctrl + Alt + ~).
The results are very informative. First of all, the Athlon 64 family processors boast a significant advantage over the competing CPUs from Intel. The most expensive and fastest Pentium 4 Extreme Edition cannot compete with the topmost members of the Athlon 64 series, not mentioning the Athlon 64 FX. As for the old Athlon XP models, they behave rather poorly. Even the fastest Athlon XP falls far behind the Pentium 4 as well as Athlon 64.
We can also see here that Intel’s new platform on the i925X Express chipset with PCI Express x16 and DDR2-533 SDRAM doesn’t guarantee a higher performance compared to the time-tested i875P platform. At the same time, the Prescott-core processors with SSE3 support and a L2 cache of 1MB are tangibly faster in Doom 3 than their predecessors on the Northwood core.
The CPUs of the Athlon 64 family have ranked up according to their rating. Irrespective of the differences in the L2 cache size and in the memory controller, CPUs with higher ratings are faster than those with lower ratings. The most expensive Athlon 64 model, the FX-53, quite naturally notches the best result of all the tested processors. In fact, we could have ended this review announcing the victory of the Athlon 64 family in our Doom 3 test. But this would be rather hasty on our side.
If critically examined, the demo1 record puts a rather small load on the CPU. And really, the CPU does skinning and dynamic shadows in Doom 3, so there should be more moving characters and many light sources, preferably moving too, like flares from weapon shots. We have little of this in demo1. That’s why we decided to record our own demo that would load the CPU more.
We played the game long to find a level where there would be many monsters to record a high-CPU-load demo. As a result, we stopped at the Central Server Banks location on the CPU level. There are many various monsters here, and we recorded a battle with them. The test results with this demo were quite unexpected:
Our demo that has a higher monster count and loads the CPU much more than the standard one (we made sure of that by disabling graphics and witnessing a very small fps rate growth) changes the situation dramatically. Now the Pentium 4 Extreme Edition is the winner, not the Athlon 64! Then, i925X Express systems are faster than i875P-based ones – this is yet another inconsistency with the previous results. As for the correlation between the processors on Prescott and Northwood cores, it persists: the frequency being the same, the Prescott is faster.
Among the Athlon 64 models, you can notice that the memory subsystem bandwidth influences the result more in monsters-heavy game scenes. For example, every Athlon 64 with a dual-channel memory controller outperforms an Athlon with a single-channel memory controller. The amount of cache memory, on the contrary, doesn’t practically affect the result. The speed of the Athlon 64 FX-53 is close to that of the Athlon 64 3800+, while the Newcastle-core Athlon 64 3000+ is but slightly slower than the Athlon 64 3200+, which has a twice-larger L2 cache.
Well, the cache size matters with Pentium 4 models. Particularly, the Pentium 4 Extreme Edition 3.4GHz is faster than the Prescott-core Pentium 4 3.6GHz.
Comparing the results in two different demo records one is left doubtful: what processors are faster in Doom 3? We need one more test to answer correctly.
The so totally different results of the first two tests made us delve deeper into the specifics of CPU benchmarking in Doom 3. And we realized that it was wrong to measure the CPU performance in Doom 3 using pre-recorded demos because the results wouldn’t reflect the real game process. The problem with the demo records in Doom 3 is in their not containing only the player’s actions but also the reaction of the outside world and the monsters to these actions. That is, part of the load that exists in real play is removed from the CPU when it is just playing a demo back. The CPU does skinning, shadows and tangent space transformations, but does not do the monsters AI and the physics modeling of the world when playing a prerecorded demo. So, the results we got in the two above tests as well as the results of many other reviewers can be justly called into question.
We had to do without demos to measure the real performance of CPUs in Doom 3. So we played the game for a while looking for moments when the CPU was under the highest load. First of all, we’d like to remind you that the rendering speed has a ceiling of 60 fps in Doom 3. In other words, there’s no sense in going for a faster processor if your current one can maintain an fps rate of 60. As we found out, the fps rate with the Medium Quality settings and with a powerful graphics card rarely falls below 60, i.e. CPU models with 3GHz and higher frequencies or 3000+ and higher ratings guarantee an acceptable performance.
However, you may sometimes experience a slowdown when the CPU is incapable of rendering those 60 frames in a second. This usually happens when there are many monsters in the same room with the player – calculating skinning and AI for the monsters puts too heavy a load on the CPU. This observation led us to the right method of measuring the CPU performance in Doom 3, in real gaming conditions.
The point of our method is in arranging a get-together of a lot of monsters in the same room and measuring the fps rate with the Fraps utility. Again, we returned to the Central Server Banks on the CPU level. Having turned on the “immortality” with the god command and having run a couple of circles around the level, we managed to get together a score or more of monsters of various kinds and characters. We then saved the game and read this save file on each of the tested systems to compare their performance in real game conditions and in one of those moments when the CPU has to crunch through a huge amount of calculations, rendering skins and shadows, modeling AI and physics. We were measuring the fps rate for three minutes and within this interval half of the monsters had killed the other half, releasing the speed to the ceiling of 60 fps. We do assert that the results we got this way allow for a more correct comparison of CPU performance in Doom 3.
First of all, here are the minimal fps rates we got with different processors:
As you see, the speed can drop much below the 60fps ceiling – some processors even get below 30fps, which is the empirical minimum, necessary for a comfortable play.
The overall picture is typical: we can’t name a definite winner. The loser is certain, though. The Athlon XP family suffers a fiasco in this test.
The results explain where the recommendation to have a processor of 3GHz frequency or 3000+ rating comes from. Note also that the Pentium 4 3GHz work faster than the Athlon 64 3000+ or even the Athlon 64 3200+. However, the top-end members of the Pentium 4 and Athlon 64 families have similar results.
And here are average fps rates for the same test:
The overall situation remains the same: the Athlon 64 models line up in order of their ratings. The Prescott-core Pentium 4 models are faster than their Northwood-core mates, the frequency being equal. The i875P platform turns to be faster than the i925X Express.
So, our “right” testing in Doom 3 suggests that there’s no CPU architecture best suited for this game. The results we got with demo records produce quite another picture, though, and that means the ordinary timedemo method is not the best or correct way of benchmarking CPUs in Doom 3.
The last small test will be about the memory amount you may want to have to play Doom 3 comfortably. More precisely, we will try to see the difference between 512MB and 1GB of memory. Is the recommendation to use 1GB a well-grounded one? We performed the test on the i975P-based system with a Pentium 4 3.4GHz processor (Northwood core). There’s no demand for a “real-game” test in this case, so we only ran our own demo record:
As you see, 512MB of system memory is enough for all the quality settings, except the maximum ones. The amount of textures the game uses in this mode exceeds 512MB and the game honestly warns you about that as you try to enable this mode. Otherwise, half a gigabyte is enough for a comfortable play.
So we can draw two points from our today’s tests. First, the standard test method that measures the average fps rate when playing a demo cannot be used for a correct comparison of CPU performance in Doom 3. More precise results can be achieved in real gaming environment only. Second, our tests don’t reveal a clear performance leader among the topmost models of the Pentium 4 and Athlon 64 series. It would be imprudent to claim that a particular CPU architecture suits better for playing Doom 3: the Athlon 64 FX-53 is about as fast as the Pentium 4 Extreme Edition and the Athlon 64 3800+ equals the Pentium 4 560. We could continue drawing parallels longer, though.
The older Athlon XP CPU family cannot compete with the new Pentium 4 and Athlon 64 models anymore. In real tests we encountered situations when the fps rate got below the comfortable value of 30 with Athlon XPs. By the way, the same can be seen with Athlon 64 3000+ and 3200+ CPUs. That’s why we recommend you to use Pentium 4 models with frequencies from 3GHz up and Athlon 64 models with ratings from 3200+ up, if you want to be absolutely warranted against any accidental slowdowns.