Scene in Wireframe mode
Scene in Smooth + Highlights mode
Scene after the Final Rendering is complete
by FastSite
05/21/2001 | 12:00 AM
In the previous article we found out what graphics card suits best of all for work in 3D MAX application. However, it is not only the graphics card performance that tells on the fastness of your 3D MAX software, but also HDD, mainboard, CPU and even the operation system may exert some influence. If you can avoid having a hard disk drive slow down your system in 3D MAX by adding more RAM, then the mainboard and CPU as well as the OS you are working in may be of the same importance as the graphics card. That is why in this article we will consider the performance of different operation systems and platforms and try to find out the best hardware and software configuration for 3D MAX needs.<%BANNER[article]%>
Choosing an operation system for 3D MAX today is not an easy task to fulfill. About a year ago there was no problem with it, since Windows 2000 wasn't born yet, the gaming graphics cards drivers (except the drivers for NVIDIA based ones) didn't allow working in 3D modeling applications under Windows 98, and the drivers for professional graphics cards were intended exclusively for Windows NT. And now you have to choose between the "gaming" Windows 98, merited Windows NT 4.0 and "relatively young" Windows 2000. Each of them has something to boast.
Windows NT 4.0 has initially been a system intended "for work only". The developers advise to use most 3D modeling applications particularly with this OS. Again, OpenGL drivers of the graphics cards intended "for work" are also optimized for this operation system. Besides, Windows NT 4.0 has been in the market for a considerable while already, much longer than all other systems discussed in this article. And it means that both: hardware and driver developers had a lot of time to optimize their products for this OS.
Windows 98 is regarded mostly as a "home" easy-to-use system. The first 3D MAX versions couldn't work in this OS correctly, because 3D MAX developers didn't take Windows 98 seriously. However, Windows 98 has one feature absent by Windows NT: DirectX support and hence the possibility to work in 3D MAX with the Direct3D driver instead of the OpenGL one.
Windows 2000 combines stability and reliability of Windows NT with the Direct3D support. But this operation system is still "too young", i.e. the drivers for Windows 2000 are not as well-optimized as those for other systems.
Last time we tested all our graphics cards under Windows NT. Since the possibility to work via Direct3D driver is the major difference of Windows NT from Windows 98 and Windows 2000, let's see if this feature is a true advantage or not.
We ran the tests on a typical scene over 6MB in size. It contained 40 thousand polygons, 6 light sources and a great lot of reflective surfaces.
At first let's compare the Direct3D drivers of different graphics cards with one another and see if their fastness differs greatly. The testbed was configured as follows:
We used the following software:
We installed all the operation systems one after another. In between, the HDD was formatted.
Here and further on in the article (as well as in the previous one) all the measurements were taken in fps, so the X-axis of all the graphs denotes the fps rate (if not stated differently).
As we can see, the GeForce2 GTS owners will hardly find working in 3D MAX with Direct3D driver a beneficial thing. The performance drop makes about 50% for all modes and in both OS. Since the results shown by one graphics card, even by the best one, do not give us any proper idea of the results we can squeeze out of other cards, let's pass over to Matrox Millennium G450 to find out more.
Although in Wireframe mode the performance drops nearly twice, in Smooth + Highlights it triples! Since Smooth + Highlights appears absolutely unsuitable for work because of extremely low speed, the increase in Direct3D could give Matrox G450 some chance, if it were not for one "but": in Windows 98 the fps rate is very unstable, which makes you feel somewhat uncomfortable.
The situation here repeats that we have just seen with GeForce2 GTS. We can hardly profit from the use of Direct3D because the performance drops twice here.
Having run the tests for all graphics cards we can see that Direct3D driver appears advantageous (which also can be argued about) only for Matrox Millennium G450 only in Smooth + Highlights mode (and Smooth only accordingly). All other graphics cards in all modes and under all operation systems showed that Direct3D driver slows down the performance quite tangibly.
Now that we have already considered the fastness of Direct3D driver, let's see what we can get in terms of image quality.
Let's discuss the positive outcome in the first place. As we found out, Direct3D drivers of the graphics cards most likely create texture miniatures in the viewports with the color depth set in the desktop properties:
Click on the screenshot to enlarge. In the right part of the image you can clearly notice the dithering effect.
In 32bit color mode there is no dithering any more.
Now let's take a look at OpenGL:
We've got the impression that OpenGL driver created miniatures in 256 colors. So, it is the indisputable advantage of Direct3D. But it is the first and the last advantage we managed to discover. Now we are passing over to drawbacks.
We revealed two major drawbacks of Direct3D drivers when displaying the viewports. The first one is the absence of textures correction.
Looks awful, don't you think so? Most boards are crooked.
And now take a look at the "correct" viewports rendering by OpenGL driver:
Here the things are alright, as you see. Since you can't see the textures laid the right way, it appears impossible to work in Smooth and Smooth + Highlights modes. So, no work connected with laying textures can be done in this case. But this is far not the end. In Direct3D most transparent surfaces are displayed as non-transparent. At first comes the scene as it should be seen:
There is a table in the middle lit by a light cone. For this "light cone" the designer use a geometric cone with the set transparency level of its sides. OpenGL driver displays this transparency correctly that's why we can see the candles and the table through this cone. And now here is the same scene displayed by Direct3D driver:
Well, instead of the light cone we see some kind of wall, which appeared because Direct3D driver turned out unable to display transparent surfaces correctly.
But this is also not the end. This driver also can't display the objects lit by multiple light sources. Here you are:
OpenGL:
Direct3D:
So, despite some quite doubtful advantages, Direct3D driver features a couple of really serious drawbacks, which make it impossible to work fruitfully. And taking into account that OpenGL drivers are considerably faster, we don't see any sense in working with 3D MAX via contemporary Direct3D drivers. Before we pass over to testing different graphics cards in different operation systems, we suggest that you should read the benchmarks description, if you haven't read our previous article "Contemporary Graphics Accelerators Tested in 3D MAX", where we have already discussed these benchmarks in detail.
The first benchmark is a certain "stress-test". It plays an animated scene in four viewports at the same time. However, the rendering methods set in these viewports are different. The scene in two upper viewports is in "Wireframe" mode, in the lower left viewport - in "Smooth + HighLights" + "Edged Faces", and in lower right viewport - in "Smooth + HighLights".
Also we enabled Anti-Aliased Lines for viewports working in Wireframe mode. Since only upper viewports worked in this mode, the thing didn't influence the lower viewports at all. This scene contains very few polygons, only 24 thousand. However, since the animation was played in all four viewports simultaneously, the resulting fps rate appeared not very high.
The second benchmark represents a scene with 7 standard primitives, which make the scene complexity equal to 10 thousand polygons.
There are six static objects in the scene and the seventh one is moving slowly across the entire scene passing through other objects. This benchmark checks if the intersecting objects are displayed correctly and if the graphics card and the driver cope with this task fast enough.
The scene from the third benchmark shows a ball, which is moving very slowly against the background geometry made of 15000 polygons.
The ball doesn't cross any other objects. Since it moves very slowly, then the "perfect" driver will make very few changes to each further frame. In other words, this benchmarks checks if the graphics card is capable of drawing anew only those objects, which got really changed.
This benchmark shows if the graphics card is good at processing very complex geometry in Smooth + Highlights mode.
The fifth benchmark is devoted to testing the graphics cards abilities in terms of processing complex geometry only. This time the amount of polygons nearly doubled and made 376 thousand. The same surface, as in the previous benchmark scene, is now covered with buildings.
This benchmark can easily bring any graphics card to its knees: the average fps rate hardly reaches 3 frames. However, you should bear in mind that this is just a benchmark testing the card's geometric performance. The file was surely created not with the 3fps. Each building was designed in a separate file and when it was added to the entire scene, some part of the geometry was disabled in order to increase the performance.
Having taken a look at the graphics cards performance when working with geometry, we suggest passing over to the imitation of multiple light sources. This test offers 8 SpotLights, which are constantly moving and lighting some kind of asteroid.
We should point out that imitating the effect made by SpotLights is a much more resource-hungry process than the imitation of Omni or Directional lighting.
Here we see the same asteroid, but this time it is lit by 8 Directional lights. Directional lights are slower than Omni lights, but faster than SpotLights.
One more time we see the same asteroid with the lights coming from 8 different sources. But now these are all Omni lights, the fastest lights in 3D MAX.
The ninth benchmark is a scene with light geometry, including only 4500 polygons, which occupies the entire viewport. It is aimed at testing the rasterizing speed.
When the camera is moving, the graphics card should rasterize big and small polygons (relative to the screen size).
The next benchmark is devoted to work with textures. The file contains a lot of textures and very little geometry (224 polygons). As for the benchmark, it is just a rotating spherical polygon with the facets covered with 48 different textures.
Very little geometry and many textures involved give us a perfect idea of how fast the graphics cards can process these textures.
This is a fully textured room with a camera moving inside. This benchmark is very close to real applications because it has a lot of textures, not very simple geometry and several light sources. It shows what graphics cards are capable of when processing pretty complex scenes in Smooth + Highlights mode.
Animated "waves" with the 114KB texture laid over them show how fast the card can deform very light geometry and modify smaller textures.
This benchmark runs with different speeds in the Wireframe mode. 111 thousand polygons in Wireframe mode will be a really tough test for any modern graphics card. Just as in the very first benchmark we enabled Anti-Aliasing here:
All the benchmarks described above are recommended by 3D MAX developers. However, as we have already seen, they are aimed at testing different functions and their implementation separately from one another. Since there are no "general" tests there, we decided to add one more benchmark to this set: a scene with 8 light sources, 61371 polygons and a great lot of transparent surfaces. The file with all textures makes 6MB total size and its complexity is quite typical of the today's 3D projects. We also included some animation to provide more realistic testing conditions: the camera is moving around the room capturing all objects. This is how the first frame looks after the final rendering is done:
We used this scene to test graphics cards in Wireframe mode as well as in Smooth + Highlights mode. As a result we've got two benchmarks:
The scene in Wireframe mode:
The same scene in Smooth + Highlights mode:
Well, since we have already proven that Direct3D driver appears totally useless in 3D MAX, we will study the graphics cards performance in Windows 98, Windows 2000 and Windows NT only with OpenGL driver.
The testbed was configured as follows:
We used the following software:
We installed all the operation systems one after another. In between the HDD was formatted.
In all the tests carried out for GeForce2 GTS, Windows NT proved faster than its competitors. In the first group of tests checking "the functions support" (benchmarks 1-3) the results obtained under Windows NT were just a bit greater than those obtained in other OS. In geometry processing (benchmarks 4-5) the difference between them grew up to 20%. And in tests processing multiple light sources (benchmarks 6-8), Windows 98 suffered a total fiasco: the performance dropped by about 50%! Windows 2000 performed more or less nicely in these tests most probably because it is built on the Windows NT core. The same thing happened in the rasterization test (benchmark 9) - a failure for Windows 98 and only 2-3% lagging for Windows 2000. In texturing tests (benchmarks 10-12) Windows 98 also occupied the last place, and Windows 2000 fell behind Windows NT only by a few percents in all the tests but one. In Wireframe mode (benchmark 13) the situation has changed a bit. Now Windows 98 and Windows 2000 show very close results and Windows NT outperforms them by about 15%. The same state of things can be found in benchmark 14, which shows the performance in real scenes in Wireframe mode. In benchmark 15 showing the performance in typical scenes in Smooth + Highlights mode Windows 98 again appears the last one, then comes Windows 2000 and at the head is Windows NT.
As a result, we see that the preferable OS for GeForce2 GTS appears Windows NT. Windows 2000 proves about 10% slower (on the average). And when working under Windows 98 the performance drops by over 50%. By the way, when we took the results in Windows 98 for GeForce2 GTS based graphics card, the max and min fps rates were very unstable, though the average rate behaved normally.
The situation with RADEON working under various operation systems is totally different from that with GeForce2 GTS. The results are almost the same in all the systems and the difference makes not more than 5%. However, in general Windows NT appears a little bit ahead anyway. The only exception is the texturing benchmark number 12, where Windows 98 fell 30% behind its rivals.
G450 shows very close results for all the three OS, which differ by 5% at the most. However, in most cases the board performs best of all under Windows 98: just look at the last two complex benchmarks. But, it is hard to work with G450 in this operation systems because of very unstable fps rate. That is why the optimal choice for this graphics card appears Windows NT.
Voodoo5 drivers are definitely not intended for work in professional applications. In Windows 2000 we couldn't work in 3D MAX because the driver displayed the program interface with numerous errors. And the main problem was noticed where there were several viewports with different rendering modes open. Namely, in some viewports there appeared some phantom objects from other viewports, which hindered proper work. In Windows NT the drivers failed to display even a single texture from benchmark 10, and anti-aliasing in all OSs turned out simply unsatisfactory (if you need more details on Anti-Aliasing, check our previous article devoted to 3D MAX).
Returning to the card's performance in different operation systems, we should point out that we obtained similar results in Windows 98 and Windows NT (and they proved equally low compared with those we got for other graphics cards).
Summing up the results of our testing in Windows 98, Windows NT and Windows 2000 we can say that for NVIDIA GeForce2 GTS based graphics cards Windows NT appears the most preferable choice, because the results obtained under this OS are almost 50% higher than the results under any other system. In case of other graphics cards tested, the performance in different operation systems will not be a determinative when deciding on this or that OS, because the performance difference is very small. However, it should mentioned that the highest results were still achieved under Windows NT. It is quite easy to explain this thing: Windows NT has been for a long time in the market already and is positioned as an operation system for "business needs". That's why the graphics card manufacturers had enough time to optimize their professional OpenGL drivers for this OS.
Well, having found out that Windows NT is preferable OS when working in 3D Studio MAX, we are passing over to testing different hardware platforms.
In our tests took part the following platforms:
The graphics card and the HDD were the same for all platforms tested:
The software:
At first we will study the dependence of the performance in 3D MAX viewports on the FSB speed. For this purpose we took the results obtained for the platform built with AMD Athlon 1.2GHz + ABIT KT7A + 256MB PC133 SDRAM working with 100MHz and 133MHz FSB.
As you may see, the FSB frequency has a great influence on the performance in viewports unlike the final rendering. In benchmarks with complex geometry the difference makes about 20%. Therefore, when choosing a platform you'd better select the one with the faster FSB, which works for both: Athlon and Pentium III CPUs. this is exactly why we suggest studying carefully all the results we got for our platforms in 3D MAX.
And now take a look at the performance of systems based on different CPUs. To give you a clearer idea of what the situation looks like we split the benchmarks into several subgroups (see the description of benchmarks above).
Benchmarks 4 and 13 present one and the same scene with mountains, though the rendering method in benchmark 4 is Smooth + Highlights while in benchmarks 13 it is Wireframe. The benchmarks run in Wireframe mode (the second result corresponds to the performance with the enabled anti-aliasing) is the test, which is most dependent on the graphics card, because the card displays the polygons calculated by the CPU without texturing them or applying any lighting effects. Benchmark 5 shows the same mountains and buildings, which doubles the geometry complexity in Smooth + Highlights mode. And what do we see? The best performance in the most processor-dependent benchmark 13 belongs to Intel Pentium 4 CPU, and the worst result was demonstrated by Intel Pentium III. In general, the systems got rated by the FSB frequency, starting from the slowest and ending up with the fastest.
In lighting benchmarks the CPU doesn't need to prepare polygons since the lit object isn't moving anywhere and lighting is calculated by the graphics card. Therefore, the results shown by all processors are identical. We still decided to present these results in order to show you what type of scenes doesn't care about the platform fastness at all.
In geometric benchmarks everything depended on the CPU, and the graphics accelerator simply displayed the ready frames, and the lighting benchmarks, where the graphics card was responsible for everything, were just the opposite. However as it came to texturing benchmarks, they appeared the happy mean. In these benchmarks the CPU and the graphics card are almost evenly occupied, because they need to modify the geometry, to lay the textures and to apply lighting effects at the same time (see the benchmarks description above for more details). Just like in the first subgroup, Pentium 4 processor took the lead and the platform built on AMD Athlon with DDR memory appears very close to the leader. Surprisingly low results belong to dual-Pentium III system, which was beaten even by a "single" Athlon CPU.
This group of benchmarks includes two: in benchmark 14 the scene is displayed in Wireframe mode (the second result corresponds to enabled anti-aliasing), in benchmark 15 the same scene is displayed in Smooth + Highlights mode. Since the scene used in these two benchmarks is very close to real applications, then it would be better to draw any conclusions about this or that platform basing on the results shown in this particular group of tests. Again Pentium 4 is ahead of all, however, this time the gap between the competitors is considerably smaller.
So, judging by the results obtained for different platforms, we see that in most scenes the memory bus bandwidth plays a very important role. Mostly due to the fastest memory bus, Pentium 4 wins in all the benchmarks. Dual-Pentium III based platforms also perform quite well even though they support PC133 SDRAM. To tell the truth, a dual-processor system features one very cool advantage: two times faster processor speed, which is very noticeable in geometric benchmarks loading exactly the central processor. Athlon based systems appeared in the middle of the rating and Pentium III 1GHz - in the very end, because it can boast neither fast FSB nor high CPU performance.
Well, having discovered that the best graphics card for work in 3D MAX appears any solution based on NVIDIA GeForce2 GTS, the best OS is Windows NT and the best platform for the same purposes is built on Intel Pentium 4, we are going to see which one is stronger: an elephant or a whale? The whale in our case will be hardware viewports rendering carried out by NVIDIA GeForce2 and the elephants role will fall upon Pentium 4 processor doing software rendering of the viewports. In fact, this hypothetic battle between the whale and the elephant isn't so senseless as it may seem at first sight: software rendering in 3D MAX has been optimized starting from the very first version and Pentium 4 proved very powerful in our tests. To make the comparison complete we added one more participant here - a platform built on Intel Pentium III 1GHz CPU with GeForce2 GTS card onboard. Now we'll see if Pentium 4 software rendering will be able to outperform hardware rendering effected by a Pentium III + GeForce2 GTS combination.
Configuration 1:
Hardware:
Software:
Configuration 2:
Hardware:
Software:
Configuration 3:
Hardware:
Software:
In all geometric benchmarks Pentium 4 software rendering is faster than hardware rendering by Pentium 4 + NVIDIA GeForce2 GTS combo. Moreover, the platform with Pentium III and GeForce2 GTS is twice as slow even. In texturing benchmarks and lighting benchmarks the software rendering is pretty slow.
Although designers major task implies creating his own world in the viewports, he sometimes needs to take a look at the scene or its parts after the final CPU rendering is complete in order to proceed with the work. Again, one day the project is finished and the last thing the designer may need to do is to carry out the final rendering. Therefore, not only the performance in viewports is worth considering but also the fastness of the scene final rendering offered by different platforms. In fact, the difference between the CPU workload when working in viewports and that when doing the final rendering is enormous. In the viewports the CPU just calculated the vertex coordinates and submits this data to the graphics accelerator, so all further work, such as texturing and lighting, is done by the graphics card. During the final rendering the CPU does all the calculations for the entire scene without the graphics card being involved. The card is only displaying the outcome on the monitor.
Well, we tested the following platforms:
The graphics card and the HDD were the same for all platforms tested:
The software:
To test the final rendering speed we resorted to a standard file from the 3D MAX 3.1 package - Anisotropic Wheel. Here is a screenshot from the final rendering of this scene:
The properties settings during final rendering looked as follows:
And now here come the results shown by these platforms (the rendering time is given in minutes and seconds, so smaller value denotes better result):
The viewports leader, Pentium 4 1.5GHz, is now on the last but one place being slightly ahead of Pentium III 1GHz. All the three Athlon based systems with different memory bus frequencies showed that the memory bus bandwidth doesn't have any serious influence on the performance during final rendering. This is once again proven by the results obtained on a dual-Pentium III 1GHz system, which won the first prize. Also during the scene final rendering the major workload falls on the FPU unlike viewports rendering.
Well, our investigation showed that the best OS for work in 3D MAX appears the good old Windows NT. It especially makes sense for the happy owners of NVIDIA GeForce2 GTS based cards, which we consider the best solution for 3D MAX users. Even though Windows 2000 is "built on NT technology, this OS failed to run as fast as its predecessor. The possibility to work in Direct3D, provided by Windows 2000 and Windows 98 isn't a great advantage at all, but for a couple of small positive things, such as high color depth in viewports. Other than that, Direct3D drivers offer only some grave intolerable errors, which makes efficient work absolutely impossible. Unlike the two other operation systems, Windows 98 suffers from fps rate instability when working via OpenGL drivers. In other words, we don't get 40fps all the way through animation, but the fps rate varied from 5-10fps up to 100fps. This fact as well as the fact that all the graphics cards tested showed the lower results under this OS make it unacceptable for 3D MAX needs.
The best graphics cards for 3D MAX users are undoubtedly those from NVIDIA GeForce2 family. Note that their performance in 3D MAX depends only on the graphics chip frequency and is in no way influenced by the memory speed or number of rendering pipelines. If you are interested in the details on graphics cards testing, please check our article titled "Contemporary Graphics Accelerators Tested in 3D MAX", because this review only touches upon that.
As for choosing the best graphics card and operation system, there is the only correct decision you can make. However, as it comes to selecting the best hardware platform, things turn somewhat more complicated. When working in viewports, i.e. when the designer is actually creating the scene, the determinative is the memory bus bandwidth, that's why Pentium 4 took the lead there. However, since the image is calculated differently during the final rendering, the FPU appears more meaningful and hence the results turn upside down. In the final rendering Pentium 4 1.5GHz performs almost on the same level as Pentium III 1GHz. And on the first position we see a dual-processor system. That is why when deciding on a platform for work in 3D MAX, you should bear in mind how long it will take the designer to create a scene and how much time you will assign to your PC to do the final rendering. Unfortunately, we can't give you any universal advice here. On the other hand, however, you could make a combined platform of two or more computers connected by a local network, some of which are optimized for work in viewports and the other ones - for doing the final rendering.
Well, this is not the last time we return to 3D Studio MAX. In the nearest future we will offer you some more reviews of different hardware tested in 3D Studio MAX.