S3 DeltaChrome S8 Gets Nitro Acceleration: Review of the Revamped S8

S3 Graphics is back and is rolling-out a yet another graphics processor today. The new chip is based on the same DeltaChrome S8 architecture X-bit labs readers are already familiar with. But the newcomer is fueled with some additional clock-speed power. Will the new DeltaChrome S8 Nitro crash competitors and get a throne in the mainstream market? Read our coverage involving 20 different benchmarks and featuring some analysis to know.

by Alexey Stepin , Anton Shilov
03/09/2004 | 02:56 AM

A few months ago in the end of 2003 we posted a detailed review of a new product form S3 marking that the company returns to the graphics processors market (see our article called The Return of S3: DeltaChrome Graphics Card Review for more details). On that day we reviewed the features and tested the performance of a new graphics adapter based on S8 DeltaChrome chip. Although this eight-pipeline newcomer was not free from certain drawbacks and bottlenecks, which we are more inclined to consider “children’s diseases”, it performed pretty well in our tests despite low working frequencies and not finalized drivers.

 

Despite some rawness of the product, DeltaChrome boasts a number of unique technologies, such as Chromotion video engine, which doesn’t have any analogues today, and is responsible for real-time video streams processing. Also the developer claimed that their solution supports complex vertex and pixel shaders going far beyond those specified by DirectX 9 spec. If you are looking for more details on DeltaChrome specifications, please see the corresponding article now.

During the tests of the new S3 features we discovered what we thought was a unique anisotropic filtering algorithm developed by S3 Graphics and implemented in their new solution, which allowed performing this filtering without any performance losses even at the maximum 16x level of anisotropy. This has always been a trump of ATI solutions only, but not any more. As for full-screen anti-aliasing, at that time they implemented only a very resource-hungry 2x super-sampling algorithm, even though the developer claimed that their solution should support multi-sampling. Moreover, this super-sampling didn’t work in resolutions beyond 1024x768 and conflicted severely with some games. Of course, we shouldn’t blame the VPU engineering team for that, it was within the software developers’ responsibility, as they were supposed to ensure the corresponding support in the drivers. However, as you know the drivers have always been a stumbling stone for S3, so this situation is not at all surprising.

Nevertheless, the launch of the new S8 DeltaChrome didn’t turn into a fiasco, like the situation with XGI Volari. Besides the above mentioned drawbacks the new chip also boasted a number of indisputable advantages, which turned it into a very attractive solution. I should admit that it was a very good start for a newcomer in the 3D graphics market, since the long-term absence of S3 Graphics on the 3D graphics stage turned it into a newcomer by now. As the desktop VPU market develops very rapidly, even a one-year absence is drastic for the company, and they will have to start pushing their way through anew.

Despite this pretty successful return of S3 Graphics, the VPU working frequency was still too low, especially against the competitors’ background. So, even despite the 8 pipelines the less efficient architecture of the new chip wouldn’t let it go neck and neck with the mainstream favorites, NVIDIA GeForce FX 5700 Ultra and ATI RADEON 9600 XT. Moreover, the production of the graphics adapters based on the new S8 DeltaChrome F1, which were intended to compete with  RADEON 9800/PRO and GeForce FX 5900 has never started, as the S3 solution with 128bit memory bus couldn’t successfully compete with the 256bit memory access of the rivals.

S3 Graphics seems to have expected a situation like that, so they decided to introduce one more product in their new family, which could compete on equal terms with the mainstream solutions from the industry leaders. In this situation the company decided to go the simplest way: to increase the chip working frequencies. The new chip is also called DeltaChrome S8, but also acquired a word “Nitro”, which stands for higher performance. This is exactly the product we are going to review today. We will try to figure out how greatly the performance of the new solution grew up compared with the predecessors, and will also see if the company’s software developers managed to eliminate the bugs in the drivers discovered earlier.

Closer Look

S3 DeltaChrome S8 Nitro arrived in our lab in a neat black box with S3 Graphics logo on it. Besides the card there was also a CD-disk with the drivers and other info, as well as a converter, which allows connecting HDTV devices to the card, such as plasma screens, projector-TVs, other display devices supporting this format.

I was pretty surprised to find out that S3 DeltaChrome S8 Nitro is an exact copy of its younger brother: S3 DeltaChrome S8 based graphics card.

  

The only remarkable difference appeared the absence of the second cooler power supply connector, a few new stickers on both sides of the PCB and a more compact cooling solution. This time our test sample featured a cooler looking very similar to Thermaltake Blue Orb, though it featured a slightly different ribs configuration and was painted black. I couldn’t remove it without applying some extra efforts, so I decided to leave it alone fearing to damage the VPU. These are the only differences we discovered between the new S3 DeltaChrome S8 Nitro and the other DeltaChrome sample we had at our disposal.

The 8 memory chips installed on the card PCB are manufactured by Samsung and marked as K4D26323E-GC2A, which means that they boast 2.8ns access time and work at 350MHz (700MHz DDR) frequency. Nevertheless, the memory actually works at 325MHz (650MHz DDR). The graphics processor also works at 325MHz frequency. These are far not the highest working speeds for the final S3 DeltaChrome S8 Nitro solutions. S3 representative claims that the card can work at much higher rates, although they decided to lock the clock rate at 325MHz to ensure maximum stability oft heir product.

When we installed the graphics card into our system we discovered that the new S3 solution works equally well with and without the connected power supply cable, just like the pervious S3 product. The new card generates somewhat less noise than the predecessor, at least this is my subjective observation. The memory chips do not require any additional cooling and the graphics core heats up just a little bit, just like in the non-Nitro version of the card.

All in all, the PCB looks pretty finalized, though I would consider it a bit too large for a mainstream product. Anyway, it doesn’t make sense to discuss this issue, as we are testing an engineering sample and not a mass production one. Maybe the final version of this product will be deprived of the useless power supply connector and the PCB will grow smaller, but it doesn’t depend only on S3 Graphics, but also on the graphics card manufacturers, who will be offering S3 DeltaChrome S8 Nitro based products. Some of them may wish to introduce their own PCB design, although this is getting rarer nowadays.

Drivers

During our test session we used the new S3 drivers, which we the company was so kind to provide us with. In terms of the available setting options the new driver version is absolutely identical to the previous one (see our Review for details), although this time we discovered one new settings page devoted to OpenGL:

Unfortunately, the variety of available settings doesn’t strike us as rich: you can only adjust the anisotropy level, enable/disable Vsync and set the level of detail. But this is definitely better than nothing. At least you can now take advantage of the S3’s excellent anisotropic filtering algorithm in OpenGL applications, too. Besides that, there are no new settings or options in the drivers. And the pages devoted to power consumption and working frequencies are still missing completely.

Testbed and Methods

Besides S3 DeltaChrome S8 Nitro we will also have the following products participating in our test session:

Our testbed was configured as follows:

Just like in our previous review we used two versions of Force Ware driver. One of them, version 52.16, was used only for performance tests in FutureMark packages, because it is exactly the version certified by the developer and it provides adequate results in 3DMark03. In all other cases we used version 53.03. The gaming tests were run with the maximum image quality settings for each game.

HDTV, DVD and DivX Video Playback Peculiarities

Since DeltaChrome graphics processor is based on a complex programmable Chromotion engine intended for video streams processing, we though it would be interesting to find out how greatly it can reduce the system CPU workload during video playback in different formats.

Of course, we were mostly interested in checking the situation with HDTV video, so let’s start here. This standard provides the highest quality video of all other standards today, because it supports resolutions up to 1920x1080. For a more illustrative comparison: the widely spread DVD format supports 720x576 in PAL and 720x480 in NTSC modes. Of course, HDTV video is much harder to decode and process than the regular DVD streams. This is exactly where Chromotion may be the right thing to help. However, if you want to take full advantage of the S3’s programmable engine, you should receive fully-fledged support from the developers of video playback software. Today we are going to find out if the advantages of S3 Chromotion engine will provide us with efficient HDTV playback. For our tests we downloaded two videos with 1920x1080 resolution and 8384Kbit/sec bitrate (you can download them from Microsoft web-site).

I should say that it would be incorrect to speak about the horizontal resolution in case of video playback by consumer electronic devices, because this parameter is not critical for TV-sets. The resolution in this case is defined as the number of displayed vertical lines. In our case there were 1080 of them. In case of video playback on a PC the horizontal resolution depends on the video format. If the screen sides ratio equals 4:3, the horizontal resolution equals 1440, and for 16:9 ratio – 1920 horizontal pixels. The first video used in our today’s test session was recorded in 4:3 format, and the second one – in 16:9 format.

  

  

Each of the two videos was about 2 minutes long. To pay both of them we used the regular Windows Media Player 9. When we played the HDTV videos the CPU utilization varied from 47% to 63% and no visual artifacts were revealed. If you took a closer look at the image, you could notice that it was a little grainy, which could be caused by the WMV format. Nevertheless, the image quality was just excellent and the colors bright and rich. When we installed a RADEON 9600 XT and then GeForce FX 5700 Ultra, the situation didn’t change and the CPU utilization remained the same. It means that the unique video processing abilities of S3 DeltaChrome do not work right now. At least it is true for the WMV format and Windows Media Player from Microsoft.

Now let’s see what happens to other formats. S3 Graphics provided us with a specially optimized InterVideo WinDVD 5 version. During the playback of the above mentioned videos, the CPU utilization stayed about 55% and all Chromotion video effects worked just fine, except Neon Forge. When we enabled Neon Forge the CPU utilization immediately grew up to 100% and the video playback started spurting forward. We observed the same effect when we enabled Deblocking Filter, although it doesn’t make much sense to apply this filter to HDTV video, because the quality of this video is high enough and doesn’t need this kind of additional processing.

Now let’s pass over to DVD playback.

During the playback of the licensed Dune movie with the optimized WinDVD 5 player version the CPU utilization never exceeded 7%. All Chromotion video effects also worked fine. When we played the same disc in Windows Media Player the CPU utilization varied from 11% to 12%. So the conclusion here is very simple I suppose: Chromotion engine does ensure some advantages during DVD playback, although it is not that critical for contemporary CPUs.

As for DivX, the CPU utilization lied between 5 and 11% when we played a file with 512x384 resolution in both: BSPlayer and WinDVD 5. All Chromotion effects worked just fine, without any problems.

So, Chromotion technology from S3 Graphics really does work, but it should be supported not only by S3 Graphics Company, but also by the video player developers. So far there is only one video player, which can take advantage of the Chromotion features: InterVideo WinDVD 5 with the special S3 Graphics drivers. Also note that the use of special effects is not absolutely free: you have to sacrifice some of the system CPU resources for that.

DeltaChrome: Beyond DirectX 9

While there are still not a lot of pure DirectX 9 games on the market, companies like NVIDIA and S3 Graphics added a number of additional capabilities exceeding DirectX 9 specifications into their latest generation of graphics processors. This was done not just in order to prolong the lifespan of the GeForce FX or DeltaChrome graphics processors, but to implement a number of specific features not directly related to PC gaming.

For instance, GeForce FX and Quadro FX graphics processors are very good in professional OpenGL applications and allow executing multitude features not found on the RADEON VPUs that are 100% compatible with DX9, not more, not less. This should help NVIDIA to address certain markets where real-time graphics was not historically available or simply make the work of graphics applications and GPUs more efficient by encouraging software makers to take advantage of the GeForce FX innovations.

S3 Graphics had a bit different reasons to add “DirectX 9+” capabilities into its DeltaChrome graphics architecture. Primarily, the list of extra features found in the DeltaChrome includes, but is not limited to: programmable render target blending (we expect to see this one in DX10); proprietary instructions for IDCT, adaptive filtering, deblocking, de-interlacing; programmable video shader; programmable depths shader; programmable table filtering and so on (you may check our initial DeltaChrome S8 Review for more details about feature-set of the chip). Basically speaking, S3 Graphics internally uses some of the additional architectural enhancements for a number of special features like Chromotion.  Moreover, the DirectX 9 capabilities along wit extra features of the DeltaChrome are key components for the recently announced PGC (Personal Gaming Console), which means that game developers may eventually start using them, in case they develop games for the PGC at all.

With rich set of features and low power consumption, the DeltaChrome architecture could be a viable choice laptops, PDAs, mobile phones, STBs and others. Even though S3 believes that the DeltaChrome is great for laptops, at present S3 Graphics is seriously concentrated on promoting the architecture as the base for VIA’s Personal Gaming Console.

Performance in Theoretical Benchmarks

As we have already told you, DeltaChrome S8 Nitro differs from the predecessor only by the clock frequencies. But nevertheless, we decided to carry out a full set of theoretical tests, in order to find out how greatly the performance of the new solution grew up due to the working frequencies increase. As usual we will start our theoretical part with the fillrate investigation.

Synthetic Benchmarks: Fillrate

As always we will start the fillrate test with the enabled color writes and Z writes:

What do we see? And we see the same architectural drawbacks, which we have already revealed in the previous version of DeltaChrome. However, the increase in the working frequencies of the chip would never actually cure them, that’s evident. When working only with the Z buffer, S3 chips perform just perfectly, but they immediately lose this advantage when there is one or two textures to be laid. This definitely indicates that the texture caches are two small. After that the graphs levels out and the performance stops its dramatic reduction. Moreover, the solution turns out even faster that the competitors when we have to lay three or four textures. Certainly, 8 pipelines is an indisputable advantage.

Now let’s disable Z writes:

Graphics processors from ATI and NVIDIA have hardly reacted to disabled Z writes. However, DeltaChrome did: the fillrate with 0 textures dropped down a lot. In fact this is not at all surprising, as it lost one of its major trumps: efficient work with the Z-buffer.

It is interesting but this time the transition from one to two textures appeared much sharper than in our previous investigation of the DeltaChrome performance, where the fillrate decreased smoothly with the growth of the number of textures. There is only one reasonable explanation to that: new drivers.

Now we are going to change the testing conditions again: let’s enable Z-writes and disable the color writes.

Usually, when we disable color writes all graphics adapters demonstrate the maximum possible performance, because they do not have to work on textures any more. However, this is not the case for DeltaChrome, which acts like that only if there is not more than one texture involved. As soon as there appears the second texture, the S3 processor slows down a lot. Moreover, the graph in this case looks completely different from what we saw with enabled color writes. Besides, this graphs also has nothing in common with the graph you can find in our previous review of S3 DeltaChrome S8: if you remember, the performance dropped even more when we had two textures to be processed. If you remember, at that time we supposed that this performance drop was caused by the drivers and could be eliminated in the new driver version. Time showed that this was a correct supposition: the situation has been significantly improved with the release of the new driver version, although it is still far from ideal. We assume that the bug in the drivers and/or the architectural peculiarities of the DeltaChrome architecture make it continues attempting to perform the color writes, even if this feature is formally disabled. That is why the performance results turn so poor. Since color writes need to be disabled pretty rarely, S3 software guys could have simply omitted it. The situation with enabled Z-writes and disabled color writes is very unlikely to occur in any game or application, that is why the results obtained in this mode are interesting mostly from the theoretical point of view.

As for the performance gain ensured by the higher working frequencies of the chip, it is most evident in case of 0 or 1 texture. As the number of textures grows, the chip gets more and more loaded with work and the architectural flaws start mattering more than higher working frequencies.

Synthetic Benchmarks: Pixel Shaders

The Fillrate Tester program we used to test the fillrate can also work with Pixel Shaders:

The performance of DeltaChrome is pretty ambiguous: it falls behind and then outpaces RADEON 9600 XT every now and then. The best result is still demonstrated with simple Pixel Shaders version 1.1. Here you can also see the performance difference between DeltaChrome S8 and DeltaChrome S8 Nitro most clearly.

When we disable Z writes, the situation remains the same.

When we disable color writes, DeltaChrome immediately loses to everyone. Unfortunately, I cannot find a suitable explanation to this phenomenon. There is absolutely no progress since the last review of the DeltaChrome solution. Either the software developers haven’t yet tackled pixel shaders optimization, or it is the architectural peculiarities of the DeltaChrome chip that cannot be improved on the software level.

We didn’t test the pixel shader performance with only one single program, of course. We have also tested the new S3 product in 3DMark 2001 SE and 3DMark03 test packages.

The new DeltaChrome won’t set the world on fire and demonstrates pretty average results: it processes pixel shaders version 1.1 as fast as GeForce FX 5700 Ultra in two resolutions out of three. I have to admit that S3 has improved the drivers since the times of the first DeltaChrome, because at that time the newcomer could only outperform GeForce FX 5600 Ultra, and now it can compete with a faster GeForce FX 5700 Ultra solution. However, DeltaChrome, as well as GeForce FX, is completely ruined by the RV360 architecture from ATI, which proves much more efficient when processing pixel shaders.

When we shift to pixel shaders version 1.4, RADEON gets so far ahead that the competitors cannot even see it any longer. Nevertheless, DeltaChrome S8 Nitro manages to stay at the level with GeForce FX 5700 Ultra and even to outpace it a little bit.

Pixel shaders 2.0 demand much more from the graphics processor architecture than the version 1.1 and 1.4. No wonder that both DeltaChrome modifications appear pretty slow here. GeForce FX benefits from high working frequencies, fast memory and shader code compiler integrated into the Force Ware driver. Nevertheless, DeltaChrome S8 Nitro performs quite fine in 1024x768, but then the gap between the newcomer and GeForce FX 5700 Ultra grows bigger due to super-fast memory of the latter. As for the advantage of ATI RADEON, no comments are necessary I suppose: the diagram is more than illustrative.

Now let’s check what ShaderMark 2.0 will reveal. As usual, this test package didn’t recognize DeltaChrome S8 Nitro as fully compatible with DierctX 9 specification and provided the following results:

The driver improvements are more than evident: with the new drivers the performance of DeltaChrome S8 Nitro is not as low as that of the predecessor. Of course, S3 graphics processors cannot catch up with RADEON 9600 XT, but they can easily compete with the praised shader compiler from NVIDIA. The situation is quite ambiguous: DeltaChrome S8 Nitro can process some shaders faster than NVIDIA GeForce 5700 Ultra, though in all other cases this VPU yields to the rival. So, S3 managed to raise the shader performance of its DeltaChrome to the level of GeForce FX 5700 Ultra, which will definitely have a positive effect on the gaming performance of this solution. We are going to dwell on this supposition and the actual games in a section devoted to gaming performance, so read on!

I don’t think DeltaChrome S8 Nitro will be able to catch up with RADEON 9600 XT: this is the price you have to pay for the flexibility of DeltaChrome architecture and higher functionality than the minimal DirectX 9 requirements. In other words, S3 DeltaChrome looks somewhat similar to NVIDIA NV3x family, although S3 stakes the 8-pipeline architecture rather than higher memory and chip frequencies.

Synthetic Benchmarks: Vertex Shaders

We also tested the performance of our testing participants when working with vertex shaders in FutureMark packages. Here are the results obtained in 3DMark2001 SE:

The 4 vertex pipelines allow DeltaChrome S8 Nitro to perform quite well in this test. Of course, it is rather far behind GeForce FX5700 Ultra, though it looks pretty OK against the background of other testing participants.

Well, DeltaChrome vertex shader processors are still not efficient enough. As the number of light sources in the test scene increases, the performance of DeltaChrome vertex shader processors drops very much below that of the RADEON 9600 XT vertex shader processors. The latter also benefits from the VPU working frequency equal to 500MHz, which makes up for few vertex processors.

DeltaChrome S8 Nitro doesn’t look that nice in the vertex shader processing test: it yields to all the rivals except GeForce FX 5600 Ultra. This is another proof to our supposition about the low efficiency of DeltaChrome vertex processors. Moreover, the memory of S3 solution is too slow and it limits the performance as well.

In the similar test from 3DMark03 test package the situation turns a little different. The test scene is quite complex here. ATI cards start suffering from their slower memory and they retain their leadership only due to higher VPU frequencies.

DeltaChrome S8 Nitro runs neck and neck with RADEON 9600 PRO and the regular DeltaChrome falls a little bit behind. Maybe the number of S3 shader processors makes up for their relatively low efficiency in this case.

Synthetic Benchmarks: Bump Mapping, Sprites, Etc.

In conclusion we would like to run different bump mapping tests and a few tests checking the sprites performance.

EMBM bump mapping is still a problem for DeltaChrome, even the S8 Nitro model falls behind every other testing participant except GeForce FX 5600 Ultra.

When we use Dot3 method the situation turns completely hopeless for the newcomer: it is the slowest, even slower than GeForce FX 5600 Ultra. If we keep in mind that Dot3 method actually uses two textures (one of them is the normal map), we can conclude that DeltaChrome owes its disaster to the low fillrate during 2+ textures processing.

The results of the Ragtroll test were actually quite predictable: both DeltaChrome solutions work at pretty low core frequencies, and the vertex pipelines are not efficient enough to make up for low clock rates. As a result the solution loses to every single competitor except GeForce FX 5600 Ultra.

The sprites test comes next. It actually also tests the vertex processors performance, though the situation here is more interesting than in all other benchmarks, I should say: in low resolutions the performance improved notably due to higher working frequency of DeltaChrome S8 Nitro. As the resolution grows this advantage vanishes because of the low efficiency of the memory subsystem. RADEON 9600 XT also suffers from its slow memory. All in all, DeltaChrome cannot boast much: it was completely defeated by all the rivals.

Summing up the results obtained in the synthetic benchmarks, we can point out only one positive improvement: DeltaChrome can now process pixel shaders somewhat faster. And that’s it. In general the situation is pretty dramatic, especially in geometric benchmarks. DeltaChrome is very unlikely to ensure high performance in geometrically complex games with the vertex processors being that slow. As for the “Nitro” word in the name of the product, we have to admit that the higher working frequency does have its positive influence on the performance of the solution, though this influence is not that great at all. 325MHz is not that fast for a video processing unit manufactured with 0.13micron technology. Although the current DeltaChrome processor still uses the old packaging. The final DeltaChrome S8 Nitro modification can be designed in an open-die package and also work at higher frequency, which has actually happened to GeForce FX 5600 Ultra, as you should remember.

Anisotropic Filtering, FSAA Performance Impact

We were extremely excited about anisotropic filtering for “free” with the DeltaChrome S8 and could not miss the opportunity to experience that thing once again. Well, we were a bit disappointed, but our expectation about anisotropic filtering without performance impact did not materialize.

If there is a technology to allow “free” anisotropic filtering on the DeltaChrome, then it does not work at this time with certain games and/or is driver-dependant. At least, UT2003 shows pretty different performance with anisotropic filtering turned on compared what we had three months ago. Another explanation to the phenomenon is that there probably was something with S3 Graphics’ drivers, which degraded performance in normal mode to the point of speed with a certain level of anisotropic filtering enabled. For instance, DeltaChrome could use 2 texture samples to emulate 8x, 16x anisotropic filtering. This resulted in pretty sharp, but a bit noisy textures. Provided that the issues are corrected now, performance should degrade according to the level of anisotropic filtering.

Please have a look on diagrams that represent dependence of UT2003 performance on the level of anisotropic filtering:

Speed of S3 DeltaChrome S8 Nitro waterfalls once anisotropic filtering is enable just like that of other graphics chips.

The situation with Inferno demo is practically the same as with Antalus level. S3 DeltaChrome S8 Nitro does not provide “free” anisotropic filtering, unfortunately.

The benchmark results from the Unreal Tournament 2004 brought us nothing new, just like the results of Max Payne 2:

In addition, we decided to try an OpenGL game with S3’s mysterious anisotropic filtering and took Call of Duty.

But no wonder happened. Speed degrades of S3 DeltaChrome S8 Nitro are in-line with other graphics processors of today.

Currently S3 DeltaChrome S8 Nitro supports only a single method of full-scene antialiasing: 2x2 supersampling. This one hits performance heavily and is not something that deserves using. S3 promises to improve FSAA performance in future, but it remains to be seen whether the company succeeds or not.

Once FSAA is enabled, speed is below any critics.

Let us try to sum up what DeltaChrome S8 Nitro has to offer us in order to improve the quality of 3D graphics. While anisotropic filtering approach seems to resemble that of NVIDIA and boasts pretty high quality amid pretty reasonable speed sometimes, FSAA is a feature we would not recommend to use on DeltaChrome because of too low performance.

Gaming Performance

We have again updated the list of games to be used for graphics processors and VGA cards tests. Here it is for your reference:

First person 3D shooters:

Third person 3D shooters:

Simulators:

Real-time strategies:

Semi-synthetic benchmarks:

Synthetic benchmarks:

We removed Star Trek 2: Elite Force game from the list and added a very popular first-person 3D shooter instead. This game is called Call of Duty and the action takes place during the WWII, just like in RTCW. Moreover, we also added the demo versions of Unreal Tournament 2004 and FarCry. You can read more about the first game in our article called Unreal Tournament 2004: The Kingdom of Massacre. As for FarCry, it is worth dwelling on now: even the demo version of this 3D shooter with the limited number of textures is very beautiful and rich in modern special effects.

Pixel shaders are everywhere in FarCry. In the very beginning of the demo level the player is surrounded by a very beautiful water surface with the highly realistic reflection of the surrounding environment. The game is very resource hungry and can use up the entire system potential even of a high-performance system like ours. It suits perfectly well for the evaluation of the contemporary graphics accelerators performance, especially since the games of the kind will become more and more numerous with the time.

Unfortunately, FarCry once again demonstrated how important it is to have hardware and game developers cooperating on a constant basis. Just take a look at the screenshots below:

  

As you see, by DeltaChrome all objects within sight are colored light-blue except those located at a distance from the viewer. We haven’t seen anything like that by RADEON or GeForce FX. The problem is evidently in the S3 Graphics’ drivers, and we hope they will eliminate it in the upcoming driver versions. As for all other games, we didn’t notice any visual problems there. The issues described in the previous S3 DeltaChrome S8 Review have been eliminated: now Unreal Tournament 2003, Tron 2.0 and F1 Challenge provide a crystal clear image without any artifacts.

Unfortunately, the new drivers again support only a very resource-hungry 2x super-sampling mode, which drops the performance down to an unacceptable level. Moreover, this anti-aliasing works only in resolutions below 1024x768 that is why there is only one single diagram for Eye Candy mode in each game, where you can see the results of 2x super-sampling and 8x anisotropic filtering. I don’t think I need to provide detailed comments to these diagrams, because the performance in every case was 2-2.5 times lower than in Pure Speed mode. In other words, it was so low that I don’t think it is of any practical value to you.

Performance in First Person 3D Shooters

Return To Castle Wolfenstein: Enemy Territory

S3 DeltaChrome S8 Nitro falls behind all the rivals despite its 8-pipeline architecture. It is the raw OpenGL drivers that cause this failure. NVIDIA GeForce FX is ahead of all, as usual, as its OpenGL driver is the best today. The results of Eye candy testing mode are not provided as S3 OpenGL driver doesn’t support FSAA.

Call of Duty

Call of Duty looks very similar to RTCW, however, in fact the differences are tremendous. In particular, the game has pixel shaders. The level with a Nazi military ship suits very well for testing purposes, because there you can see not only nice geometry but also water surface created with pixel shaders. It is exactly these pixel shaders that determine RADEON’s victory over NVIDIA GeForce FX, while the overall complexity of the environment caused a dramatic failure of S3 DeltaChrome S8 Nitro. Higher working frequencies allow the new S3 solution to outperform its predecessor, but you can actually see it only in high resolutions. The results obtained in FSAA+AF mode are not provided for the same reason as in case of RTCW: Enemy territory.

Unreal Tournament 2003

In Inferno test the newcomer suddenly performed very well and even appeared among the three fastest testing participants.

S3 software developers did a great job on the drivers: DeltaChrome has now turned into a worthy competitor to the other tested solutions, while at first it was behind everybody, as you remember.

The advantage of S3 DeltaChrome S8 Nitro over the predecessor doesn’t require any additional comments of ours.

Antalus, is another demo where S3 DeltaChrome S8 Nitro feels at home. 8 pipelines and new drivers let it run almost as fast as the leaders do, falling just a little bit behind GeForce FX 5700 Ultra with its high-speed memory.

When we enable anti-aliasing and anisotropic filtering the performance drops down dramatically. Extremely resource consuming FSAA implementation, not very efficient graphics processor performance and low memory frequency do their black deed. The performance of our hero is so low, that it doesn’t make sense to comment on each of the demos separately.

Unreal Tournament 2004 Demo

In the new version of the popular on-line shooter from Epic, S3 DeltaChrome S8 Nitro also performs quite fine and can successfully compete with ATI RADEON 9600 PRO.

We see the same situation in the second test, although the results here are overall higher because the story takes place indoors.

In “heavy” modes the situation we have just discussed in UT2003 repeats: the performance is unacceptably low, though only 2x anti-aliasing is enabled.

Halo: Combat Evolved

Halo is a very demanding game that is why here S3 DeltaChrome S8 Nitro yields a little to its rivals. Nevertheless, its performance is close to that of RADEON 9600 PRO, which is not bad at all, keeping in mind that we are testing only an engineering sample of this card now.

Tron 2.0

Geometry processing speed is not that high by S3 DeltaChrome, but it is still higher than that of GeForce FX 5600 Ultra. So, this explains the results you see on the diagram.

The newcomer manages to outpace GeForce FX 5600 Ultra, but can it be called a victory at all? I don’t think so, because the NVIDIA solution provides better image quality.

Highly Anticipated DX9 Game 1

In the next generation games we will not run the tests in only two records, because the third record is a bit excessive, I suppose. Having run the tests in the third record, I saw that it doesn’t reveal anything new in the performance picture.

Next generation games will be very resource hungry. Most users know it already. NV3x architecture is not very well suited for Next Generation DirectX 9.0 Game, which we have already discussed in detail in our article. And the drivers, which were promised to improve the DirectX 9.0 performance in this and a few other applications are not available yet. S3 DeltaChrome S8 Nitro does its best to work in DerctX 9 mode and… it succeeds! It is not as successful as ATI based graphics cards, but it easily outperforms GeForce FX 5700 Ultra, which can hardly reach 30fps in the lowest screen resolution. GeForce FX 5600 Ultra, which could work in this game only with the DirectX 8 rendering mode enabled shows better numeric results but sacrifices a lot of its image quality for that.

In Under Two test the situation is just the same: S3 DeltaChrome S8 Nitro outperforms GeForce FX 5700 Ultra but falls behind all other testing participants.

Highly Anticipated DX9 Game 2

This game focuses primarily on the geometry and texturing. You can read more about it in our article called Yet Another DirectX 9 Game: Lost Oblivion in Chernobyl.

S3 DeltaChrome S8 Nitro performed very slowly here, but it is too early to make the final verdict yet. The game as well as S3 drivers are not finalized yet, so the situation can change a lot in the future. Unfortunately, the game refused to run on DeltaChrome in 1600x1200 resolution.

A relatively slight performance drop compared with the Pure Speed mode is no consolation. On the contrary, I was pretty puzzled with it.

Everything repeats in Escape scene. No comments are necessary.

The same is true for Eye Candy mode as well.

FarCry Demo

Unfortunately, this game is only available as a demo version. However, despite its far not the best quality textures, it can already be used as a reference for 3D image quality evaluation. Like most contemporary games, which use new technologies very actively, FarCry has some problems with proper implementation of anisotropic filtering and full-screen anti-aliasing. Neither FSAA, nor AF work correctly in this first-person 3D shooter game. So, we are offering you only the results obtained in the Pure Speed mode:

Oh, this is a total disaster, no doubt. In fact it is not at all surprising, as the game is simply stuffed with shader effects and bump mapping. Of course, ATI R3xx/RV3xx architectures feel best of all here.

Performance in Third Person 3D Shooters

Tom Clancy’s Splinter Cell

The results demonstrated by S3 DeltaChrome S8 Nitro are pretty low and stay at the level of GeForce FX 5600 Ultra. The situation may improve in the future, but so far the performance of DeltaChrome in this game is beyond all criticism.

Tomb Raider: Angel of Darkness

For some unknown reason the game refused to run on DeltaChrome in any resolutions above 1024x768. I believe it is the drives that we should blame for this. The results obtained in the only operational resolution were a pleasing sight to our eye: S3 DeltaChrome S8 Nitro fell just a tiny bit behind RADEON 9600 PRO. As for the enabled FSAA and anisotropic filtering, the game refused to enable these technologies on S3 DeltaChrome.

Prince of Persia: Sands of Time

The newcomer runs almost as fast as RADEON 9600 PRO in this game. Moreover, in 1600x1200 S3 DeltaChrome S8 Nitro even catches up with RADEON 9600 XT.

Max Payne 2: The Fall of Max Payne

Unfortunately, in Max Payne 2 S3 DeltaChrome S8 Nitro gave up its positions and appeared lagging behind all the rivals except the morally outdated GeForce FX 5600 Ultra. To our great disappointment, we have to admit that the cards on S3 DeltaChrome S8 Nitro will have to compete with the whole bunch of other products, but definitely not with GeForce FX 5600 Ultra, when they appear in the market. In this respect, I don’t want even to think about the possible performance of DeltaChrome S4 here.

Here the performance drops down dramatically again. It is impossible to play like that, no doubt.

Performance in Simulators

IL-2 Sturmovik: Forgotten Battles

In Il-2 Sturmovik S3 DeltaChrome S8 Nitro rolled back to the very last spot having yielded even to GeForce FX 5600 Ultra. Well, it looks as if S3 Graphics will have to work really hard on its newborn before it becomes truly competitive.

The results need no comments of ours, I suppose.

X2: The Treat

Everything I have just said about Il2 is also true for the X2: The Threat space simulator. Although S3 DeltaChrome S8 Nitro is faster than the regular DeltaChrome S8, both these graphics solutions are pretty slow. DeltaChrome is not suitable for playing this game at all.

The results here are not so tragic, as in Il-2 Sturmovik, though it is still impossible to play at such a low fps rate.

F1 Challenge 99-2002

The situation in F1 Challenge looks somewhat different. S3 DeltaChrome S8 Nitro runs quite acceptably although it fails to catch up with GeForce FX. However, as the resolution grows, the newcomer slows down yielding to the rivals because of its not very efficient memory subsystem.

The gaming performance is below the acceptable level: no comments are necessary.

Performance in Real-Time Strategies

C&C Generals: Zero Hour

In this game our hero lacks shader power: it is defeated by all its competitors. The situation may change for the better in the future, but so far S3 DeltaChrome S8 Nitro disappointed us with its level of performance.

Everything we have just said is also true for the Eye Candy mode. The performance is twice as low, so forget about comfortable gaming. S3 DeltaChrome S8 Nitro cannot offer anything other than that here.

Performance in Semi-Synthetic Benchmarks

Final Fantasy XI Official Benchmark 2

S3 DeltaChrome S8 Nitro has nothing to boast here: the performance of the new solution can only be compared with that of GeForce FX 5600 Ultra.

Another complete fiasco. And what else did you expect from super-sampling? We can only wait for S3 Graphics software people to implement more optimal anti-aliasing techniques in their drivers, as they have promised.

Aquamark3

Hm, this is also far from positive: S3 DeltaChrome S8 Nitro can be compared with only that of GeForce FX 5600 Ultra, even though S3 DeltaChrome S8 Nitro is a flagman of the DeltaChrome graphics processor family. As for Eye Candy mode, Aquamark simply reported that DeltaChrome doesn’t support FSAA at all.

Performance in Synthetic Benchmarks

Futuremark 3DMark03 build 340

We installed Force Ware 52.16 for GeForce FX solutions to run the 3DMark03 tests, because FutureMark approved of this particular version for the tests in 3DMark03 package. For the other graphics cards we also used the drivers approved by the benchmark developer.

S3 failed to achieve any encouraging results here. S3 DeltaChrome S8 Nitro competes only with GeForce FX 5600 Ultra yielding hopelessly to all the other testing participants.

In the first relatively simple test the newcomer managed to perform quite OK, and yielded just a little bit to the leaders. We failed to obtain any results in Eye candy mode, because 3DMark03 reported about the absence of FSAA support.

The second benchmark is more complex, as it uses shaders and performs the rendering in a few passes. Since S3 DeltaChrome S8 Nitro architecture is not very efficient in such working mode, we shouldn’t be surprised to see its failure. However, higher working frequencies and new drivers make it not as hopeless as the regular DeltaChrome S8.

In the third test, which is not very much different from the second one, actually, the situation repeats.

The fourth gaming test reveals the architectural drawbacks of all graphics processors in the most efficient way. Only the architectures that boast optimal pixel shader performance can win here. In our case we can see that S3 DeltaChrome S8 Nitro is not any worse than GeForce FX 5700 Ultra despite its higher working frequencies and monstrous DDR-II memory.

Conclusion

It is really hard to draw a definite conclusion about S3 DeltaChrome S8 Nitro performance according to the results we obtained in our today’s test session. The product performed too ambiguously as you may have noticed.

In some games, such as Unreal Tournament, ?ALO, Tron 2.0, Max Payne 2, F1 Challenge and Prince of Persia: Sands of Time, the new S3 product performed pretty well. On the other hand, in many other games the newcomer suffered a total fiasco despite its higher working frequency and new driver.

OpenGL driver is still very raw, so the performance in games using this API is extremely low. As for full-screen anti-aliasing, we can say that S3 doesn’t support FSAA at all now. The only available 2x super-sampling mode failed to provide an acceptable performance level in any of the contemporary games supporting FSAA. So, it doesn’t make sense to enable FSAA on about S3 DeltaChrome S8 Nitro in most cases.

Yes, DeltaChrome features go beyond DirectX 9 specification, but will these features be really demanded even when the game developers start supporting them? Of course, not, if the performance level of this solution remains as it is now.

If we take a more abstract look at the whole situation, we will see that the most powerful S3 Graphics product, which is intended to compete with RADEON 9600 XT and GeForce FX 5700 Ultra, doesn’t suit well this goal. It can compete with the morally and technically outdated GeForce FX 5600 Ultra and RADEON 9600 PRO, but the updated mainstream solutions from ATI Technologies and NVIDIA Corporation do not leave S3 DeltaChrome S8 Nitro a single chance. There is even not a single piece of evidence that the situation may change for the better in the future. Of course, we are a little bit optimistic about S3 claiming that its S3 DeltaChrome S8 Nitro can support working frequencies higher than 325MHz, but the question about its ability to eliminate the architectural issues remains open. What we can say about the future of S3 products is that the company is looking forward to deliver faster chips for PCI Express x16 bus later this year. Moreover, the firm is already working on the next-generation architecture beyond DeltaChrome and DirectX 9. In addition to hardware S3 Graphics needs to polish the drivers really well, and luckily we do see some progress in this direction already. But a lot still needs to be done before they come to the final driver version.

The only remarkable thing about the video processing unit from S3 Graphics so far is its unique opportunities to play and process video data, as well as hardware image rotation, which can be quite useful for the owners of LCD panels supporting portrait mode. The mobile device builders will also like low power consumption of the new S3 solution. All other feature of the new DeltaChrome are not exclusive and are also available in competitors’ products. The S3 DeltaChrome S8 Nitro product is pretty good, but it is evidently being late for the mainstream market, which may turn fatal for it, as the market develops very rapidly nowadays.

Initially S3 DeltaChrome S8 graphics cards will be available in Europe, Asia and North America in 2004 in that order respectively (with S8 Nitro to be available later). Pricing will be set to DeltaChrome S8 - $139~$149 and S8 Nitro at $169. Besides Club 3D S3 Graphics is currently working with a major AIB in Taiwan. This and other AIB partners are expected to be disclosed in near future.

Will DeltaChrome remain up-to-date by the time mass products hit the streets? Everything depends on S3 Graphics now and its ability to act real fast to win its niche in the desktop 3D graphics market.