The Return of S3: DeltaChrome Graphics Card Review

We have got a very special guest in our graphics laboratory today - S3 Graphics with its DeltaChrome graphics processor. The long awaited VPU the world has been waiting for about 3 years is now here and so is the legendary developer of VGA solutions. But is this comeback a worthy one? Find it out with X-bit labs!

by Alexey Stepin , Anton Shilov , Tim Tscheblockov
12/30/2003 | 11:55 PM

S3 Graphics. This means a lot to the elder generation of the PC enthusiasts, because in those good old times most personal computers were equipped with the graphics accelerators from this company. I believe that each of you who used to be interested in hardware in those times can easily recall such names as S3 Trio and S3 Virge.

 

As soon as 3D era arrived, the company didn’t want to stay behind such well-known developers as 3dfx, NVIDIA and ATI and launched a pretty successful solution from the technical point of view aka S3 Savage3D. I have to stress here that S3 was a pioneer in the texture compression techniques, so that its new 3D solution could process high-resolution textures at the expense of very small performance reduction. One of the texture compression methods is still called S3TC, which indicates its origin very clearly. The first game to support S3TC appeared Unreal, and the bundle of another Epic game, Unreal Tournament, even included a CD-disc with high-resolution textures, which allowed the happy owners of S3 Savage3D and Savage4 graphics adapter enjoy high-quality clear-cut textures in full. However, Savage3D has never become as popular as it actually deserved. At that time the game developers didn’t pay due attention to the features offered by S3TC, and raw drivers turned into a true disaster for the S3 Savage3D graphics card owners. Unfortunately, the next baby of S3 Graphics, Savage4, followed in the footsteps of the predecessor. Even though it demonstrated pretty high performance and boasted one of the best levels of image quality in the market, constant problems with the drivers did their dirty deed: the newcomer from S3 failed to win the users’ hearts and become truly popular.

The end of 1999. NVIDIA is rushing to finish the new GeForce256 chip, which was claimed to be a revolution in the 3D graphics field. The official product announcement is scheduled for September 1, 1999. Suddenly on August 30 S3 Graphics launches Savage2000, the world’s first graphics processor featuring hardware means for geometry processing. It was just one single day, but it completely ruined all NVIDIA’s attempts to make a loud announcement. It looked like S3’s triumph, but… Regrettably, the situation was not that rosy at all. The thing is that the first graphics cards on S3 Savage2000 were shipped with the disabled T&L engine, which completely eliminated all the advantages of the new solution.

S3 was little by little losing its ground and finally, on April 11, 2000 they announced their desire to sell the Company’s graphics division to VIA Technologies for about $323 million. As a result of the internal S3 restructuring the company was transformed into SONICblue focusing on consumer electronics and at the same time being one of the S3 Graphics shareholders alongside with VIA Technologies. The new S3 Graphics Company didn’t manage to stand out in any way in the desktop graphics market, although it proved pretty active in the integrated chipset field and mobile graphics market. As you know, ProSavage and Twister product lines were built with the graphics solutions developed in the times of savage4 and Savage2000.

The situation remained almost unchanged up to January 7, 2003: on that remarkable day S3 Graphics announced a new DeltaChrome graphics core supporting Microsoft DirectX 9 API. Then on March 13, 2003 the company announced DeltaChrome one more time. Now they were talking about the entire chip family targeted for the desktop 3D graphics market. The top solution in this family was DeltaChrome F1 with the following specifications:

As you see, the chip looks like a pretty serious solution, at least on the paper.

The next model, DeltaChrome S8, differs from F1 only by the size of the supported graphics memory (up to 128MB). DeltaChrome S4 features only 4 rendering pipelines and is the last model in the new product family.

This serious announcement gave everybody to understand that the company was planning to return to the positions they almost lost a year ago. Unfortunately, it turned out a paper announcement and the cards based on DeltaChrome chip didn’t hurry to turn up on the store shelves. The first DeltaChrome previews started appearing on the web in September, claiming that the new graphics cards on S3 chips will start selling in commercial quantities in October 2003. However, the mass shipments of DeltaChrome based solutions began neither in October, nor in November. Moreover, the first contract for DeltaChrome chips supplies has been made public only on December 17, 2003. The first partner of S3 Graphics in Europe appeared the Club 3D Company, which you should have already read about on our site (see our article called Club3D RADEON 9800 PRO: European Finest Graphics Card?). The first graphics card from this manufacturer using S3 Graphics technologies will be based on the top DeltaChrome chip – F1. It is quite possible that they will later introduce a few more solutions based on other S3 chips: DeltaChrome S8 and S4.

A short comment: we have just found out that some sources claim DeltaChrome F1 based graphics cards may never ever see the world. The thing is that this graphics solution family was initially positioned for the lower High-End market. Unfortunately, 128bit memory bus will hardly be able to help the solution conquer this market segment occupied by such successful products as ATI RADEON 9800 and NVIDIA GeForce FX 5900.

S3 graphics chip family now has one more member aka S8 Nitro. It will differ from its younger brother by higher working frequencies of the core and memory.

Will S3 be able to return to the desktop graphics market successfully? In fact, this is an extremely hard task nowadays, because the discrete graphics solutions market has been fully dominated by ATI Technologies and NVIDIA Corporation, which are very unlikely to give way to a new player here. Everything actually depends on the ability of S3 Graphics to provide its new processors with quality support and on the ability of its start-up partners to provide the graphics card makers with sufficient amounts of chips. Anyway, S3 Graphics will have to work really hard to win back the user’s hearts.

Our today’s review is devoted to S3 DeltaChrome S8 graphics solution. We hope that it will help you to find answers to most questions that trouble you today.

DeltaChrome As Is

We managed to get our hands on the graphics adapter based on DeltaChrome S8 chip, which is the second fastest VPU from S3G positioned as a direct competitor to mass solutions of the 3D market leaders, namely ATI Technologies and NVIDIA Corporation. As a result we decided to compare it with graphics solutions based on RV350/360 and NV31/36 respectively. If we approach this comparison from a theoretical point of view, DeltaChrome will look not any worse than the solutions from the two largest graphics processor developers.

Moreover, S3G newcomer boasts a number of interesting technologies. The most remarkable one, which we should mention separately is the Chromotion engine. Starting with R300, ATI’s VPUs are known to support real time video streams processing with the help of pixel shaders unit. S3 Graphics engineers have significantly enhanced the idea about involving pixel shaders into video processing. Having implemented this idea in real silicon they called it Chromotion. What can this wonder-technology actually do? In fact, it can do quite a lot of things: the use of pixel shaders allows performing the following with the video streams:

All in all, DeltaChrome is very rich in video processing features and boasts the following additional things:

S3 has definitely outperformed all its competitors here, although it still failed to become the first in the field of pixel shaders for video streams processing. The leadership here surely belongs to ATI Technologies with its VideoShader, which has hardly been used by the software developers at all (except DiVX Player). We hope that the arrival of the new VPU featuring such wide range of options for video processing will manage to stimulate the developers’ interest and we will finally see the whole bunch of software products using brilliant opportunities offered by VideoShader and Chromotion technologies.

We can’t help mentioning the support of vertex and pixel shaders with the parameters beyond DirectX 9.0 specifications: this is where S3 developers did a really great job. Just take a look at the comparative spec table for NVIDIA, ATI and S3 graphics processors:

 

ATI RADEON 9800

S3 DeltaChrome S8

NVIDIA GeForce FX 5950

Manufacturing technology

0.15micron

0.13 micron

0.13micron

Number of transistors

110-115mln

80mln

135mln

Chip frequency

412MHz

300MHz

475MHz

Graphics memory controller

256bit
DDR SDRAM

128bit DDR SDRAM

256bit
DDR SDRAM

Graphics memory frequency

730MHz
(365MHz DDR)

650MHz
(325MHz DDR)

950MHz
(475MHz DDR)

Peak memory bus bandwidth

23.4GB/s

10.4GB/s

30.4GB/s

Max graphics memory size

256MB

256MB

256MB

AGP interface

AGP 3.0 4x/8x

AGP 3.0 4x/8x

AGP 3.0 4x/8x

Pixel Pipelines, Pixel Shaders

Pixel pipelines

8

8

4, 8

Texturing unites per pipeline

1

1

2, 0

Max number of textures during multi-texturing

8

8

8

Texture filtering types

bi-linear
anisotropic
tri-linear
tri-linear + anisotropic

bi-linear
anisotropic
tri-linear
tri-linear + anisotropic

bi-linear
anisotropic
tri-linear
tri-linear + anisotropic

Max anisotropy level

16

16

8

Pixel shaders version

v.2.0 (f-buffer, v2.0+)

v2.0+

v.2.0+

Branching, subroutines and loops

none

n/a

none

Max number of textures per shader

16

16

16

Max number of texture instructions

32

n/a

1024

Max number of arithmetic instructions

64 (+64)

n/a

1024

Max number of instructions per shader

96 (+64)

n/a

1024

Registers

2 color registers,
32 constant registers,
8 texture coordinates registers,
16 TMU identification registers,
12 temporary registers,

4 resulting color registers,
1 resulting Z register

n/a

2 color registers,
512 (1024) constant registers,
8 texture coordinates registers,
16 TMU identification registers,
16 (32) temporary registers,

4 resulting color registers,
1 resulting Z register

Data representation formats

Fixed point,
16bit floating point,
32bit floating point

Fixed point,
16bit floating point,
32bit floating point

Fixed point,
16bit floating point,
32bit floating point

Internal Pixel Shader Pipeline Precision

96bit pixel precision/24bit floating point precision

96bit pixel precision/24bit floating point precision

128bit pixel precision/32bit floating point precision, 16bit floating point precision

Vertex pipelines, vertex shaders

Vertex pipelines

4

4

3

Vertex shaders version

v.2.0

v.2.0+

v.2.0+

Branching, subroutines and loops

Static

Dynamic, 16x16 maximum loops

Dynamic

Max number of instructions per shader

256

n/a

256

Max number of instructions with loops extension

65536

65536

65536

Registers

16 input registers,
12 temporary registers,
256 constant floating-point registers,
16 constant integer registers,
16 Boolean registers,
1 address register,
1 loops counter register,

8 output registers for texture coordinates,
1 fog color output register,
1 vertex position output register,
1 pixel size output register,
2 output registers for diffuse/mirror color component

16 input registers,
16 temporary registers, etc…

16 input registers,
16 temporary registers,
256 constant floating-point registers,
256 constant integer registers,
256 Boolean registers,
1 address register,
1 loops counter register,

8 output registers for texture coordinates,
1 fog color output register,
1 vertex position output register,
1 pixel size output register,
2 output registers for diffuse/mirror color component

Data representation formats

32bit floating-point

 

32bit floating-point

Full-Scene Anti-Aliasing

FSAA methods

Rotated grid multi-sampling
(RGMS)

Supersampling, Multisampling

Supersampling,
Ordered grid multi-sampling
(OGSS, OGMS)

Number of samples

2, 4, 6

2

2 (OGSS, OGMS),
Quincunx,
4 (OGSS, OGMS, OGSS+OGMS),
6 (OGSS+OGMS),
8 (OGSS+OGMS)

Technologies aimed at higher memory bandwidth efficiency

Hidden Surfaces Removal (HSR)

Yes

Yes

Yes

Frame-buffer compression

Yes

n/a

Yes

Z-buffer compression

Yes

n/a

Yes

Other techniques

Yes

Yes

Yes

As you see, the abilities of DeltaChrome solution in terms of shader processing is far beyond the DirectX 9.0 minimal requirements. Will the game developers really take advantage of it? It is hard to answer this question, actually. But if this is ever destined to happen we will have to wait for quite a long time, maybe until these features become a widely spread standard.

Besides that, DeltaChrome features bi-directional color and Z buffers, supports hardware GDI+ acceleration, can lay up to 16 textures per single pass and perform 2x supersampling and up to 16x anisotropic filtering. Unfortunately, the current driver version doesn’t support multi-sampling, which the company also claims as one of the features.

Theoretically, DeltaChrome looks like a very successful solution: the chip boasts a lot of interesting features and strikes as pretty competitive. However, it is high time we shifted from beautiful and promising but “theoretical” tables to the real graphics accelerator and find out what the new S3 baby is actually capable of.

S3 DeltaChrome S8 Graphics Card

Our today’s hero arrived in our lab absolutely “naked”: just a graphics card and nothing with it, even the CD-disc with the drivers was missing. The matter is that this sample of DeltaChrome based graphics card was not a retail product, but a pre-production sample. Nevertheless, the card looked very nice and made a very positive impression right away.

 

The aristocratic PCB color, neat mounting, no temporary wiring, all this seemed very pleasing. It is interesting that the PCB is equipped with a variable resistor. I don’t know what it was actually meant for, but decided not to touch it in order not to damage the sample.

The PCB design is very simple. It is hardly any heavier than the design of the RADEON 9600 XT. The BIOS chip is located on the right together with the Silicon Image Sil64CT64 TMDS-transmitter. All power supply circuits are laid out in the upper left corner of the PCB. Judging by the number and size of the elements, DeltaChrome really does consume very little power, just like S3 Graphics claims. Despite the 0.13micron manufacturing process, the card is still equipped with an additional power supply connector, which has already become a common thing nowadays, I should say. Nevertheless, ATI RADEON 9600 PRO and 9600 XT do not have this connector and are quite happy with the power they get through the AGP slot.

The PCB is equipped with 8 K4D263238E-GC2A memory chips from Samsung with 2.8ns access time. The memory working frequency in our case was 300MHz (600MHz DDR). The core also worked at 300MHz. Unfortunately, we didn’t manage to remove the cooler from the chip, because it was deadly stuck to it, that is why we cannot tell you what chip revision our sample is based on. The cooler was a ChromeOrb from ThermalTake, however, I am sure that it is a temporary solution, because it is pretty tall and occupies the next PCI slot. Besides, it produces quite loud noise, because of the high-speed fan used in it. Since the chip consumes only about 4W of power, it will feel quite fine with a small cooling solution like those used on RADEON 9600 PRO, for instance.

I didn’t manage to carry out any overclocking experiments with our solution, because even though the latest PowerStrip version did recognize the graphics adapter correctly, it refused to allow any overclocking: both engines responsible for core and memory frequencies tweaking stayed passive gray. As for 2D image quality, it proved pretty good: in all resolutions up to 1600x1200x75Hz the text remained pretty clear. In 1600x1200x85Hz we noticed some slight blurring. Of course, it is hard to evaluate the 2D image quality with just an engineering sample at hand, because it will finally depend on the graphics card manufacturer. It is quite possible that some DeltaChrome based graphics cards will boast crystal-clear image, while the solutions from less responsible manufacturers who decided to save on the PCB design may “please” the users with irritating blurring in 1024x768 and up. Anyway, if you have an LCD display with DVI interface, you do not have any causes for concern anyway, because the image quality in your case doesn’t depend on the graphics accelerator at all.

Drivers and Settings

The driver version we received from S3 and used for our test session had quite many pages:

 

 

 

Each page is responsible for a list of corresponding functions, but the most interesting ones are S3Chromo and S3Config D3D.

The first page is devoted to Chromotion management:

 

There are two sections. The first one is called Artistic License Effects and is intended for real-time video effects application. The second one, Deblocking Filters, serves to improve the quality of low-quality videos.

S3Config D3D name speaks for itself: this page contains Direct3D settings.

There are not too many of them here, but all the necessary settings are available. The anisotropy level can be adjusted from disabled to 16x, while for FSAA mode you will be able to set only one option: 2x supersampling.

As for OpenGL settings, there are simply no settings of the kind, which made it simply impossible to adjust any anisotropic filtering or anti-aliasing modes as well as enable/disable Vsync. I assume that they are going to add the corresponding page later in the future S3 driver versions.

Testbed and Methods

Since S3 DeltaChrome is a new solution we decided to carry out the complete test session, including the full range of theoretical tests, which will indicate the efficiency of the architectural solutions introduced by the developer. We will compare DeltaChrome S8 with the mainstream graphics accelerators on ATI RADEON 9600 PRO, ATI RADEON 9600 XT, NVIDIA GeForce FX 5700 Ultra and NVIDIA GeForce FX 5600 Ultra.

Our testbed was configured as follows:

Performance in Theoretical Benchmarks: Fillrate Examination

As usual we will start the theoretical tests with the fillrate investigation. Let’s check if the claimed number of 2.2GPixels is true. At first come the data obtained with disabled multi-texturing:

Low VPU frequency does its dirty deed: the newcomer falls behind all the rivals. With enabled FSAA the fillrate is considerably lower: all the competitors use a more optimal multisampling mode, while DeltaChrome uses supersampling.

Now let’s enable multi-texturing:

As you see, DeltaChrome shows results, which are very close to the theoretical ones. 8 rendering pipelines allow the newcomer to outperform all the competitors. Even RADEON 9600 XT with the VPU working at shy-high 500MHz frequency couldn’t demonstrate such impressive performance. Unfortunately, with enabled FSAA the solution from S3 lost its positions immediately. This diagram can actually be a perfect illustration to an article about the advantages of contemporary FSAA methods.

Now we are going to test the new product with the help of a special test utility, which we have already introduced to you in our article called ATI RADEON 9600 XT vs. NVIDIA GeForce FX5700 Ultra. Here is the result obtained with color writes and Z writes enabled:

As you see, starting with two textures DeltaChrome outperforms the rivals, which is not at all surprising keeping in mind its 8-pipeline architecture. The remarkable thing is that DeltaChrome runs pretty fast when it works only with the Z-buffer, however, even when it has to lay only one single texture the fillrate drops by about 50% below the maximum theoretical value of 2400 Mpixels/sec. The chip probably has not very big texture caches and the caching algorithms it uses for textures, pixels and Z-buffer is not as optimal as the competitors’ solutions.

However, there is actually nothing to be surprised at. ATI and NVIDIA have already released a few generations of successful graphics processors, while S3Graphics can only boast Savage4 and Savage2000.

With disabled Z writes the results looked as follows:

The graphics accelerators from ATI and NVIDIA hardly even noticed that Z writes got disabled, however, DeltaChrome reacted in an absolutely different manner: with no textures the fillrate dropped quite significantly. Now let’s enable Z writes and disable color writes:

Again the picture is pretty unexpected: DeltaChrome loses its advantage in case of more than 1 texture, while all other testing participants do not notice this change at all. The maximum performance can be achieved in case of no textures and in case of only 1 texture. I should say that DeltaChrome behaves very strangely: the results show that it still continues working on the textures even when the color writes is disabled. Moreover, the results achieved in this case are hardly related to those obtained with enabled color writes. Hopefully this can be cured in the next driver versions. Although I cannot quite understand why the driver can matter at all for a low-level benchmark like that, if the graphics processor architecture is adequate enough.

Synthetic Benchmarks: Pixel Shader Performance

This utility can also tell us quite a bit about the performance of this or that graphics adapter during shader processing:

During per-pixel lighting processing DeltaChrome appears in between GeForce FX 5700 Ultra and RADEON 9600 XT. With simple pixel shaders 1.1 DeltaChrome showed its best performance. It is not so successful with shaders version 2.0, although even in this case its performance is close to that of RADEON 9600 PRO. Not bad for S3 Company, which hasn’t been in this market for three years!

With disabled Z writes nothing actually changes.

And as soon as we disable color writes the situation no longer looks so rosy: S3 DeltaChrome is a complete loser in all tests. The new VPU runs the worst during per-pixel lighting processing. Maybe it is the low working frequencies that cause this disaster, or again, the raw drivers.

In order to represent the results in a more convenient way, let us take a look at Pixel Shader performance in 3DMark 2001SE and 3DMark03 benchmark suites:

In case of pixel shaders 1.1, S3 DeltaChrome S8 can hardly boast any remarkable achievements, although it manages to outperform GeForce FX 5600 Ultra in two modes out of three.

When we use pixel shaders version 1.4 the situation improves a little bit: 8 rendering pipelines allow DeltaChrome to retain the parity with GeForce FX 5700 Ultra despite the pretty low working frequencies.

The situation with pixel shaders version 2.0 looks much worse, I should say. Only in the lowest resolution DeltaChrome appears a little bit faster than GeForce FX 5600 Ultra. I have to admit that pixel shaders processing is not among the trumps of the new S3 DeltaChrome chip. However, NVIDIA receives help from ForceWare compiler, while S3 VPU has to rely only on itself. Maybe S3 Graphics software developers will follow in NVIDIA’s footsteps and create a compiler similar to the one integrated into ForceWare, which will improve the performance of DeltaChrome during shaders processing. In fact, there is a lot of room for experimenting as the new solution we are reviewing today boasts much more advanced features list than the DirectX 9 specifications.

In fact, the results obtained in theoretical benchmarks are far from being optimistic. Now let’s test our solution in ShaderMark 2.0. Just like in case of NVIDIA graphics adapters, this test package refused to recognize full compatibility of S3 DeltaChrome VPU with DirectX 9:

In fact, the reason seems to be just the same: the current S3 driver doesn’t support floating point textures and render targets under DirectX 9. It must be corrected in the upcoming driver versions. Nevertheless, let’s have a look at the performance of our today’s hero during pixel shaders processing:

DeltaChrome demonstrated pretty low results here, however, it outperformed GeForce FX 5600 Ultra almost everywhere except Per Pixel Wood shader, where it suffered a complete fiasco having performed unacceptably slow for a contemporary mainstream VPU. I should admit that the newcomer coped pretty well with per-pixel diffuse lighting: it even defeated the solution from NVIDIA here. Nevertheless, there are not that many reasons for optimism: a product tending to become an up-to-date graphics solution for next generation games will never elbow its way through with shader performance like that. It could be flexible but very complex DeltaChrome architecture that caused this devastating result, as well as low working frequencies. The availability of 8 pipelines can somehow make up for the latter, while the efficiency of shader code processing is something the company still has to work hard on.

Anyway, the chip as is looks not that bad against the background of NV31/36.

Synthetic Benchmarks: Geometry, Vertex Shaders Speed

We will start with measuring the efficiency of the classical T&L algorithm:

In this geometrical test from the 3DMark 2001SE DeltaChrome performs not bad at all: with the VPU working at only 300MHz it manages to outpace RADEON 9600 PRO. It probably also owes this success to the 4 vertex pipelines it features.

However, as we get more light sources involved, the new chip immediately slows down and appears in the very last position. The vertex processors are not efficient enough to ensure proper lighting processing. There is another proof to the point: despite 4 vertex processors, DeltaChrome S8 is as fast or slower than RADEON 9600 PRO with two pipelines and 400MHz working frequency.

Vertex shaders unit performance again doesn’t impress us. However, in 640x480 and 1024x768 DeltaChrome manages to outperform GeForce FX 5600 Ultra. Is it the price they have to pay for the flexibility of the vertex shader unit or the consequence of the low VPU and memory frequencies? Unfortunately, I do not know the exact answer to this question, but the results we obtained are far from nice.

In 3DMark03 DeltaChrome demonstrates much better vertex shaders processing speed than in 3DMark 2001SE. The new solution appears quite a worthy competitor to ATI RADEON 9600 PRO. The performance of DeltaChrome S8 is very likely to depend a lot on inefficient internal architecture aimed at higher flexibility rather than faster speed. In other words, we have another GeForce FX here. The chip can theoretically support a number of improvements compared with DirectX 9.0, however, these advantages are never used because of low performance. Even though the performance drop is not so dramatic as by competitors when we shift from shader version 1.1 to version 2.0.

Bump Mapping, etc

Now let’s take a closer look at the rest results obtained in theoretical 3DMark 2001SE and 3DMark03 tests:

EMBM bump mapping can also hardly be called an advantage of DeltaChrome solution. In this benchmark our hero was the last to arrive, even though it appears not so far behind GeForce FX 5600 Ultra. Maybe it is the not quite optimal textures processing algorithms that matter in this case.

In case of Dot3 the situation turns out even worse: the S3 solution is 1.5 or more times slower. I have to admit that S3 DeltaChrome is not that strong here, too.

Ragtroll test is known to be a combined test loading both: the system CPU as well as the vertex shader unit of the graphics adapter. Since DeltaChrome works at pretty low frequencies and cannot boast highly efficient vertex shader unit, the results turn out quite predictable.

The newcomer has hard times processing sprites. Even ATI Technologies’ graphics solutions, which are usually slower than their competitors from NVIDIA in this test, easily defeat DeltaChrome here. Again this could have been caused by the caching and vertex power problems. In fact, this benchmark tests the fillrate and vertex processors.

Image Quality: Problems in Games

Now we should dwell on the performance of S3 DeltaChrome in games. Here I have to point out one important thing first: the driver version we had at our disposal during this test session had no page with OpenGL settings, although Direct3D page was there. Moreover, we discovered quite a few visual problems in many games. In particular, F1 Challenge 99-02 didn’t have the image in back-view mirrors.

In Tron 2.0 some textures and neon lighting were missing in case we enabled FSAA:

Besides, the resolutions menu didn’t allow setting 1280x960. It looks as if the current S3 driver simply doesn’t support this mode at all.

In Unreal Tournament 2003 the Link Gun ray is not visible:

In X2 – The Threat test some space crafts suddenly turned red and a few textures disappeared in the middle of the game. It happens very rarely, but it does happen. Unfortunately, we failed to take a screenshot of this phenomenon.

IL-2 Shturmovik: Forgotten Battles game and Final Fantasy XI Official Benchmark 2 “pleased” us with a black screen when we enabled FSAA.

HALO game warned us that the current graphics adapter will not provide acceptable gaming performance. We ignored this warning and the game started OK.

FSAA mode worked only in 1024x768. And every time we tried to start a Direct3D game in a higher re solution resulted into game termination and returned us to the desktop.

Image Quality: Tri-Linear and Anisotropic Filtering

All contemporary VPUs can perform anisotropic filtering. Enabling this function can significantly improve texture quality in almost any game or application.

In a way DeltaChrome is a kind of S3 Savage4 and Savage2000 successor that is why we expected the new S3 chip to be able to perform tri-linear filtering within a single pass of the pixel pipeline, just like any other S3 graphics chip can do. And our expectations came true. Now you will hardly surprise anyone with a feature like that: the texturing units of contemporary graphics processors can also perform single-pass tri-linear filtering. However in the times of Savage4, this was a truly unique feature of the product. Firstly, because Savage4 were the only graphics processors in the consumer graphics market before GeForce 256, which could perform tri-linear filtering at all. And secondly, and the most pleasing fact, the best tri-linear filtering provided by this solution was almost free of charge!

Let’s dwell on this feature of Savage4 and DeltaChrome solutions for a while. The “basic” texture filtering method is bilinear filtering with MIP-mapping. To eliminate the noticeable effects remaining when we shift to the next MIP-level (as the camera moves on the texture level of detail changes dramatically from one MIP-level to another), we use tri-linear filtering.

In fact, tri-linear filtering is a kind of “addition” to bilinear filtering, which serves to eliminate the visual bugs when we shift between the MIP-levels. For this purpose two values for the texture color are calculated: bilinear filtering is performed on two neighboring MIP-levels at a time. Then both texture color values with the corresponding weight coefficients are summed up: we perform linear interpolation between them. This is where the name of this texture filtering algorithm actually comes from: tri-linear filtering is bilinear + linear filtering.

Contemporary graphics processors perform tri-linear filtering directly: they select 8 samples from the texture (2x2 block on each of the neighboring MIP-levels), perform two bilinear interpolations and then perform linear interpolation between the two obtained values. Both operations are implemented on the hardware level, “in silicon”, and are carried out within one pass of the pixel processor.

Savage4 and DeltaChrome use a different algorithm for tri-linear filtering. It is known that each ongoing MIP-level of the texture is a less detailed version of the previous one. Moreover, the color of each texel on any of the texture MIP-levels is an average value of four (2x2) texels of the previous MIP-level.

This way, the texel colors from the 2x2 block taken from any of the MIP-levels is an average color value of the 16 texels (4x4 block) from the pervious MIP-level. In other words, if you have the values for 16 colors of any MIP-level, you will be able to calculate the colors of 4 texels from the next MIP-level. As a result, with the data for only one MIP-level at hand we can perform bilinear filtering on two MIP-levels located next to the given one.

All this little by little reveals the idea behind tri-linear filtering implemented in DeltaChrome: instead of selecting 4 texturing samples from each MIP-level, DeltaChrome can select 16 samples from only one MIP-level and then sum them up with corresponding weight coefficients, so that the obtained result is math1ematically correct and corresponds to the result of “traditional” tri-linear filtering. All the calculations involved into tri-linear filtering algorithms from S3 Graphics are not that complicated, so that even Savage4 learned to perform then within a single pass.

Of course, DeltaChrome can also perform “classical” tri-linear filtering, but their own “home” implementation algorithm boasts a lot of advantages over the traditional ways. First, in any moment of time the pixel pipeline texturing unit requires only one MIP-level for any type of filtering that is why the texturing cache can be used in a more optimal and efficient way. DeltaChrome doesn’t have to store two MIP-levels of one texture in the cache, so it can use the available space for additional data from the same MIP-level. This increases the probability that the data requested by the texturing unit in a given moment of time will be already stored in the cache, which will save time and trouble requesting this data from the graphics memory and waiting for it. This definitely reduces the amount of data to be transferred along the memory bus. Second, it is very convenient to use texture compression together with this tri-linear filtering method. For example, the basic S3TC algorithm works exactly with 4x4 texel blocks. Texturing data can be compressed “on the fly”, transferred along all channels (even the sampled texturing blocks can be compressed), and unpacked directly before use. It will also significantly reduce the graphics memory bus workload.

Anisotropic filtering as performed by DeltaChrome also boasts a few remarkable peculiarities. The details about anisotropic filtering implementation haven’t been made public yet, but we still have the opportunity to evaluate some specific advantages of this algorithm. At first have a look at the screenshots taken in 3DMark 2003, which illustrate the work of bilinear, tri-linear, anisotropic, and a combination of tri-linear + anisotropic filtering algorithms. For your convenience the MIP-levels are highlighted.

Bilinear filtering

S3 Graphics

NVIDIA

ATI

 

   

In case of bilinear filtering DeltaChrome didn’t surprise us with anything.

Tri-linear filtering

S3 Graphics

NVIDIA

ATI

 

   

Here DeltaChrome evidently used the “classical” tri-linear filtering algorithm. If it had applied its “own” algorithm, the image would have been just the same as in case of “regular” tri-linear filtering. However, the highlighted MIP-levels would have shown that it has used only bilinear filtering. The explanation of this phenomenon is very simple: since DeltaChrome uses the info taken from only one single MIP-level to perform tri-linear filtering, the “highlights” for different MIP-levels will never get mixed together.

2x anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

   

4x anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

   

8x anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

   

16x anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

                                      

 

Well, the quality of anisotropic filtering provided by DeltaChrome is at least not any worse than that from ATI and NVIDIA solutions. I would like to point out right away one of the advantages of DeltaChrome VPU over the ATI chips, which is the absence of “inconvenient” angles, and over NVIDIA chips, which is the support of higher anisotropy level.

2x tri-linear + anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

   

4x tri-linear + anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

  

8x tri-linear + anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

   

16x tri-linear + anisotropic filtering

S3 Graphics

NVIDIA

ATI

 

                                    

 

Judging by the screenshots with highlighted MIP-levels, DeltaChrome doesn’t use tri-linear filtering together with anisotropic filtering, despite the settings. However, since there are no typical MIP-banding artifacts on the non-highlighted screenshots, we still can conclude that tri-linear filtering is not missing. All this means that anisotropic filtering algorithms used by S3 Graphics DeltaChrome chip are based on the company’s brand name tri-linear filtering algorithms.

We don’t know the details behind this algorithm yet, but we dare suppose that unlike NVIDIA solutions, DeltaChrome doesn’t use 8 fully-fledged bilinear or tri-linear sample blocks when performing 8x anisotropic filtering, for instance. Instead, the graphics processor selects only two “brandname” 4x4 sample blocks from a single MIP-level. The reference points in these sample blocks are located along the viewing axis projection on the texture. And in these points they perform certain actions imitating tri-linear filtering. In other words, the weight coefficient matrix of the texels taken from 4x4 blocks is the only thing that actually changes during anisotropic filtering. “Large size” of the 4x4 block, if I can say it this way, probably allows using only one block even in case of 4x anisotropic filtering.

As for the texture quality in the DeltaChrome pictures above, there are two things worth mentioning: first, there are no “inconvenient” angles (the anisotropic filtering algorithms by ATI lose their anisotropy level at certain angles), and second, there are no “simplified” tri-linear filtering like by NVIDIA’s solutions. These simplifications just make no sense for DeltaChrome with its tri-linear filtering algorithms. The only drawback we noticed in images provided by DeltaChrome is extremely high level of texture detail (LOD). You can tell this by the distance to the MIP-level borders compared to what we see on the screenshots taken from competitor products, and by the level of “noise” on the textures. Hopefully, in the next driver versions S3 Graphics will reduce the LOD to a common denominator or will allow adjusting it manually.

As for the performance with enabled anisotropic filtering, DeltaChrome prepared us a very pleasant surprise: the performance hardly reduces even a little bit when we enable anisotropic filtering. Moreover, it doesn’t at all depend on the level of anisotropy. To see my point, take a look at the benchmarks results for DeltaChrome and competing solutions from NVIDIA and ATI with forced anisotropic filtering of different levels:

Well, there is really something unique about DeltaChrome: none of the two industry leaders has yet offered “free” anisotropic filtering. Moreover, as we see, the quality of anisotropic filtering from S3 Graphics is beyond any praise. Keeping in mind that ATI and NVIDIA do introduce certain quality limitations in order to increase the performance of their solutions, we have every right to state another victory of S3 Graphics: the “free” tri-linear filtering in 1999 has now grown into “free” anisotropic filtering! This is an excellent achievement for a company, which has been de facto absent in the discreet graphics market for a while.

In the upcoming reviews, when new more stable DeltaChrome driver versions appear, we will do our best to devote more time to consideration of such interesting things as anisotropic filtering and full-screen anti-aliasing implementation by DeltaChrome. And now let’s pass over to the gaming benchmarks and the performance of our today’s hero in this type of applications.

Performance in Gaming Benchmarks

Since DeltaChrome drivers do not allow using multisampling yet, and enabling supersampling could be unfair to the new newcomer, we decided to run the tests in Pure Speed mode only, that is without FSAA and anisotropic filtering. Especially, since the tricky anisotropic filtering algorithm implemented by S3 Graphics doesn’t lead to any performance drops. As for the gaming tests used this time, the list looks as follows:

First person 3D shooters:

Third person 3D shooters:

Simulators:

Real-time strategies:

Semi-synthetic benchmarks:

Synthetic benchmarks:

Just a reminder: IL-2 Sturmovik: Forgotten Battles and Final Fantasy XI Official Benchmark 2 were not included into the list of our today’s tests because of image quality problems. The same happened to the Highly Anticipated DX9 Game, where the image quality was simply terrible. Note that the results in the OpenGL games shouldn’t be regarded as final, because the OpenGL ICD from S3 hasn’t been finalized yet and even doesn’t have a separate page in the drivers, which makes it impossible to adjust any settings.

DeltaChrome runs pretty slow here, but please do not forget that it is because of the raw OpenGL driver. As soon as the new S3 Graphics drivers come out, the situation may change dramatically.

The same is true for StarTrek: Elite Force 2 game, which also uses OpenGL. We have to be patient and wait for the new S3 Graphics drivers.

And in Unreal Tournament 20003 DeltaChrome feels quite at home: the newcomer is capable of competing successfully with RADEON 9600 PRO, even though it is still a little bit behind the latter. As the resolution grows, the gap becomes bigger. Both: DeltaChrome S8 and RADEON 9600 PRO feature graphics memory working at 300MHz (600MHz) frequency. However, ATI is known for its efficient technologies for saving the memory bus bandwidth. The techniques used by S3 seem to be less efficient here. But nevertheless, this S3’s baby looks quite attractive.

Now let’s find out what the situation in Antalus benchmark looks like:

In this case S3 DeltaChrome S8 is also far from fiasco: it lags less than 10fps behind RADEON 9600 PRO. And this is even despite the fact that ATI Technologies boasts a huge experience in the field of powerful video processing units development, while DeltaChrome is actually one of the first S3’s attempts in this field.

Well, the things are not that bad at all in HALO game, which is the “hungriest” of all. DeltaChrome falls 5fps at the most behind RADEON 9600 PRO and runs almost as fast as GeForce FX 5600 Ultra. Anyway, only the RADEON 9600 XT can ensure the acceptable minimal performance here.

As is known, Splinter Cell likes ATI VPUs. This simple truth got proven once again here. DeltaChrome managed to perform as fast as GeForce FX 5600 Ultra and fell just a tiny bit behind GeForce FX 5700 Ultra. Note that NVIDIA based graphics cards were greatly supported by the software shader code compiler, while the solution from S3 had to rely only on itself.

In Tomb Raider: Angel of Darkness the S3 DeltaChrome also performed very well. The only exception appeared 1600x1200 resolution when the game simply refused to start. It is probably the raw drivers again that spoilt the show. As for the performance, it almost corresponds to that of GeForce FX 5700 Ultra again.

In Tron 2.0 DeltaChrome failed to show its best: although the card outperformed GeForce FX 5600 Ultra in 1024x768, it fall a bit behind it in 1600x1200. Unfortunately, this game uses 1280x960 resolution instead of the wider spread 1280x1024. The current S3 driver doesn’t support this gaming mode at all that is why we failed to run the benchmark in this resolution and hence have no results for it.

Command & Conquer Generals is a strategy. The distinguishing feature of these games is the presence of a plenty of various models on the screen. Moreover, the viewing angle is mostly the same throughout the entire game. Maybe the failure of DeltaChrome has to do with the low efficiency of its vertex processors, but I assume it is again the drivers, which are to blame here.

This air-battle simulator is very friendly to NVIDIA’s solutions and doesn’t like ATI that much. DeltaChrome performed not bad here in all resolutions except 1600x1200, where the less efficient algorithms for work with the memory subsystem did its black deed.

I cannot find any reasonable explanation for this more than double failure other than the raw drivers.

In Aquamark3 DeltaChrome doesn’t impress us that much, although its performance is comparable to that of GeForce FX 5600 Ultra. Again the slowdown in high resolutions is inevitable.

The newcomer from S3 looks pretty modest in the first benchmark from the 3DMark03 test package. It falls behind all the rivalry products. In fact, this is not a very interesting benchmark, because it uses only DirectX 7. DeltaChrome will hardly be a good choice for those games, which use this old Microsoft API.

In the second gaming benchmark the situation is just the same. As you know, the rendering algorithms used by Futuremark in the second and third tests of the 3DMark03 test package are far from being optimal.

In the third gaming benchmark DeltaChrome is again unable to demonstrate any competitive results. Inefficient work with the textures? Possible, but I wouldn’t claim that this is it yet.

The new S3 solution looks surprisingly nice in the Nature test. It doesn’t strike us with outstanding results, but looks very nice running neck and neck with GeForce FX 5600 Ultra, without any additional support from anything like shader code optimizers or the like.

Now we have a more or less clear picture. DeltaChrome feels best of all in contemporary games and applications using the features of the latest DirectX version, in particular, pixel shaders 2.0. Of course, the results of the new VPU are still not very high, and are even a complete disaster in some tests, but we could explain it by the not finalized drivers, which has always been the weak point of all s3 solutions: the company introduced great chips and then provided them with awful drivers.

Conclusion

We have just tested the first graphics adapter based on the newest graphics processor from S3 Graphics – DeltaChrome. Having taken a closer look at the product, we arrived at the following conclusion: this is a very interesting graphics processor featuring a bunch of interesting technologies. I would like to specifically stress highly efficient real-time video editing and very interesting anisotropic filtering algorithm ensuring high image quality without any noticeable performance losses.

So, what does the new S3 Graphics DeltaChrome S8 offer us today?

All in all this is very nice. Low results in some benchmarks show not the weak spots of the DeltaChrome architecture that much, but the low-quality drivers, which we had at our disposal. Today these drivers are still very raw: take the absence of the OpenGL settings, for instance. If the S3 Graphics software developers manage to eliminate these drawbacks in the nearest future, then DeltaChrome has every chance to become a very attractive solution and win a stable position in the desktop VPU market.

However, the developer will have to work really hard to achieve this goal. Here I imply not only developing good, properly working drivers, but also establishing cooperation with the graphics card manufacturers, which are known for their lack of trust in the new solutions appearing in the market.

If S3 Graphics manages to cope with this task, if the graphics cards based on DeltaChrome chip cost reasonable money for the performance level they provide, and if the drivers are not any worse than those from ATI Technologies or NVIDIA Corp., then S3 graphics solutions will definitely win their market and become popular in certain user groups, which value multi-display configurations, advanced video processing options and DirectX 9+ features support. Due to low power consumption the new chip will also find its niche in the systems where the economical factor is vital, such as SFF-systems and notebooks.

So, what are those features, S3 Graphics DeltaChrome solution cannot offer you so far?

It will be tough for S3 to win the users’ hearts, because the company has been absent in the market for a significant period of time, so that many users have already forgotten the name of S3 Graphics Company. The way back to public acknowledgement will be a hard one but S3 is very determined to succeed. The major thing for the company now is to avoid the mistakes, which have already caused a failure once.

We believe that S3 Graphics Company should now focus on developing quality drivers for their new solution, because the chip already looks very mature and has every chance to become a success.

We would like to welcome S3 Graphics back in the desktop graphics market! :)