by Alexey Stepin , Anton Shilov , Tim Tscheblockov
01/15/2004 | 01:47 PM
It has happened so historically that the cut-throat competition in the 3D graphics market is mostly created by such monsters as ATI Technologies and NVIDIA Corporation. However, other, less popular players also pop up there from time to time, representing serious threat and trying to win a tasty piece of the market for themselves. We have already told you about S3 Graphics Company, which has undertaken an attempt like that (see our article called The Return of S3: DeltaChrome Graphics Card Review). Today we are going to talk about XGI Technology.
The name of XGI Technology hides none other but the graphics division of Silicon Integrated Systems (SiS), which has been announced a separate independent company in the middle of 2003, and Trident Graphics, which was purchased from trident Microsystems last summer.
SiS and Trident Companies were pretty well-known in the graphics market in mid 90s of the past century. However, when the graphics market evolved to 3-dimensional environment they lost their entire popularity by the end of the decade. Although they undertook numerous attempts to return to the 3D graphics market, none of them appeared a success. If Trident Graphics chips enjoyed some demand in the mobile PC market, then in the desktop field NVIDIA and ATI didn’t leave them any hope at all.
Nevertheless, the engineering team from SiS, which has been long working on the new graphics processors, continued developing their Xabre project. The first supplies of these DirectX 8.1-compatible VPUs were shipped in April 2002.
This product family was quite progressive from the technological point of view, however, it was not free from a couple of serious drawbacks, unfortunately. One of the major disadvantages of the Xabre solution from SiS appeared the absence of hardware vertex shader unit, so that the system CPU had to do this work itself. But a real stumbling stone for Xabre, which then determined its failure in the market, was low-quality drivers and the so-called turbo-texturing mode. Hunting for high competitive results at any rate, the company introduced the so-called turbo-texturing mode, which boosted the performance by reducing the texturing quality quite significantly. Of course, turbo-texturing allowed Xabre solutions to run as fast as NVIDIA GeForce3, but the resulting 3D picture was a total disaster: the image quality was simply catastrophic! Moreover, the chip couldn’t perform anisotropic filtering, and that was in 2002 already! Together with the absence of the quality settings in the drivers, all these issues couldn’t help telling on Xabre’s fortune: the solution hasn’t become any popular in the market. The launching of Xabre 600 didn’t change the state of things, because the only difference of this chip from the predecessors was its higher working frequency and integrated VIVO support. (Read more about Xabre solutions from SiS in our articles called Xabre The Brave: New 3D Hero from SiS and SiS Xabre600 Graphics Chip Review: SiS' Newcomer against Competitors from NVIDIA and ATI)
Finally, SiS decided to separate its graphics division into an independent company, which still belonged to SiS. So, this happened. The new company was called XGI (eXtreme Graphics Inc., according to the company officials) and started working on June 1, 2003. The rumored Xabre II project started taking a more or less material form inside the new company, having lost its initial name though, because XGI had no right for the Xabre trademark. Moreover, on June 12, 2003 the company announced their acquisition of the Trident Microsystems graphics division, which was supposed to strengthen its potential.
Note that the year 2002 was extremely unhappy for Trident: DirectX 8.1 chip aka Blade XP4 suffered a total fiasco. It failed to get through to the desktop market and at the same time didn’t become any popular in the mobile PC market.
A few months after XGI Company had been established, they felt so confident that they announced their vital intention to become profitable within the next three years. The company really decided to seriously tackle the 3D graphics market, instead of focusing on the low-end products, as they had been doing before. This was a very optimistic announcement for a company, which has been very unlucky in the graphics market for a long time.
Then September came…
On September 16, 2003 the rumors have finally come true: XGI announced a new DirectX 9.0 graphics processor family. Since they couldn’t use the Xabre name for a number of reasons, the newcomer was called Volari. According to the graphics market traditions, the family was split into a few models differing from one another by working frequencies and pipelines configuration.
The top model in the new family appeared Volari V8 Ultra working at 350MHz core frequency and featuring 8 rendering pipelines. Volari V8 differed from the top model by the working frequency only, which was equal to 300MHz. As we have expected, Volari V5 Ultra also worked at 350MHz core clock frequency, but was equipped with only 4 rendering pipelines. And Volari V5 in its turn differed from the Ultra version by lower working frequency.
Unfortunately, the architecture of these new solutions caused some concerns about their performance. The thing is that Volari featured only two vertex processors, while RADEON 9500/9700/9800 boasted four of them, and GeForce FX 5700/5900/5950 featured three vertex processors. As for the pixel shader units, Volari V8/Ultra were provided with 4 units like that, and Volari V5/Ultra – with two. Besides, the entire family featured 128bit DDR or DDR-II memory bus for each chip.
Well, it looked as if they has already lost the fight even before it actually started. But then XGI suddenly made an unexpected move. The company made it possible to use the new chips in dual-chip configurations, which has been forgotten since the times of 3dfx Voodoo5 5500.
The new DirectX 9.0 features, enhanced memory controller, dual-chip configurations support: can it be true that XGI Volari is more than SiS Xabre II?
Dual-processor graphics cards have never become very popular and widely spread, because they consumed a lot of power, cost more and besides, the driver quality left much to be desired, and the performance growth they provided was incomparable with the price growth all these innovations caused.
Theoretically, a dual-chip solution on Volari V8 Ultra looks not bad at all: two 128bit memory buses provide the total of 256bit with the bandwidth of 32GB/s (in case they use GDDR-II 1000MHz). And the number of rendering pipelines increases up to the impressive number of 16, thus ensuring enormous fillrate of 5.6Gpixels/s. As for the supported memory types, single-chip Ultra solutions were supposed to work with DDR memory chips working at 750MHz, and the non-Ultra models – with 650MHz memory. The more powerful dual-chip products were intended to work with GDDR-II supporting 900MHz or even 1000MHz frequency.
The approach that allowed XGI to host two chips on a single PCB was called BitFluent. According to the technical documentation available on XGI web-site, the chips are connected with one another with a special 32bit bus working at 133MHz (the efficient frequency is equal to 533MHz), and providing the bandwidth of 2.13GB/s. From the hierarchical point of view, the first chip connected to the AGP bus is the primary one, while the second one connected to the primary chip with the above described bus is the secondary one. Here I would like to point out that AGP 8x and BitFluent feature a lot of similar technical details, that is why the latter is very likely to be almost a full analogy of the AGP 8x.
According to XGI, each VPU processes its frame. Moreover, both VPUs work in parallel and neither of them has to wait for the counterpart to complete its work. Of course, each VPU features its own frame-buffer and memory bus.
BitFluent technology boasts a number of enhancements, which allow minimizing the latencies and increasing the inter-chip bus efficiency.
It is a pretty funny observation, I suppose, but even though all multi-processor graphics systems intended for both: the mainstream users and professionals, use different algorithms ensuring parallel functioning of the two chips, the questions about their efficiency are always the same.
Here is a list of potential problems XGI’s solution may face:
Well, now we should only wait and see if our concerns about the low efficiency of the new Volari Duo solution will be proven or denied by the benchmark results. And in the meanwhile, let’s dwell a little bit more on the peculiarities and features of the new graphics chip from XGI.
According to the developer, BroadBahn Memory Architecture optimized the work of the memory subsystems due to a number of technologies based on compression algorithms. Well, this is at least a kind of attempt to optimize the things, where NVIDIA and ATI spent thousands of hours of work.
I would also like to point out Cipher Video Processor technology intended to improve deinterlacing quality during DVD playback. This technology uses Pixel shaders 2.0 and ensures excellent image quality. XGI claims that the image quality in this case is as good as that provided by the top-end DVD players.
You know very well now that developing a VPU is only half of the work you need to do to successfully enter the graphics market. You also have to attract people’s attention and make potential partners interested in the new product. It is really surprising but almost immediately XGI managed to sign a very successful agreement with one of the largest European graphics card manufacturer: Club3D Company. This company is not only known for high-quality products, but also easily gets very excited about new non-standard solutions. Otherwise, how would you explain the fact that Club3D will offer the users not only the whole bunch of solutions on NVIDIA and ATI chips, but also some products based on the graphics processors from the less famous developers, such as SiS, S3 Graphics and even PowerVR?
Anyway, this company got really interested in the new Volari family. So far, Volari based graphics cards are manufactured and supplied by Club3D and CP Technology. But the rumors about ASUS, ABIT, Gigabyte, Chaintech and even MSI considering the possibility to design their cards on the XGI products continue circulating around the Web.
XGI is still very optimistic and claims that they will be able to significantly increase their supplies in 2004. Will they manage to win a part of the market where the two giants, NVIDIA and ATI, continue their eternal rivalry? What gives XGI the confidence: the technological advantages or anything else? We will do our best to answer these questions in our today’s article devoted to Club3D Volari Duo V8 Ultra.
Now let’s take a look at the features of the new Volari solution from XGI. In this case we are going to talk about the top model in the family: V8 Ultra, because the Club3D Volari Duo V8 Ultra is based on two chips like that. Since the developer positions this solution as the most expensive and powerful one, we will compare it with the competitors’ High-End solutions.
ATI RADEON 9800
Club 3D Volari Duo V8 Ultra
NVIDIA GeForce FX 5950
Number of transistors
Graphics memory controller
Graphics memory frequency
Peak memory bus bandwidth
Max graphics memory size
AGP 3.0 4x/8x
AGP 3.0 4x/8x
AGP 3.0 4x/8x
Pixel Pipelines, Pixel Shaders
Texturing unites per pipeline
Max number of textures during multi-texturing
Texture filtering types
Max anisotropy level
Pixel shaders version
v.2.0 (f-buffer, v2.0+)
Branching, subroutines and loops
Max number of textures per shader
Max number of texture instructions
Max number of arithmetic instructions
Max number of instructions per shader
2 color registers,
2 color registers,
Data representation formats
Internal Pixel Shader Pipeline Precision
96bit pixel precision/24bit floating point precision
128bit pixel precision/32bit floating point precision, 16bit floating point precision
Vertex pipelines, vertex shaders
Vertex shaders version
Branching, subroutines and loops
Max number of instructions per shader
Max number of instructions with loops extension
16 input registers,
16 input registers,
Data representation formats
Rotated grid multi-sampling
Number of samples
2, 4, 6
2 (OGSS, OGMS),
Technologies aimed at higher memory bandwidth efficiency
Hidden Surfaces Removal (HSR)
Well, everything looks quite impressive, I should say. The only exception is probably the 4x maximum anisotropic filtering mode – it is not that cool for the end of 2003-beginning 2004. Other than that, this is a pretty common set of features for the graphics accelerator compatible with DirectX 9. The VPU working frequency is not impressively high, especially compared with that of GeForce FX 5950 Ultra, however, Volari is equipped with two VPUs instead of one, so it should make up for this drawback. Anyway, let’s take a closer look at the Club3D graphics card on XGI chips.
We received this product in the full retail package including a colorful box designed in minimalistic-aristocratic style typical of this particular VGA card maker. Unlike the package designed for RADEON 9800 PRO (see our article called Club3D RADEON 9800 PRO: European Finest Graphics Card?), the today’s one was decorated with a blue stripe instead of the red one, which indicated that the graphics card inside the box is based on an XGI chip:
Well, this is a pretty simple and nice way of informing the user about the graphics chip maker the product is based on. By the way, Club3D graphics cards on S3 DeltaChrome chip will be packed in boxes marked with white stripes.
The card itself looks more than respectful. If I am not mistaken I had a similar feeling when I first saw some products from Leadtek. But Club3D Volari Duo V8 Ultra went even farther:
Here is a list of items I found inside the box:
Despite the first impression, the PCB of this graphics accelerator is only 5mm longer than that of the GeForce FX 5950 Ultra card. This is why all the electronic components are located so close to one another on the PCB, as XGI developers managed to fit here two graphics processors with their power supply circuitry and half of all memory chips.
In the right part of the PCB you can see SiS301MV chip, which is responsible for multi-display configurations and TV-Out support. And below it there is a flash-memory chip with the BIOS. The entire upper part of the PCB is loaded with capacitors, regulators and coils building the power supply circuitry of the graphics adapter. All critical elements are equipped with heatsinks, which is a very pleasing sign. In the left part of the PCB there are two standard four-pin power supply connectors. The manufacturer didn’t manage or didn’t want to unite them into a single connector. This layout can actually cause some troubles, because it might require two connectors on the PSU. According to XGI’s recommendations, Volari Duo V8 Ultra will work fine if your system PSU is at least 400W.
As for the coolers, they deserve a separate mention. A massive copper plate serves as a general heatsink for both graphics chips and the memory chips located on the front side of the PCB. The contacting surfaces of the graphics chips are covered with a layer of thermal interface and the memory chips contact the heatsink plate through flexible rubber-like thermal pads. There are two sets of thin copper ribs soldered to the heatsink plate in a circle. The whole construction is cooled down by two fans with semitransparent blades of unusual shape, which look very much like snowflakes. The fans are decorates with blue lighting and XGI name. There are three wires connecting each of the fans to the PCB, which gives us some hope for the implemented hardware fan rotation speed control system. Here is what we see when the cooler has been removed:
The gigantic size of the graphics chips is striking. The die size of each Volari VPU is even bigger than that of RADEON 9800 XT. In order to protect the dies against physical damage each Volari chip is covered with a small frame.
GDDR-II memory chips from Samsung feature 2.0ns access time and can work at the nominal frequency of 500MHz (1000MHz), though in our case their working frequency is 450MHz (900MHz).
The reverse side of the PCB looks a little simpler: we could only point out Conexant Bt835KRF chip responsible for the implementation of VIVO functions. The memory chips are covered with a massive copper plate with ribs. As is known, GDDR-II memory dissipates much more heat that is why the use of this powerful cooling solution is absolutely justified. Nevertheless, we will be able to evaluate its efficiency only during the test session, which is about to start a little bit later.
All in all, the design of this product can be called successful, although we cannot disregard its massiveness. However, despite the seemingly huge size, the card is not that big. As we have already said it is only 5mm longer than the PCB of NVIDIA GeForce FX 5950 Ultra. It is not a small card, but not an unbelievably big one for today. Club3D added a few of their own ideas to the Volari Duo V8 Ultra reference design. Dark-brown PCB lacquer, protective cooler bars of the same color – all this makes the card even more charming, so that to catch the eye of a potential customer.
When we started the system for the first time, we got truly perplexed: the noise was simply unbearable! It was so noisy that we sincerely felt like comparing it to the “legendary” NVIDIA FlowFX system. Unfortunately, since we didn’t have the reference GeForce FX 5800 in our lab at that moment, we couldn’t find out which one of them was louder. But the high-quality wailing of the two fans can hardly be considered harmless for one’s ears.
A little while later the GDDR-II memory showed its “hot” temper. Both, the front and the back heatsinks got tangibly warm, and after about half an hour of testing the thermal sensor indicated that they got up to 60oC. All this was accompanied by worried beeping of the electronic thermometer set to this temperature.
Well, Club3D Volari Duo V8 Ultra cannot be called a cold graphics solution. By using it in your system you threaten the thermal conditions inside the PC case quite severely.
As for 2D image quality, we didn’t find any issues about it: it was up to the already existing standard. The picture was very clear in all resolutions up to 1600x1200x75Hz. However, a lot also depends on the monitor and the quality of the cable connecting it to the graphics adapter.
We used the last version of the official drivers – XGI Reactor v1.01.51, which is available for free download on the company’s official web-site. When we installed the drivers Volari Tray icon appeared in the system tray. By clicking it you can access the following settings:
On the Application page you can access such functions as screen menu and hot keys, and get the info about the active AGP mode, enable multiple desktops support and launch PowerManager application:
This utility is intended for monitoring of the graphics card critical parameters and adjusting the fan rotation speeds. On the first launch of this utility the second fan suddenly stopped, which immediately told on the temperature of the graphics processor located closer to the end of the PCB. After the temperature of the second VPU reached 50 degrees, we had to close this utility and remove it from the memory, which immediately brought the second fan back to life.
Unfortunately, the option of determining the fan rotation speed is inactive yet or works incorrectly. The scales, which are supposed to reflect the status of these parameters stayed gray, and only a few times a green column appeared on one of them. PowerManager Monitor page reports about the type, family and working frequency of the CPU in the system.
Display Property menu allows adjusting the graphics card driver settings, which you can also access via the Display Properties – Advanced – Utility Manager:
From here you can control all the parameters of Club3D Volari Duo V8 Ultra, namely: gamma-correction of the image, overlay settings and displaying devices parameters.
3D Setting option is worth mentioning separately, because is contains three most interesting and important pages about the Dierct3D and OpenGL modes:
As you see, the card allows enabling FSAA up to 4x and anisotropic filtering also up to 4x. The latter was a bit disappointing for us, because all contemporary graphics accelerators have already learned to perform 8x and even 16x anisotropic filtering. However, we are going to talk more about it in the corresponding section of our article. Getting a little bit ahead of time, I would like to mention that we couldn’t enable anisotropic filtering as well as full-screen anti-aliasing at all with the older XGI Reactor 1.00.00 drivers. When we installed the latest XGI Reactor v1.01.51 drivers anisotropic filtering had finally appeared, but FSAA remained unavailable: none of the existing games reacted in any way to different positions of the slider responsible for the FSAA modes. We are absolutely sure that this issue will be eliminated with the time, and the new XGI Reactor versions will not suffer from it any more. As for the page responsible for clock frequencies, we have to confess that we didn’t try overclocking the card fearing that it will negatively tell on its “health condition”, because it already got extremely warm even at the nominal frequencies with the fans rotating at their maximum speed.
Volari based graphics cards are still considered something exotic, they have just appeared in the market. Therefore, we decided to go deep into details with our piece, which we usually do with all new solutions that get into our lab. For multiple reasons we included into our comparative analysis high-end graphics cards, such as NVIDIA GeForce FX 5950 Ultra and ATI RADEON 9800 XT, alongside with mainstream ones, represented by ATI RADEON 9600 XT and NVIDIA GeForce FX 5700/5600 Ultra.
Our testbed was configured as follows:
All drivers are the latest official versions downloaded from the developers’ web-sites.
As usual, we will start our theoretical investigation with fillrate and texturing speed. Let’s have a look at Volari Duo V8 Ultra here:
This benchmark displays a number of planes with “semitransparent” textures. In multi-texturing mode contemporary graphics accelerators allow laying 8 textures per pass, so the test scene consists of 8 planes with 8 textures on each, which makes the total of 64 texture layers. When a single texture is laid over a polygon, the scene includes 64 surfaces with one texture layer on each.
As a rule, graphics cards perform better in this test in case of multi-texturing than in case of single-texturing. When we have multi-texturing, the previous Z value and the color should be read only once for each 8 textures. However, without multi-texturing this reading should be carried out for each texturing layer anew. As a result, the graphics cards performance with enabled multi-texturing appears less dependent on the memory bus bandwidth and texturing cache structure.
What does Volari show here? The maximum result we managed to obtain in this benchmark equals 3110Mtexels/s. the theoretical maximum for our graphics solution makes (8 pipelines * 1 texturing unit) * (2 chips) * (350MHz) = 5600Mtexels/s. In other words, the obtained result makes only 55.5% of the theoretically possible maximum, which demonstrates not very high efficiency of the Volari architecture.
By the way, this result also ruined our supposition about the processors of our card having only 4 pipelines each instead of the 8 pipelines claimed by XGI. If this had been true, then the theoretical performance maximum could have been equal to 2800Mtexels/s. The practical results are higher than that, which means that we were wrong.
Now let’s take a closer look at the performance with enabled multi-texturing and the changes that take place if we increase the texture color depth, frame buffer and Z. we can clearly see that the performance of our Volari solution mostly suffers from the shift from 16bit to 32bit textures, i.e. when the chips have to request twice as many texturing data. This is an evident indication of the inefficient caching algorithms implemented in XGI Volari.
A good example illustrating this statement is tri-linear filtering. To perform tri-linear filtering graphics chips require 8 texturing values instead of 4 values usually used for bilinear filtering (except DeltaChrome with its unique tri-linear filtering algorithm, see our Review for details). If our supposition about Volari’s bottleneck is true, then enabled tri-linear filtering should cause a significant performance drop.
To check this out we decided to run a texture filtering quality test from 3DMark03 test package with included bilinear and tri-linear texture filtering:
The results prove our supposition: enabled tri-linear filtering slows down Volari’s texturing speed by 37-47%.
And as a practical proof of the results obtained in this synthetic test we ran Return To Castle Wolfenstein game with enabled bilinear and tri-linear texture filtering:
The performance of the card also gets considerably lower here when we enable tri-linear filtering, however, it is not so dramatic as in the previous test, anyway. The situation in real game is „less synthetic“.
But let’s return to our synthetic benchmarks:
Yeah, the fillrate of the newcomer is far from perfect. When there are no textures it can be compared to that of RADEON 9600 XT, however, as soon as we need to lay at least one texture the performance immediately falls down beyond all acceptable limits and stays there as the number of textures increases. We can’t explain this disaster by inefficient texturing algorithms and small texturing caches – the performance drop is too huge even compared with the results of the previous tests. Volari Duo looks OK when performing some operations with the Z buffer, but only against the background of mainstream solutions. Even the 16 pipelines do not help here, as their efficiency is pretty low, too.
Now we disable Z writes:
Nothing changes: the card still slows down when we have to process at least one texture. Now we enable Z writes, but disable color writes:
The situation here is surprisingly similar, even though all other testing participants run close to their theoretical performance having got rid of the texturing workload.
As we see, the TMU of Volari graphics chip can hardly be called efficient, which is the first big drawback of the new XGI solution.
Now let’s check our well our today’s hero can cope with pixel shaders:
The per-pixel lighting using shader version 2.0 appears a total fiasco: even the morally and technically outdated GeForce FX 5600 Ultra easily outperforms the new Volari solution. However, when we shift to simple shaders our hero proves much more successful and surpasses both youngest NVIDIA cards.
I dare suppose that XGI Volari Duo V8 Ultra features similar weak spots as GeForce FX, because its graph is shaped just like the one of GeForce FX: it looks like a small hump, where the decrease indicates noticeable performance drop when we shift to shader version 2.0.
When we disable Z writes the overall picture remains unchanged.
However, when we disable color writes, all graphics cards except Volari speed up almost hitting their theoretical maximum. Frankly speaking, the graph for Club3D solution makes us think that it doesn’t work correctly in our benchmark. Or maybe its “unique” architecture simply doesn’t allow disabling the color writes completely. Or maybe it is all about the unfinished drivers, although this is the most unlikely supposition for this low-level benchmark.
The benchmark for version 1.1. pixel shaders from 3DMark 2001 SE test package again brings Club3D to the very last place. It is not too far behind GeForce FX 5700 Ultra, but the latter belongs to mainstream solutions, while XGI positions its Volari Duo V8 Ultra as a high-end one. As we see, the benchmark results do not prove this statement yet.
Pixel shaders version 1.4 are also far from being among the trumps of Club3D Volari Duo V8 Ultra. The card fell behind all its rivals and its results can only be compared with those of GeForce FX 5700 Ultra.
Pixel shaders 2.0. This is a total disaster for Volari Duo V8 Ultra. No more comments here.
Now let’s take a look at the results in ShaderMark 2.0.
At first ShaderMark gave out the same report as we acme across during our tests of GeForce FX 5700 Ultra and S3 DeltaChrome. Only the details were a little bit different this time:
Before the testing actually started, we had already understood that we will not see some shaders at least with this driver version. Here are the results we obtained in this test:
Well, this is a pretty curious picture, I should say. Volari Duo V8 Ultra is certainly not a RADEON 9800 XT, and even not a RADEON 9600 XT, but it is also not a GeForce FX 5700 Ultra: the newcomer doesn’t process the shaders fast enough to compete with the invincible ATI Technologies, but at the same time it is quite capable of competing with the mainstream NVIDIA solutions. However, this can hardly comfort us, as it is priced as a good high-end product. Besides, ShaderMark is actually almost the only pixel shader test, where Volari manages to get a little ahead of NVIDIA. I doubt that you will be willing to pay almost $500 just for the sake of learning yourself that a much less expensive RADEON 9600 XT or GeForce FX 5700 Ultra can offer you the same or even higher performance.
Anyway, the results shown by Club3D Volari Duo V8 Ultra are extremely low. In games using high-resolution textures, as well as in games with pixel shaders version 2.0 this solution will not grant you high level of performance.
Well, it is high time we tested the vertex processors and geometrical capabilities of the new XGI Volari Duo V8 Ultra.
This is a geometrical benchmark, therefore, it demonstrates how efficient the vertex processors of the tested solutions are. Volari Duo vertex processors are very slow, even though the two chips installed onto a single PCB feature 4 of them.
At first glance it seems as if the card from Club3D were catching up with the competitors. However, in reality it is the competitors, which slow down as the number of light sources in the scene increases. We have every evidence that Volari vertex processors are slow and not very efficient, but it is very strange to see it perform equally fast with one and 8 light sources. If it doesn’t result from some cheats in the drivers, then Volari really should have some potential here. However, in the meanwhile we get the impression that something prevents T&L units (or their vertex shader emulations) from showing their real best, because as we see the number of light sources in the testing scene doesn’t have even the tiniest influence on Volari’s performance.
Vertex shader version 1.1 benchmark once again demonstrates that vertex processors of our Volari Duo are far from the fastest, even though this time the lag is somewhat smaller compared with what we saw in High Poly Count. All in all, this is not a bad result for an average solution. But again: Volari Duo is positioned as a high-end product.
The benchmark for vertex processors 2.0 included into 3DMark03 showed us nothing new: Volari’s vertex processors are really slow. All other testing participants work at much higher clock frequencies, but this cannot serve as an excuse for a product tending to compete with contemporary high-end solutions.
Now we are going to run a few other benchmarks from 3DMark2001 SE and 3DMark03 test packages:
Volari Duo cannot cope well with EMBM bump mapping, too. To be more exact, it cannot cope with it at all: 26.7fps in 1600x1200 is an indisputable failure. And we see a 500-dollar graphics card perform like that! Is it another consequence of the low texture caching efficiency?
Dot3 bump mapping finally gives us a break: Club3D Volari Duo V8 Ultra performs not bad here, although it is still behind RADEON 9800 XT and GeForce FX 5950 Ultra.
Point Sprites test is actually none other but a test of vertex processors efficiency and fillrate. Remembering the previous results we expected Volari Duo V8 Ultra to suffer a total fiasco here, too. However, the situation didn’t turn dramatic this time, and the card worked pretty fast, although only in 1024x768. After that it yielded to everyone except RADEON 9600 XT. I would blame not very efficient work with the memory here. Maybe two 128bit buses do not work as fast as 256bit buses of RADEON 9800 XT and GeForce FX 5950 Ultra. Or maybe the relatively narrow bus between the two chips with the bandwidth of a little bit more than 2GB/s limits the performance.
The results in Ragtroll test prove our supposition about architectural inefficiency of Volari Duo V8 Ultra, and once again indicate how slow its vertex processors are. Even the availability of two VPUs as well as DDR-II memory working at 450MHz (900MHz) doesn’t help here. All in all, Club3D Volari Duo V8 Ultra performed far from acceptably: it managed to compete only with a morally and technically outdated GeForce FX 5600 Ultra and sometimes with newer mainstream solutions, such as GeForce FX 5700 Ultra and RADEON 9600 XT.
Well-known Villagemark test was developed by PowerVR Company, an expert in tile architecture, which can perfectly remove invisible parts of the scenes. This test shows how well the GPU can identify invisible surfaces and remove them thus optimizing the performance of the memory subsystem.
As we see, Volari doesn’t strike us with brilliant performance here, which means that we shouldn’t hope for any breakthroughs in contemporary games with high overdraw coefficient. Volari’s results are twice as low as those of GeForce FX 5600 Ultra: this is an indisputable fiasco!
This section of our review is devoted to the quality of anisotropic filtering as implemented by XGI Volari Duo V8 Ultra. To test the quality of anisotropic filtering we took the same scene from the 3DMark2003 test package, as for S3 Graphics DeltaChrome.
No surprises from Volari here.
Volari is quite adequate with enabled tri-linear filtering.
2x anisotropic filtering
Here is the first surprise: anisotropic filtering by XGI is the same bilinear filtering but with a bit higher level of texture detail. Do you see how far MIP-level borders have been moved?
4x anisotropic filtering
MIP-level borders on the images obtained from Volari are even farther now, but texture quality hasn’t got any better on the remote surfaces. Well, this is a highly original view of anisotropic filtering algorithms, isn’t it?
2x anisotropic filtering + tri-linear filtering
No surprises here. This is true tri-linear filtering with higher level of detail at the foot of the player. No anisotropy whatsoever.
4x anisotropic filtering + tri-linear filtering
Well, anisotropic filtering by Volari has nothing to do with real anisotropic filtering, but is just a simple change of the level of detail (LOD). We can prove it very easily. Take a look at the pics showing the pixel-by-pixel difference between the images obtained as a result of bilinear filtering and 4x anisotropic filtering performed by Volari (on the left) and GeForce FX (on the right):
The comparison of these images produced by Volari shows that the difference exists only in those places where MIP-level borders have been moved. In case of GeForce FX the difference can be seen in the entire scene.
What was that? Unfinished drivers or “XGI’s vision of anisotropic filtering”? I would prefer to believe in the first, rather than the second. But is we remember how big the performance drop is when we enable tri-linear filtering, then we can actually imagine how dramatic the situation will turn if we enable “true” anisotropic filtering. Therefore, the only possible conclusion will sound as follows: Volari doesn’t know to perform “true” anisotropic filtering at all, or performs it with such awful performance losses that the software developers simply had to disable this option in the drivers.
Unfortunately, just like in case of S3 DeltaChrome, XGI Volari Duo V8 Ultra also had some start-up problems in a few contemporary games.
In particular, Tomb Raider: Angel of Darkness refused to start with old XGI Reactor v1.00.000 drivers, and when we installed the newer v1.01.510 driver it did start but froze every now and then. One of the next-generation games we had at our disposal refused to start with any driver version we installed, and in a few other games we could notice some unpleasant visual artifacts. Among them are Tron 2.0 and Command & Conquer Generals: Zero Hour:
In case of Zero Hour the problem is connected with the water surface display. It probably has something to do with the fact that the current driver version doesn’t support some shaders. As for Tron 2.0, the picture here is not so illustrative, although the reference image and the image produced by Volari do differ greatly. In the first place the differences are noticeable in the neon lighting and ghosts around many objects in this game.
There is one more issue, which has nothing to do with games, but which nevertheless should be mentioned. During the playback of some movies we watched a black screen with the sound from the movie. The problem could be solved by disabling overlay in the player settings, and it worked for Winamp 5.01, for BSPlayer and for Windows Media Player 9. The problem is likely to be originating from the incorrectly working Cipher Video Processor technology, and should be soon solved in the new driver versions. However, the fact that we discovered it now doesn’t add to the developer’s score.
Unfortunately, the rumors circulating around the web and claiming that XGI decided to go in NVIDIA’s footsteps by optimizing the drivers, appeared absolutely true. xrvkp.sys file included into the first version of XGI Reactor contained the following:
3DMARK03.EXE; 3DMARK2001SE.EXE; 3DMARK2001.EXE; AQUAMARK.EXE; C4DEMO.EXE; GMMARK2.EXE; MAXPAYNE.EXE; BENCHMARK.EXE; UT2003.EXE; BEND3DIM.EXE; MADDEN04.EXE; HALO.EXE; 3DNASYS.EXE; CODECREATURES BENCHMARK PRO.EXE; CODECREATURES; SERIOUSSAM.EXE; SPLINTERCELL.EXE; C4.EXE; TRAOD.EXE; TRAOD_P3.EXE; TRAOD_P4.EXE; GIANTS.EXE; SEP.EXE; CLIENT.EXE; DFBHD.EXE; TOEE.EXE; MAJESTICCHESS.EXE; AOWSM.EXE; MGS2; SR2.EXE; F1_2001.exe; SPEEDDEMO.EXE; BF1942.EXE; X2-DEMO.EXE; XIII.EXE; FFXIWINBENCH.EXE; SIMCITY4.exe; GAME.DAT; WAR3.EXE; SM.EXE; FREELANCER.EXE; NR2003DEMO.EXE; FS9.EXE; DCT2.EXE; HCT.EXE.
More than suspicious, don’t you think so? The new XGI Reactor v1.01.51 doesn’t have these lines any more, but the concerns we had inspired us to check if the cheats had really been removed completely. To find this out we ran all gaming tests twice. First we ran the tests with standard names of the exe-files, and then we renamed the start-up exe-files and repeated the test cycle again.
Therefore all graphs and diagrams for the performance of our testing participants in gaming applications contain two positions for Club3D Volari Duo V8 Ultra: the first one illustrates the regular result, while the second one shows what was obtained after we renamed the files of the tested application. If the second position is empty, then it means that the performance and image quality in both cases was the same. When we discovered any optimizations, we took a couple of screenshots to make sure that the performance gain demonstrated by Volari caused noticeably image quality worsening. Namely, that they had to sacrifice tri-linear filtering and enable bilinear instead, just like in the first XGI Reactor driver version. By the way, they replaced the older driver with a new version very quickly, almost immediately after the cheats had been discovered.
The list of gaming applications we used remained the same, except for the games that refused to run on Volari based graphics card.
First person 3D shooters:
Third person 3D shooters:
As we have already said, we didn’t use one next-generation game this time, because the system froze too often when running it. We also removed Tomb Raider: Angel of Darkness from the list of our gaming apps. All games were run with maximum graphics quality settings. Since FSAA doesn’t work yet in the XGI Reactor drivers, we ran all tests in the “Pure” mode only.
The performance of Club3D Volari Duo V8 Ultra in the previous generation OpenGL game leaves much to be desired to put it mildly. The software developers still have to work quite hard on the OpenGL driver optimization.
We didn’t find any specific optimizations here, however, the overall performance of Club3D Volari Duo V8 Ultra turned out pretty low. It was much lower than that of RADEON 9600 XT, which has been considered an outsider in this benchmark.
We get the impression that OpenGL is a stumbling stone for all manufacturers trying to conquer the graphics market. As you remember, S3 Graphics also failed to develop a good driver for this API, although its performance is a little bit better than that of the XGI’s one. Anyway, since we have already seen the reaction of XGI Volari on enabled tri-linear filtering, we can also explain its failure in this OpenGL test by inefficient texture caching algorithms.
In Unreal Tournament 2003 the newcomer manages to perform a little bit better than in OpenGL games: it runs almost neck and neck with other mainstream graphics solutions. When we used the XGI Reactor v1.00.000 drivers, Club3D Volari Duo V8 Ultra performed much faster, but it was only possible at the expense of image quality, which got notably worse. In the new drivers, XGI corrected its mistake and now the picture looks quite OK. However, the performance dropped down to the speed of the today’s mainstream solutions.
Despite the fact that the image quality improved, it is still far from ideal. Take a look yourselves:
The problems with texture level of detail and scene lighting are undeniable. The latter is especially evident in Antalus screenshots. This way, we clearly see that even with the new drivers the image quality leaves much to be desired.
And here is the first disclosure: HALO game started running much slower on Volari after we renamed the halo.exe file. What allowed the card to run as fast as GeForce FX 5600 Ultra? Take a look at the screenshots below and you will get the point:
The image quality on the first screenshot is simply awful: the textures are greatly washed out, and this blurring starts almost at the foot of the player. Moreover, the scaling effect can be noticed on the entire image: look at the polygon edges and at the target cross. On the left you see enlarged fragments of the “optimized” image. On the right – the image without “optimizations”.
In the screenshots with “optimizations” you can clearly see alternating strings: one string, then two same strings, then one string again and then two same strings again… This results from picture scaling along the vertical axis, when the picture was enlarged by 1.5 times. This way, it is evident that XGI did more than simple texturing optimizations. They even resorted to such a measure as building the scene in lower resolution first and enlarging it then to normal size. Having analyzed the resulting images we discovered that in HALO the resolution is reduced only along the vertical axis and only by 1.5 times. But it means that theoretically Volari can build an image in any “convenient” conditions and then enlarge it to normal dimensions. The visual evidence of this approach being use is dramatically broken polygon borders.
Please, welcome anti-Anti-Aliasing from XGI!
On the screenshots obtained without any optimizations on Volari’s part, everything looks OK, but the performance is simply unacceptable for comfortable gaming. Even to make these screenshots I had to make a lot of efforts, because the image was moving jerkily.
One more game, this time it is Splinter Cell, and another evidence proving unfair policies pursued by XGI, trying to increase the performance of its Volari solution by right or wrong. They didn’t even care about the image quality here. Look:
As you see, when we renamed the starting exe-file the image quality changed a lot. In particular, the lighting and textures got much clearer, the missing shadows returned to their places. However, the performance dropped down to unacceptable level.
Tron 2.0 game uses 1280x960 resolution, which is not supported in the current XGI Reactor driver version. That is why we are offering you only the results for 1024x768 and 1600x1200:
There isn’t anything Volari Duo V8 Ultra could boast here. Besides the visual issues already mentioned above, the performance again turned out a total disaster.
IL-2 Sturmovik: Forgotten Battles also uses this nonstandard resolution: 1280x960.
As you see, there is hardly anything to comment on. Again a failure. I don’t think that flight simulator fans will ever decide on a solution from XGI.
Another flight simulator and another defeat. The newcomer is almost twice as slow as the former outsiders: GeForce FX 5600 Ultra and RADEON 9600 XT in the lowest resolution. In the higher resolutions it is even slower than that. Luckily, no optimizations have been noticed here, otherwise…
Space simulator fans also have nothing to hope for with XGI Volari: the performance of the top model, Volari Duo V8 Ultra, is extremely low in this one of the most beautiful games of the kind. Even in the lowest resolution you will not be able to play X2 – The Treat on Club3D Volari Duo V8 Ultra.
Command & Conquer Generals: Zero Hour is a real time strategy, which uses 1280x960 resolution, considered non-standard by XGI Reactor driver.
Do you want me to comment on that? Why is the performance so low? Is it because of inefficient Volari architecture or raw drivers? I hope that the problem lies with the drivers, although the low-level fillrate and texturing tests proved the other way.
In Final Fantasy XI Official Benchmark 2 Club 3D Volari Duo V8 Ultra looks pretty fine, although it is not the best here. It performs as fast as RADEON 9600 XT. It really surprised me a lot, but the image quality didn’t arouse any reproach on our part.
In Aquamark3 the graphics card from Club3D performs not bad at all and gets close to RADEON 9800 PRO and GeForce FX 5900. Nevertheless, the first XGI Reactor driver version was optimized exactly for Aquamark3 test, besides others. The new version of the driver is free from any cheats: at least we didn’t see any performance changes after the aquamark.exe file had been renamed. For a more illustrative comparison the diagrams above also contain the results shown by Volari Duo with the old XGI Reactor driver and renamed aquamark.exe file.
Now let’s check the image quality in this test:
The problems with the texture level of detail and lighting are evident here. Unfortunately, XGI’s attempts to help its Volari run as fast as the competitors were either vain or required significant image quality losses.
At first have a look at the general performance picture in 3DMark03:
The situation is very close to what we have just seen in Aquamark3: our today’s hero performs quite fast, being in between the mainstream and high-end solutions. However, the gap between XGI Volari Duo and the mainstream competitors is not that big at all. In this case we decided to check once again if this level of performance is still achieved at the expense of image quality, despite the fact that we couldn’t prove any optimizations with all certainty.
To our disappointment, we once again saw that XGI cheats with textures quality. We revealed some problems with lighting and texture level of detail. Now let’s take a closer look at 3DMark03 results:
Nothing extraordinary: Club3D Volari Duo V8 is not bad at all here, but only with all the hidden optimizations enabled. But when we disable them, the newcomer slows down to unbelievably low rates.
In the second gaming test the situation is just the same as in the first one, despite the overall lower values. Volari Duo V8 Ultra can be regarded as competitive in terms of performance, only if we forget about acceptable image quality. And the image quality gets really bad in this case.
In the third gaming test the situation is just the same as in the first two.
The fourth gaming test brings another devastating failure. On the one hand, the results shown by Volari Duo V8 Ultra are not that bad, but on the other hand, the price we pay for this performance is unacceptable.
This is very sad news, but having tested Club3D Volari Duo V8 Ultra we arrived at the conclusion that this product hardly has any chance in the today’s market.
Club3D did everything they could to make the new product as attractive as possible for the end-user. And we should admit that they really did a great job on that. The card looks very impressive and aristocratic. However, it is not only the exterior that you usually value about the graphics card. It is not for nothing that you prefer performance and good image quality to the nice looks and PCB design. Unfortunately, XGI can’t boast much here.
The current XGI Reactor driver version ensures the performance level comparable to that of the mainstream solutions from ATI Technologies and NVIDIA Corporation, however, it sacrifices image quality for the sake of extra fps. Note that we revealed not only texturing quality problems, but also problems with the lighting, problems resulting from the attempts to speed the solution up by rendering in lower resolutions and then enlarging the image to the necessary dimensions, problems with “fake” anisotropic filtering.
XGI didn’t have any pains of conscience when they introduced “driver optimizations” for specific games and test applications. Moreover, it was done so crudely, that it immediately caught our eye. We could not only see the file names in one of the files included with the XGI Reactor drivers, but also the image quality got so much worse that it simply couldn’t remain unnoticed.
In the last version of XGI Reactor, there was no evidence such as file names any more, but the performance was sometimes still improved by reducing the image quality. In other words, the tactics remained the same.
Of course, the XGI baby can also work without these “optimizations” providing normal image quality. But as soon as we disable the “optimizations”, the performance of Volari Duo V8 Ultra gets a few times lower, it drops beyond any possible minimum. Moreover, we are talking about the most powerful product model here. I am already terrified at the mere thought of how low the performance of Volari Duo V5 or single-processor Volari models could be in reality.
What causes such low performance or a seemingly promising graphics processor? We assume that it is primarily because of raw and unfinished drivers. However, it would be unfair to blame only drivers for this failure.
The major reason of this low performance has probably a lot to do with the Volari’s inefficient architecture, starting from the elementary caching algorithms and finishing with complex pixel shader 2.0 processing. The fillrate problems also contributed to the product’s low speed. And the reason for such poor fillrate comes from the very low levels, which is perfectly proven by the driver “optimizations”. As you remember, XGI Reactor drivers recognize familiar applications and reduce the texture level of detail, which automatically reduces the caches workload, because there are less texturing data to be processed. The performance grows up. This definitely indicates that Volari does have problems with texturing speed.
Yes, formally the card features two memory buses 128bit each. However, in this case 128x2 doesn’t make 256, because the graphics processors of the dual-chip solution are connected with one another via a very narrow channel, which bandwidth is as high as that of the AGP 8x. Using fast and expensive GDDR-II memory doesn’t make much sense, although it is quite logical, as speeding up external memory could theoretically help solve the problem with the internal cache. However, the practice showed that GDDR-II, which failed to save NV30, doesn’t save XGI Volari Duo today.
So, no wonder happened. Volari Duo V8 Ultra didn’t become a worthy competitor to RADEON 9800 PRO/XT and GeForce FX 5950 Ultra, although XGI had promised that. This “hot” and expensive dual-processor monster is most likely to follow in the footsteps of ATI Rage Fury MAXX and 3dfx Voodoo5 5500. In other words, it is most likely to become a museum rarity interesting for us from the theoretical point of view, nothing else. Although unlike the latter solutions, it will not be able to enjoy the laurels of a leader even for one single day. Well, another attempt to introduce dual-processor solutions in the 3D worlds suffered a total fiasco.
It is very sad that XGI didn’t lean the lesson of SiS Xabre and again resorted to the so-called “turbo-texturing”, although it was a little bit modified this time. Moreover, they kept it secret hoping that no one would notice the image quality worsening. No doubt that the company, which used to be known as Xabre Graphics Inc, will hardly be able to win the customers’ and potential partners’ hearts this way. Secrecy is not the best thing to start with for a company, which is going to compete seriously with such graphics giants as ATI Technologies and NVIDIA Corporation. Moreover, “turbo-texturing” implemented in SiS Xabre has already demonstrated this back in 2002.
So, what can XGI actually offer us today?
What can’t XGI offer us today?
As a result, we see that XGI developers didn’t address the matter seriously enough. The attempt to present the solution as if it were much better than it actually is, is the worst thing that could be done, because the users, as well as analysts are usually very suspicious towards the manufacturers caught cheating. In the end, I do not understand why they decided to integrate DirectX 9.0 feature into a solution, which cannot do texturing, lacks memory bandwidth saving technologies, etc.
If XGI doesn’t improve the situation with the image quality and performance of their Volari in the near future, we will have every reason to state that their debut in the consumer 3D graphics market turned out a failure. We will carefully follow the events here, to make sure that you always know the latest news.
Although we are not so positive about the situation with XGI Company, we have to stress one very important thing, though. When we were working on this article, the top model from Volari family, Volari Duo V8 Ultra, cost $499. In fact, this is much higher than most of the today’s graphics accelerators actually cost. Take for instance, RADEON 9800 XT or GeForce FX 5950 Ultra, which sell for less, and at the same time offer much better performance and image quality.
In conclusion, we have to say a few words about the retail solution based on the notorious XGI Volari Duo V8 Ultra chip: Club3D Volari Duo V8 Ultra.
As you can notice, the list of drawbacks is considerably longer than the list of advantages, which will hardly help Club3D Volari Duo V8 Ultra to become a successful product.