SiS’s Xabre 400 graphics processor
There’s more to SiS’s Xabre than support for DirectX 8.1 and AGP 8X, and only diving into the registry can reveal the cunning conspiracy. Years ago, one of SiS’s on board graphics chipsets was caught rendering only every other frame in ZD’s then-popular 3D Winbench performance test. It was cheating, and a mark was made on SiS’s permanent record. Now, with Xabre, SiS is again using questionable tactics to deliver better frame rates in 3D applications.
Is SiS cheating again? What does the performance picture look like when we force a level playing field? Find out the answers to these questions and more as we take an in-depth look at SiS’s Xabre 400 graphics chip.
SiS hasn’t introduced a new GPU since its 315 chip, so the Xabre is a pretty big deal. Xabre is currently available in three flavors, the Xabre 400, 200, and 80, but we’ll be concentrating on the Xabre 400 for the purposes of this review. As you might expect, the Xabre 200 and 80 run at lower clock speeds. The Xabre 80 also has only a 64-bit DDR bus and is only AGP 4X-compliant.
Unlike ATI and NVIDIA, who outsource their chip production to specialists like TSMC and UMC, SiS fabs the Xabre chips itself.
Here are some of the Xabre’s key features:
- A 4×2 rendering pipeline SiS has given Xabre four pixel pipelines, each capable of laying down two textures per pass. As far as I’m aware, there’s no functionality that allows Xabre to “loop back” and lay down multiple textures without a second rendering pass. For complex shader calculations, Xabre will have to write to the frame buffer, read back the results, and possibly lose some color precision in the process.
Today’s games don’t lay down that many textures per pass, but future titles using more complex shader programs will. Then again, this is a budget graphics part, and the only announced graphics card with any kind of really hard-core internal precision is ATI’s high-end Radeon 9700 Pro.
- Shaders Xabre features version 1.3 pixel shaders, which have a maximum instruction length of 12 instructions (8 arithmetic, 4 texture address). Pixel shaders are at the heart of DirectX 8’s advances, because they enable more complex pixel processing (and thus tastier eye candy) in 3D applications.
What’s really interesting is that Xabre has no vertex shaders. DirectX doesn’t have any facility for emulating pixel shaders (they could be emulated via multi-pass rendering, but DX8 doesn’t break down pixel shader programs into multiple passes, probably for performance reasons). However, DX8 does have a software vertex shader implementation, so cards without vertex shaders can lean on the CPU for vertex shader handling. SiS kept Xabre’s transistor count low by leaving out vertex shaders. Clever, no? This is one of the benefits of using a clean-slate design for a budget chip instead of recycling an older graphics core, like NVIDIA did with the GeForce4 MX.
There is a hitch, however. The lack of a vertex shader makes the Xabre look a lot more like the GeForce4 MX than a DirectX 8.1-compliant graphics card, and it could run into some problems with some DX8 titles. Applications will have to be intelligent enough to recognize that the Xabre has pixel shaders but not vertex shaders. If applications don’t get that right, they may either turn off shader-specific features or fail to run at all.
SiS would have been wise to include a software vertex shader in its driver software, so Xabre could expose both pixel and vertex shader capabilities to applications. A driver-level vertex shader would also present opportunities for tweaking, extensive SIMD support, and the like. Instead, SiS chose to rely on Microsoft’s DX8 vertex shader.
- Bandwidth conservation Xabre uses “Frictionless Memory Control” to conserve memory bandwidth, but details on just what makes the system frictionless and what they’re doing to conserve bandwidth are scarce. The most we were able to squeeze out of SiS’s engineers is that FMC is a 2-channel architecture. The Xabre also attempts to conserve bandwidth by using Z-compression, fast Z-clear, and their own proprietary hardware occlusion culling algorithm.
Like every other graphics chip in its class, Xabre has a 128-bit memory bus, which means there should be a decent amount of memory bandwidth with DDR SDRAM and reasonable memory clock speeds.
- AGP 8X SiS is talking a lot of smack about AGP 8X with the Xabre, and to its credit, Xabre was the first AGP 8X graphics card available. However, with a shortage of AGP 8X motherboards on the market, the fact that Xabre was first is a bit of a moot point.
As you may have already guessed, AGP 8X delivers twice the bandwidth of AGP 4X, which puts it right around 2.1GB/s. Greater AGP bandwidth will theoretically let Xabre pull more data over the AGP bus, but what kind of impact this will have on performance is unclear. To be faster than AGP 4X, an application needs to be saturating the AGP bus with over 1GB/s of data, and I’m not sure applications are quite there yet.
- Multi-monitor and video capabilities Xabre features an integrated 375MHz RAMDAC and MPEG decoder, which means cards should support a good range of resolutions and refresh rates, plus DVD playback. There’s also a TV encoder on the chip, but Xabre needs the companion SiS 301 chip to output signals to a TV. The 301 chip also powers a DVI output or second VGA monitor if you want to use multiple monitors.
SiS has programmed in some per-pixel motion detection de-interlacing that kicks in when you use the integrated MPEG decoder. This feature should improve the quality of incoming video streams.
Now that we have an idea what Xabre is all about, let’s take a look at how it stacks up against its competition, theoretically speaking. Fill rates and memory bandwidth don’t guarantee real-world performance, but they will give us a good launching point for our benchmarks.
|Core clock (MHz)||Pixel pipelines||Peak fill rate (Mpixels/s)||Texture units per pixel pipeline||Peak fill rate (Mtexels/s)||Memory clock (MHz)||Memory bus width (bits)||Peak memory bandwidth (GB/s)|
|GeForce4 MX 440||270||2||540||2||1080||400||128||6.4|
|GeForce4 Ti 4200 128MB||250||4||1000||2||2000||444||128||7.1|
|GeForce4 Ti 4200 64MB||250||4||1000||2||2000||500||128||8.0|
|SiS Xabre 400||250||4||1000||2||2000||500||128||8.0|
|Radeon 9000 Pro||275||4||1100||1||1100||550||128||8.8|
|GeForce4 MX 460||300||2||600||2||1200||550||128||8.8|
|Radeon 8500 128MB||275||4||1100||2||2200||550||128||8.8|
With four pixel pipelines laying down two textures per pipe, the Xabre 400 lines up perfectly with ATI’s Radeon 8500LE and GeForce4 Ti 4200s in terms of single and multi-texturing fill rate. With 8GB/s of memory bandwidth, things remain competitive across the board. These are just theoretical peaks, so don’t get too excited just yet.
Now let’s take a look at the reference card.
I’m not a huge fan of purple, but I can live with just about any color on my GPU heat sink as long as the fan isn’t overly annoying. Unfortunately, the fan on our reference card’s heat sink was beyond annoying. Whenever you fire up a 3D application, the temperature-sensitive fan kicks into high gear and starts whining louder than a Morpheus user complaining about high CD prices. Really, it’s that loud.
So the fan is loud, but doesn’t the large size at least cool the memory?
Nope. As you can see, there are no heat sinks on the card’s rear memory chips. Normally this would mean that there’s an imbalance between the memory cooling on the front and rear memory chips, but that’s actually not the case here.
It doesn’t really matter that there are no heat sinks on the rear RAM chips because the heat sink doesn’t actually make physical contact with memory chips on the front of the card. The gap between the heat sink and memory chips isn’t a result of a sloppy heat sink installation; the top of the GPU and memory chips simply don’t line up in a manner that lets the flat base of the heat sink make contact with both.
Our Xabre reference card features all the usual ports. Given the low price targets of Xabre-based cards, I wouldn’t expect most manufacturers to include the DVI-to-VGA adapter necessary to run two VGA monitors.
The Xabre 400’s memory spec only calls for a 250MHz memory bus, but the 64MB worth of chips on our reference card were rated at 300MHz. That should be one heck of an easy overclock, though I doubt third party cards will be so generous with their memory chips.
Driver and platform quirks
SiS’s drivers for the Xabre are about as lean as drivers come these days. There are no tuning options for any 3D features. Multimonitor control is especially weak, and there’s not even access to vsync. In a world where NVIDIA and ATI keep raising the bar for driver software, SiS’s effort has much to be desired.
The driver CD for our review sample came with a copy of SiS’s 3D Wizard software. Although this utility isn’t available for download from SiS, I would expect it to ship with cards based on the Xabre chipset. 3D Wizard gives you control over antialiasing, 3D stereo glasses, and overclocking, but its support of controversial wireframe and transparency modes is sure to ruffle a few feathers.
In the past, wireframe and transparency modes have been justified by manufacturers as enabling users to learn more about 3D rendering. However, SiS includes access to these features under the 3D Wizard’s “Eagle-Eye” tab, which doesn’t sound all that educational to me.
Normally we like to make sure graphics cards we’re testing are displaying textures with the highest possible quality setting because high scores in benchmarks mean nothing if rendering quality is poor overall. Since SiS’s drivers don’t include any texture sliders, we had to hack our way through the registry to access a setting X-bit Labs found in their Xabre review called TexTurbo. On our review system, the SiS.3D.TexTurbo key was found in the following registry location:
In a nutshell, TexTurbo controls the quality of textures, which in turn affects frame rates and image quality. SiS’s latest WHQL drivers default to a TexTurbo setting of 3, and we’ve tested that. We’ve also run through a complete line of tests with TexTurbo set to 0, and we’ll be comparing the two along the way in both performance and image quality to see how they differ.
Having to use the registry to change a driver setting is worrisome, but we’ll have to go through some benchmarks and image quality tests before passing judgment. TexTurbo could be a big deal or a minor inconvenience.
Stability was a big problem with the Xabre on Abit’s AT7 motherboard, which uses VIA’s KT333 chipset. The card was stable on Abit’s KT266A-based KR7A-RAID, but I’m hearing there are some issues with Xabre-based cards on VIA and nForce platforms in general. If problems plague both VIA and NVIDIA chipsets, chances are pretty good that the blame lies with the SiS rather than with the chipsets themselves.
Our testing methods
As ever, we did our best to deliver clean benchmark numbers. All tests were run three times, and the results were averaged.
Intel Pentium 4 2.26GHz
Intel 82845E (MCH)
Intel 82801DB (ICH4)
512MB (2 DIMMs)
CAS 2.5 PC2700 DDR SDRAM
GeForce4 Ti 4200 128MB
|Graphics driver||SiS 3.03||
IBM 60GXP 40GB 7200RPM ATA/100 hard drive
Windows XP Professional
Today we’re comparing the Xabre 400 with the most recent crop of budget video cards from ATI and NVIDIA. Some of these cards are price-competitive with products based on the Xabre 400, while others will show just what kind of performance gains to expect if you’re willing to spend a little more.
The Abit BD7II-RAID motherboard we used for testing doesn’t support AGP 8X, which may put the Xabre at a bit of a disadvantage. However, considering how few applications saturate an AGP 4X bus, I think we’ll be alright.
We used the following versions of our test applications:
- MadOnion 3DMark2001 SE
- VillageMark 1.17
- Quake III Arena 1.30
- Jedi Knight II: Jedi Outcast
- Serious Sam SE
- Comanche 4 demo benchmark
- SPECviewperf 7.0
The test systems’ Windows desktop was set at 1024×768 in 32-bit color at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Most of the 3D gaming tests used the high detail image quality settings in 32-bit color.
All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
In 3DMark2001 SE, the Xabre 400 doesn’t perform well at all; there’s also quite a performance hit when TexTurbo is set at 0. Though the Xabre 400 claims to be a fully DirectX 8.1-compliant product, it’s the slowest DirectX 8.1 part in 3DMark2001 SE. In fact, with TexTurbo set to 0, the Xabre 400 is slower than the GeForce MX cards and the Radeon 7500, neither of which implement the full DirectX 8.1 feature set.
It’s interesting to note that MadOnion had to issue a patch for 3DMark2001 SE to make it work correctly with the Xabre. Totally new GPUs like Matrox’s Parhelia haven’t required patching, so SiS must be doing something radically different from other graphics companies. My guess would be the “software” vertex shader, which is certainly a nonstandard way of fulfilling that particular DirectX 8.1 requirement.
3DMark2001 SE – Game benchmarks
In 3DMark2001 SE’s first game scene, the Xabre is only slower than the Radeon 8500 and GeForce4 Ti 4200 cards. With TexTurbo at 0, things get a little more dire, and performance drops considerably.
The Xabre 400 rides Dragothic’s dragon all the way to last place among DirectX 8.1-compatible cards in the second game test, and it’s the slowest of all the cards we’ve gathered with TexTurbo 0.
In the Lobby game scene, the GeForce4 MX 460 just squeaks by the Xabre 400, which continues to scrape the bottom of the barrel when TexTurbo is set to 0.
3DMark2001 SE’s Nature scene requires vertex and pixel shaders, so the GeForce4 MXs and Radeon 7500 are out of the picture. With them gone, the Xabre 400 brings up the rear with a performance that’s well off its competition. The Xabre 400 may be fully DirectX 8.1-compatible, but it doesn’t perform that well when DirectX 8.1’s advanced features are used.
3DMark2001 SE – Fill rate
Earlier we looked at the theoretical fill rates of each of the cards in our stable today, and now it’s time to see how much of those theoretical peaks each card is able to realize in 3DMark2001 SE’s fill rate tests.
The Xabre 400’s single texturing fill rate is poor, and grossly inefficient regardless of which TexTurbo value is used.
With multiple textures, things get better for the Xabre with TexTurbo set to 3. Setting TexTurbo to 0, however, cripples multi-texturing performance and we realize less than 25% of the Xabre 400’s theoretical multi-texturing fill rate.
3DMark2001 SE – Transform and lighting
SiS claims the Xabre has a third-generation T&L unit, and its performance isn’t too bad.
In a more complex scene with eight lights, the Xabre 400 moves up right behind both GeForce4 Ti 4200s and Radeon 8500s. Note here that TexTurbo has no impact on T&L performance; it’s just a texture thing.
3DMark2001 SE – Bump mapping
The Xabre 400 performs admirably in the DOT3 bump mapping test, that is until we set TexTurbo to 0. Notice the huge variation between the Xabre 400’s two scoresone almost doubles the other.
Only the GeForce4 MXs can’t complete the environmental bump mapping test. The Xabre’s performance here is, well, pretty horrible. Even a Radeon 7500 is more than twice as fast.
3DMark2001 SE – Advanced features
In 3DMark2001 SE’s point sprite test, the Xabre 400 only manages to edge out the Radeon 7500, which barely produces a score.
The results of the vertex shader test are particularly interesting because both the Xabre and GeForce4 MX cards are emulating a vertex shader in software and doing all the work on the CPU. Xabre runs a little bit slower than a GeForce4 MX and quite a bit slower than the cards that have true hardware vertex units.
3DMark2001 SE – Pixel shaders
The Xabre 400 has pixel shaders, but their performance is over three times slower than the pixel shader-enabled competition.
I would have liked to include Codecreatures scores because it’s a nifty benchmark that really stresses vertex and pixel shaders. Unfortunately, though the Xabre 400’s shaders are detected and the card generates a score, the rendered output doesn’t look anything like what it should. Entire sections of the benchmark scene are blank; if Codecreatures were a game, it would be completely unplayable.
3DMark2001 SE had to be patched to work with the Xabre, and maybe that’s what Codecreatures needs to render things properly. Software shouldn’t require an update just to run on a new graphics card, though. I’m not sure whether to blame the software developers for hard coding elements to certain graphics chips or SiS for doing something wrong with the Xabre.
Villagemark shows a huge disparity between the performances of the two TexTurbo modes we’re testing. With TexTurbo set at 3, the Xabre 400 is the fastest card of the lot. Setting TexTurbo to 0, however, drops the Xabre 400 to the bottom of the pile.
Let’s take a more condensed look at how the Xabre 400’s more direct competition fares over several different resolutions.
Here we can clearly see the extreme disparity between the performance of TexTurbo modes. The Xabre 400 is either solidly above or below its direct competition depending on which TexTurbo setting you use.
Quake III Arena
First off, you’ll notice that the Xabre 400 with TexTurbo 3 scores a zero for 640×480 in Quake III Arena. No amount of coaxing could get the timedemo to complete without crashing the system at that resolution.
In resolutions above 640×480, the Xabre 400 with TexTurbo 3 consistently finds itself behind the GeForce4 Ti 4200 cards. With TexTurbo set to 0, it drops to the bottom and effectively ties the Radeon 7500 for the lowest score at each resolution.
At TexTurbo 0, the Xabre 400 follows the Radeon 7500’s performance quite closely through all the resolutions. The GeForce4 MX 460 provides the best competition for the Xabre 400 with a TexTurbo setting of 3, but the Xabre 400 has a slight edge.
Jedi Knight II showcases the disparity in TexTurbo scores in a Star Wars setting, but this time the Xabre 400 with TexTurbo set to 3 is able to beat out even the GeForce4 Ti 4200 cards. As you might have expected, setting TexTurbo to 0 throttles the Xabre 400’s performance and drops it near the bottom. Available video memory really matters here. Notice how the performance of the GeForce4 Ti 4200 64MB and 128MB cards diverge at high resolutions. A Xabre 400 with 128MB of DDR SDRAM should fare much better at high resolutions in Jedi Knight II.
Though the Xabre 400’s performance with TexTurbo set to 0 starts really dropping off after 800×600, setting TexTurbo to 3 lets us maintain more consistent frame rates until the highest resolution. We are using Jedi Knight II’s “Very High” texture quality setting here, which accounts for low scores at high resolutions. Once we’re done with the performance benchmarks, we’ll get into just what kind of image quality price we’re paying for these higher frame rates with TexTurbo set to 3.
Serious Sam SE
Even a TexTurbo setting of 3 can’t help the Xabre 400 in Serious Sam SE, and it brings up the rear at almost every resolution. As we’ve seen throughout testing so far, a TexTurbo setting of 0 is much slower, especially as the resolution increases.
It looks like the Xabre 400 is limited by driver throughput at Serious Sam SE’s 640×480 resolution, because both TexTurbo setting lines have similar slopes to the competition after 800×600.
With Serious Sam SE we get one more graph than usual: a plot of frame rates over time during a benchmark demo. This plot will show us where during the benchmark demo the cards excel, choke, and just how erratic the frame rates are.
The same pattern we’ve seen so far rears its head again, and the Xabre 400 with TexTurbo set to 0 peaks and dips lower than everything else. With TexTurbo set to 3, the Xabre 400 manages to keep itself relatively competitive, but really only against low end GeForce4 MX 440 and Radeon 7500 cards at this resolution.
Serious Sam SE had some unusually jerky console performance in high resolutions on the Xabre 400. You have to drop into the console after each demo run to record average frame rates, and as I ran through each resolution several times, I noticed the console rendering getting slower and slower with each resolution increase. This problem was only on the Xabre, and I didn’t observe any slowdown with any of the other cards.
Console rendering isn’t a huge deal, but applications that use the same kind of transparency seen in Serious Sam SE’s console window could run into problems.
Comanche 4 didn’t like the Xabre much, and it didn’t recognize any shader support. This could be Novalogic’s problem, something that will just require a small patch like 3DMark2001 SE, but it illustrates how application support for the Xabre’s unique approach to shaders may be spotty.
The Xabre 400 is reduced to the same graphics settings as the GeForce4 MX in Comanche 4, and its performance is disappointing. Not even setting TexTurbo to 3 makes things respectable, and frame rates really slide as the resolution scales up.
The Xabre 400’s performance with TexTurbo set to 0 drops off right away, while the other cards have more gradual performance declines as the resolution increases. Setting TexTurbo to 3 keeps the Xabre 400 from falling too far behind the value competition, but it’s still getting beat.
Something that the benchmark results don’t show, but that’s worth mentioning anyway, is a lot of choppiness in Comanche 4 when the benchmark fades in and out of different scenes. Only the Xabre 400 seemed to slow down during these transitions, when one scene fades into the next, and where I would assume transparency is being used extensively.
SPECviewperf’s results vary a little, but the Xabre 400 is at or close to the bottom of the pile in each test; this is clearly not a card you want to use for 3D workstation applications. TexTurbo settings have no impact on the Xabre 400’s performance in SPECviewperf, further confirming that texture quality is the only thing being manipulated with that switch.
The antialiasing graphs are pretty big, but they show the same kind of performance we’ve seen from Xabre throughout our benchmarks. Antialiasing on the Xabre 400 is faster when TexTurbo is set to 3, just like everything else, but it’s still not enough to vault the card to the top of the pile. With TexTurbo set to 0, performance drops dramatically and the Xabre 400 really only competes with the Radeon 7500a card that’s far too old to be running with a new DirectX 8.1-compliant AGP 8X graphics card.
All the price-competitive value graphics cards show relatively similar drops in performance going from no antialiasing to 2X and 4X AA at 1024×768. The whole point of antialiasing is how it makes the scene look, so let’s take a look at some screen shots.
ATI’s SMOOTHVISION continues to produce the best looking antialiasing, followed closely by the GeForce4 MX and finally SiS’s Xabre. 4X AA doesn’t look too bad on the Xabre 400, but the 2X AA sample shows a lot more jagged edges on the lamp than either ATI or NVIDIA. There doesn’t appear to be much difference between the Xabre 400’s no AA and 2X AA samples at all, and the very subtle reduction in jagged edges certainly doesn’t justify 2X AA’s performance hit.
Image quality is up next, and we’ve got a bunch of uncompressed image files to illustrate a few of the Xabre 400’s texture quality shortcomings. If you’re on broadband, you’ll be fine, but modem users will have to wait quite a while for everything to download. You can skip past this next section if you’d like. We would have offered up compressed JPEGs, but compressing files designed to illustrate image quality sort of defeats the purpose.
This is where we uncover the price for all that extra performance having TexTurbo set to 3 was able to give us in our performance benchmarks. Pay special attention to Max’s hand and necklace as we take a look at some screenshots from 3DMark2001 SE.
There you have it: the image quality price that comes along with a TexTurbo setting of 3. Max’s hand and necklace clearly illustrate that the Xabre 400 is using lower quality textures. Some users may be hard pressed to pinpoint these lower quality textures in a game setting, but things feel noticeably ‘off’ during gameplay.
TexTurbo doesn’t only affect the Xabre 400’s texture quality, it also limits texture filtering as a little Quake III Arena r_colormipmaps fiddling shows.
The Xabre 400 can’t do trilinear filtering with TexTurbo set to 3, and that results in some ugly transitions between mip map levels. Again, you may not be able to notice these rough transitions during a heated deathmatch, but the image quality is noticeably poorer than on other cards, even when you’re frantically running around trying to evade rockets.
Quake III Arena’s sky doesn’t look too hot either, no matter which TexTurbo setting you use with the Xabre 400.
The Xabre 400’s clouds are a lot rougher than what you’ll find on the other cards (and the Radeon 9000 Pro in this specific instance).
The Xabre 400’s performance isn’t impressive, but maybe a little overclocking will help matters. In testing we were able to get the Xabre 400’s core clock speed all the way up to 295MHz without any artifacting or instability. Curiously, the memory chips rated to 300MHz topped out at only 260MHz, and the card crashed consistently with memory clocked any higher.
An almost 20% core clock speed boost helps the Xabre put in a better performance in 3DMark2001 SE, but it’s still the slowest DirectX 8.1-compliant card that we tested.
In Quake III Arena, there’s a small performance gain, but it’s not much. After seeing what setting TexTurbo to 3 does to image quality, I think it’s pretty safe to say that those scores aren’t all that important to consider unless you really don’t care about image quality.
So is SiS cheating?
Testing shows that SiS sacrifices image quality to gain higher performance in 3D applications with its TexTurbo setting. A TexTurbo setting of 3 yields higher frame rates at the expense of texture quality. But is that wrong? The answer isn’t quite as obvious as the differences in TexTurbo image quality and performance.
Many graphics drivers, including those from ATI, have a texture quality slider that lets you specify a range of values between “Quality” and “Performance” settings. It’s no secret that downgrading texture quality can increase performance, and some gamers don’t mind if things look horrible. If TexTurbo is really no different than ATI’s texture quality slider when it comes to its impact on image quality and performance, where’s the beef?
Here: The TexTurbo switch is buried in the registry, and its default setting is 3.
The fact that you have to wade through the registry to change the texture quality is bad enough, but since SiS has used a TexTurbo setting of 3 as the default, you’re stuck with poor image quality until you change the registry setting. Now hacking the registry isn’t a big deal for most enthusiasts, but the Xabre 400 is a low-end card that will find its way into the value market. Many users aren’t as comfortable with modifying the registry, and they shouldn’t have to.
I can’t justify calling TexTurbo cheating, but how SiS have implemented this texture quality setting is sneaky and abhorrent, and they have no plans to let users directly access this setting. Letting the users adjust texture quality easily from the driver control panel is all SiS needs to do to correct the problem, since doing so will give users a choice between texture quality and performance. As far as I’m concerned, having to modify the registry doesn’t constitute choice.
I have to admire SiS’s novel approach to shaders. They save transistors by emulating a vertex shader in software, but they include hardware pixel shaders to achieve DirectX 8.1 compatibility. It’s an interesting way to do things, but it looks like a lot of software will need to be patched before the Xabre 400’s shaders are universally detected. Until then, you’re left with the performance picture we’ve painted today, and it ain’t pretty.
The Xabre 400 occasionally performs very well with TexTurbo set to 3, but if you’re at all concerned with image quality, you’d do well to just ignore those results. Performance with TexTurbo at 0 is uglyas ugly as texture quality with TexTurbo set to 3and the Xabre 400 is consistently the slowest DirectX 8.1-compatible card. Against ATI’s Radeon 9000 Pro, the Xabre 400’s closest competitor in terms of price and features, the Xabre is outclassed in every department.
Maybe the Xabre 400 would have performed a little bit better if we’d tested with an AGP 8X platform, but even if it did, any praise would be conditional. SiS is playing games with texture qualitydon’t let them pull a fast one on you. TexTurbo needs to be in the driver control panel, or it needs to default to 0. Anything less is deceptive and unacceptable.