Home SiS’ Xabre600 GPU
Reviews

SiS’ Xabre600 GPU

Geoff Gasior
Disclosure
Disclosure
In our content, we occasionally include affiliate links. Should you click on these links, we may earn a commission, though this incurs no additional cost to you. Your use of this website signifies your acceptance of our terms and conditions as well as our privacy policy.

MUCH HAS BEEN MADE of NVIDIA’s coming GeForce FX processor and the challenges the company has faced producing the new chip on a 0.13-micron manufacturing process. But just because NVIDIA won’t have its 0.13-micron GeForce FX out until January or February of next year doesn’t mean that the 0.13-micron process is somehow unattainable for graphics chips today. In fact, I’ve got a 0.13-micron graphics core running right next to me. It’s SiS’ Xabre600, a DirectX 8.1-compliant GPU that seems to have no problem chugging along on a 0.13-micron core.

SiS might seem like an unlikely candidate to have the first 0.13-micron desktop GPU, but they were also the first to support AGP 8X with the Xabre’s initial launch, so they’re no stranger to pushing the envelope. Unfortunately, the Xabre’s initial launch was plagued by poor performance, application incompatibilities, and questionable driver shenanigans that largely overshadowed SiS’ unique graphics offering.

Now, months after the initial Xabre400 release, SiS has added the faster Xabre600 to its arsenal of graphics products. This new Xabre comes with a revamped set of new drivers, too. Can a shrunken GPU and new drivers make the Xabre600 a credible, compatible, and perhaps even desirable value offering? Read on to find out.

The chip
SiS’ Xabre600 is essentially just a faster version of the Xabre400 graphics chip that we reviewed back in August. Nothing has changed in the rendering pipeline or overall chip architecture. This Xabre is just built on a 0.13-micron process rather than the 0.15-micron manufacturing processes used for the Xabre400 and other low-end Xabre chips.

SiS don’t need no stinkin’ paste Apparently, SiS’ engineers are so confident in the 0.13-micron core’s ability to run cool that they didn’t even bother putting any thermal interface material between the Xabre600’s GPU and heat sink. That SiS was able to move the Xabre600 to a 0.13-micron process before graphics heavyweights like NVIDIA and ATI is impressive, but don’t expect this chip to take on high-end offerings like the GeForce FX. The Xabre600 is positioned against more mid-range offerings like the Radeon 9000 Pro, which is why you won’t find it running at 500MHz with a Dustbuster strapped on its back.

It’s been a few months since we last looked at an Xabre GPU, so I’ll give you a quick refresh of the chip’s unconventional architecture.

  • A 4×2 rendering pipeline — Each of the Xabre’s four rendering pipelines is capable of laying down two textures per pass, which gives the card a competitive fill rate, at least on paper. As far as I know, the Xabre isn’t capable of “looping back” within the rendering pipeline to lay down additional textures per pass. That shouldn’t be a problem in today’s games, which generally don’t lay down more than a couple of textures per pass. However, the Xabre could take a significant performance and image quality hit in future titles that require more textures per pass than the Xabre can handle without having to write out to the frame buffer.
  • Shaders — SiS has taken an interesting approach to shader support with the Xabre. The chip itself features version 1.3-compatible pixel shaders, but there’s no hardware vertex shader. Instead of performing vertex shader calculations on the GPU, the Xabre uses DirectX 8’s software vertex shader and performs the calculations on the CPU. In the end, SiS gets away with a simpler GPU core and a smaller die size, but still retains compatibility with DirectX 8 titles.

    NVIDIA’s GeForce4 MX also uses DirectX 8’s software vertex shader to offload vertex shader calculations to the CPU, but that trick doesn’t work for pixel shaders. Because DirectX 8.1 can’t emulate pixel shaders in software, the GeForce4 MX isn’t a complete DirectX 8.1-compatible part.

    Because of SiS’ unconventional approach to shaders, application compatibility has been an issue. Applications may correctly detect the Xabre’s pixel shaders, but be unable to recognize or utilize DirectX 8’s software vertex shader, which could degrade performance and eliminate some in-game eye candy. The Xabre400’s initial drivers had problems with some 3D applications not correctly recognizing the Xabre’s shader support, but hopefully the Xabre600’s newer driver revision does better in that department.

  • AGP 8X — SiS started the AGP 8X party when it first announced the Xabre, and since then AGP 8X has become a must have checkbox feature for new graphics products. We’re still waiting for applications to actually take advantage of the extra bandwidth made available by AGP 8X, though.
 

The card
The Xabre600 comes on an eye-catching black PCB, complete with a polished gold heat sink. The leather jacket/gold medallion look is new for SiS, but we’ll see just how tough this card is a little later on.

Pimp, baby Thankfully, the Xabre400’s whiny, annoying fan hasn’t carried over to the Xabre600. The Xabre600’s GPU fan is no louder than what you’d expect to find on cards sporting ATI and NVIDIA chips, and it’s certainly a lot quieter than Abit’s funky new OTES.

Given the 0.13-micron manufacturing process, and the lack of any thermal interface material between the GPU and heat sink, I have to wonder how well the Xabre600 would do with a good passive cooling solution. GPU fans aren’t loud, but it would be interesting to see if you could use a passive heat sink with some good heat sink paste and not have to sacrifice any clock speed, stability, or image quality with the Xabre600.

Where’s the RAM? It’s been a while since I last saw a graphics card with memory on only a single side of the PCB, but the Xabre600 manages to keep all the chips on one side with this 64MB card.

Memory chips apparently rated at 2.8ns SiS has gone with new BGA memory chips on the Xabre600, which should help the card hit higher memory clock speeds. Apparently, the chips are rated to 2.8ns, though I can’t find an official part number that exactly matches the chips.

The 301 companion chip Like the Xabre400, the Xabre600’s multi-monitor and TV output is handled by SiS’ own 301 video chip. The Xabre600 actually works just fine without the 301 companion chip, you just don’t have access to DVI and video outputs, which some low-end cards and integrated motherboards may not even need.

Everything you’d expect for output ports The Xabre600 features all the usual suspects on the port back plane. There are VGA, S-Video, and DVI ports to play with, and since this is a mid-range part, I really can’t justify ragging on SiS for there being only one DVI output. Since all I’ve tested is a reference card, I can’t comment on whether third party board manufacturers will be including DVI to VGA adapters with their Xabre600 cards. Given the card’s rather sparse multimonitor features, I wouldn’t expect it.

 

New drivers
SiS has revamped its Xabre drivers, and at least on the surface, fixed a lot of problems with previous versions. Those initial drivers required a registry hack to enable anything other than a low texture quality setting for 3D applications. This was a particular sore point in our initial Xabre400 review, but these new drivers let you adjust DirectX and OpenGL texture quality with an easy slider right in the driver control panel. That’s really the way it should be; having to resort to Regedit to unlock high quality textures is unacceptable.

In addition to adding a texture quality slider, SiS has also include an application that lets you automatically throttle the Xabre’s clock speeds when you’re not using 3D applications. It’s an interesting feature, but one that’s probably better suited for mobile or small form factor applications where power consumption and heat are more of a concern.

Our testing methods
Our test systems were configured like so:

Processor Pentium 4 2.26GHz
Front-side bus 533MHz (133MHz quad-pumped)
Motherboard SOYO SY-P4X400 DRAGON Ultra
North bridge VT8754
South bridge VT8235
Chipset drivers VIA 4-in-1 4.43
Memory size 512MB (2 DIMMs)
Memory type CAS 2.5 PC2700 DDR SDRAM
Graphics card SiS Xabre400
SiS Xabre600
NVIDIA GeForce 4 Ti 4200 64MB
NVIDIA GeForce4 MX 460
ATI Radeon 9000 Pro
Graphics driver Xminator II 3.07 Detonator 40.72 CATALYST 2.24, 2.23
Storage Maxtor DiamondMax Plus D740X 7200RPM ATA/133 hard drive
OS Microsoft Windows XP Professional
OS updates Service Pack 1

We don’t have any official pricing from SiS on the Xabre600, but according to SiS’ roadmaps, the card should be priced to compete with ATI’s Radeon 9000 Pro, and NVIDIA’s high-end GeForce MX and low-end GeForce4 Ti 4200. The Xabre600 is the only AGP 8X card of the lot, but that shouldn’t make a huge difference in the benchmark results.

You’ll notice that two sets of CATALYST drivers were used for the Radeon 9000 Pro. For some reason, the Codecreatures test just wouldn’t run with the latest 2.24 drivers, even after a fresh, driver-less re-imaging of the test system. Because of this, I tested the Radeon 9000 Pro with the 2.24 drivers for everything but Codecreatures, where I used the 2.23 drivers.

We used the following versions of our test applications:

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.

 

Fill rate
This next section presents an interesting challenge. You see, according to SiS’ press kit, the Xabre600 comes with core and memory clock speeds of 300MHz, respectively. The reference card I have is clocked at 315/315. It gets better. On a fresh OS, driver, and card installation, the clock speeds sit at 315/315. If you go into SiS’ overclocking utility and click on the “Default” clock speed button, the clock speeds drop to 310/310. Powerstrip, on the other hand, pins the default clock speeds at 315/315.

Confused yet? I am. I’ve contacted SiS about this, and haven’t heard anything back yet, so I’m not sure what final Xabre600s will be clocked it. Manufacturers may in fact be free to set Xabre600 clock speeds higher than 300MHz if they so choose, which might explain the discrepancies. Since the default installation clock speed of the Xabre600 is 315/315, and since the card is perfectly stable at those speeds, I’ve run through all the benchmarks in that configuration.

  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 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
SiS Xabre 600 300 4 1200 2 2400 600 128 9.6
SiS Xabre 600* 315 4 1260 2 2520 630 128 10.1

Of course, those are just paper fill rates. Theoretical peaks dictated by spec sheets aren’t always a good indicator of overall 3D performance, especially since graphics cards don’t always realize all of their theoretical fill rate in the real world.

The Xabre600’s actual single and multi texturing fill rates are far below its theoretical peaks, which suggests that the Xabre is largely unoptimized. Notice how in the multi texturing test, the Radeon 9000 Pro and GeForce4 cards turn in actual fill rates very close to their theoretical peaks. The Xabres don’t even realize half of their potential fill rate in single or multi texturing tests, which confirms just how important it is to look beyond paper specifications.

 

Occlusion detection
You have to get into something like NVIDIA’s GeForce FX or ATI’s Radeon 9700 Pro to see some really cool Early Z occlusion detection in action, but the Kyro tile-based rendering architecture proved that eliminating overdraw can be especially benefical for even low-end graphics cards with limited fill rates.

SiS doesn’t actually talk about occlusion detection at all in the Xabre600’s documentation, so I’m going to have to assume that not much is going on in that department. Let’s see what the VillageMark results look like.

The Xabre600’s higher scores may be more a result of its higher theoretical fill rate than any kind of occlusion detection scheme, especially since clock speed seems to have a big impact on the Xabre’s performance. You’ll notice that the GeForce4 Ti 4200 doesn’t register a score for our 640×480 resolution; it’s simply too fast. VillageMark doesn’t seem to accurately report frame rates above 200 frames per second.

 

Pixel shader performance
The Xabre’s big selling point, at least when compared to the GeForce4 MX, is its DirectX 8-class pixel shaders. The Xabre features version 1.3 pixel shaders, which can handle a maximum of 12 instructions (8 arithmetic, 4 texture address), but just how fast are they?

In 3DMark2001SE’s pixel shader tests, the results aren’t pretty. Though the Xabre600 is able to maintain more consistent frame rates than the competition, those frame rates are also consistently much slower than the competition.

Does the Xabre600 fare any better in NVIDIA’s ChameleonMark pixel shader tests?

Nope, not in any of the three ChameleonMark tests. The Xabre600 may have pixel shaders that are being correctly detected, but so far their performance just isn’t all that impressive.

It’s also worth noting that the relatively close performance between the Xabre400 and 600 suggests that a clock speed boost isn’t going to help matters much here.

 
Polygon throughput and vertex shader performance
3DMark2001’s vertex shader test results will be particularly interesting because we’re essentially looking at two different hardware vertex shader implementations, and two different software vertex shader implementations. Both Xabre cards use DirectX 8’s software vertex shader, and NVIDIA’s GeForce4 MX 460 also offloads vertex shader calculations to the CPU.

Clearly, NVIDIA and ATI’s GPUs perform far better than our Pentium 4 2.26GHz when it comes to vertex shader calculations. What’s interesting to note here is that the Xabres are slower than the GeForce4 MX 460, despite the fact that both are using the CPU to emulate a vertex shader.

Next up we have some DirectX 7-class transform and lighting tests, which may be more important than the vertex shader tests for determining performance in current games and applications that don’t really take advantage of DirectX 8-class hardware.

The Xabre’s results here are particularly interesting here because they differ so much from the norm. The Xabre must have one hell of a T&L unit, which is why the performance is so good at low resolutions. However, as the resolution scales up, it doesn’t look like the card’s actual fill rate can keep up, which kills performance.

AGP write performance
Serious Magic’s AGP texture download tests are in for the sake of completeness. Here we’re looking at how fast a graphics card can transfer textures to main memory from the graphics card. This isn’t something that’s especially important for gaming or most 3D applications, but it’s really startling to see just how little texture download bandwidth can sometimes be available, even with AGP 8X graphics cards. Please see our article on this subject if you want to know more.

Despite being an AGP 8X part, the Xabres are way behind NVIDIA’s offerings when it comes to texture download speed. NVIDIA used to be just as bad as the rest, but a driver update has really improved their performance in this area.

 

Games
Enough with the synthetic performance tests. Game performance is what we’re all really after, isn’t it?

Jedi Knight II

Jedi Knight II performance doesn’t seem overly limited by any of the graphics cards we’re using today. While it’s a fairly recent game, but it’s based on the Quake III engine, which is starting to show its age. The Xabre600’s DirectX 8 features really aren’t giving it any help here.

Comanche 4

Comanche 4’s engine is newer than Jedi Knight II, and it shows; frame rates are lower across the board. Initially, Comanche didn’t even detect the Xabre’s shader support, but all that has been fixed with the latest WHQL Xabre driver. The new drivers enable shader support in Comanche 4 for both the Xabre400 and 600.

Unfortunately, even with its shaders being detected properly, the Xabre600 is still quite a bit slower than its closest competitor in Comanche 4. In fact, the Xabre600 can’t even manage 30 frames per second at 640×480.

Codecreatures Benchmark Pro

Codecreatures requires pixel shaders, and didn’t used to run properly with the Xabre drivers. With new drivers, Codecreatures runs perfectly, but the performance isn’t all that impressive. Kudos to SiS for at least getting its compatibility issues worked out with new drivers.

 

Unreal Tournament 2003
Unreal Tournament is probably the most graphically demanding game on the market, and thanks to [H]ard|OCP’s handy benchmarking tool, testing the game is a snap. We’ve run through a set of resolutions for the DM-Antalus and DM-Inferno levels with the default low and high quality graphics settings to get an idea of how the Xabre600 performs with what will likely be the dominant first person shooter until Doom III hits.

With low quality graphics settings, the Xabre600 actually holds its own at high resolutions, but it looks like SiS could still stand to do some driver optimizing to improve performance at lower resolutions. Dropping the resolution usually results in much smoother gameplay, but that may not be the case with the Xabre600 if it’s being held back by drivers rather than fill rate.

With high detail graphics settings, the Xabre600 just keeps up in DM-Antalus, but falls further behind in the more demanding DM-Inferno test. Even in the high quality tests, the Xabre600 seems more competitive at higher resolutions. Unfortunately, at those high resolutions, the frame rates aren’t good for much more than walking around and gawking at the Unreal Tournament 2003’s gorgeous levels.

 

Serious Sam SE
Because we maximize the graphics quality settings on all of our other tests, I used the Extreme quality add-on for Serious Sam SE. This add-on sets all Serious Sam SE’s graphics options to their highest quality levels, limited of course by the capabilities of each card.

The performance of the Radeon 9000 Pro and GeForce4 Ti 4200 looks a little slow here, but that’s because both cards support varying levels of anisotropic filtering, which Serious Sam SE is taking full advantage of. The Xabre600 doesn’t support anisotropic filtering, which gives it an edge in the performance department, but even that’s not enough. Of course, since the Xabre600 doesn’t support anisotropic texture filtering, its textures will be quite a bit blurrier than the Radeon 9000 Pro or GeForce4 Ti 4200.

Serious Sam SE also lets us graph frame rates over the entire length of the benchmark demo, which will give us a good idea what contributes to the above average frame rates. Let’s take a look.

Over the length of the benchmark demo, the Xabre600 doesn’t exhibit any unusual spikes or dips in performance relative to the other cards.

 

3DMark2001 SE

SiS will likely tout the Xabre600’s performance in 3DMark2001 SE, since the card does relatively well overall when compared with the GeForce4 MX 460 and Radeon 9000 Pro. How does the card perform in 3DMark’s simulated game tests?

In Game 1’s car chase scene, the Xabre600 is pretty competitive with the GeForce4 MX 460 and Radeon 9000 Pro throughout the resolutions we tested.

In Game 2’s Dragothic scene, the GeForce4 Ti 4200 steals the show, but the Xabre600 holds its own and turns in another competitive performance.

In the Matrix-inspired Lobby scene, the Xabre600 manages to stay one step behind the GeForce4 MX 460 and Radeon 9000 Pro.

3DMark2001 SE’s Nature test requires pixel shaders, so the GeForce4 MX 460 can’t play this time around. Nature does, however, highlight the Xabre600’s weak pixel shader performance when compared with the Radeon 9000 Pro and GeForce 4 Ti 4200.

 

Workstation-class applications

The Xabre600 runs at the back of the pack in all but a couple of SPEC’s viewperf workstation 3D tests, which we’ve included for the sake of completeness. Suffice to say that you’re not going to see any graphics workstations using Xabres in the near future.

 

Antialiasing
The Xabre600 doesn’t do anisotropic filtering, but it will do 2X and 4X antialiasing. Of course, the question is, what impact do these AA modes have on performance?

Surprisingly, the Xabre600 holds its own in our antialiasing tests. We are using 3DMark2001 SE’s low detail Game 1 test, which probably has something to do with the Xabre600’s better overall performance.

 

Overclocking
Our sample Xabre600’s lack of thermal interface material between the GPU and head sink doesn’t exactly scream out for overclocking, but I thought it would be interesting to see just how far I could push the Xabre600’s 0.13-micron core with the card’s limited cooling. In testing, I was able to get the core clock speed stable at 340MHz, and the memory stable at 330MHz (660MHz DDR). Not bad for not even having thermal paste.

Unfortunately, the clock speed boost doesn’t do much for performance in 3DMark2001SE, and it probably won’t help many other applications until the Xabre’s driver bottlenecks are cleared somewhat. Still, the results do bode well for running the card at stock speeds with only a passive heat sink.

Conclusions
The Xabre600’s pixel shaders give it an obvious edge over the Geforce4 MX in a feature category that will only become more important as time goes on. Sure the GeForce4 MX 460 is faster now, but it may not support all the new eye candy in future DirectX 8.1 titles.

Against the Radeon 9000 Pro, the Xabre600 starts to look a lot worse. Here the DirectX compatibility playing field is level, but the Radeon 9000 Pro’s pixel shader performance is much better, as is its performance in real world applications. Even if the Xabre600 is able to achieve price parity with the Radeon 9000 Pro, ATI’s value offering is still going to be a better deal.

Let’s not even get into how the Xabre600 compares with the GeForce4 Ti 4200, because it really doesn’t. The GeForce4 Ti 4200 is likely to be the most expensive of the Xabre600’s closest competitors, anyway.

The fact that the Xabre600’s performance can’t keep up with the competition doesn’t mean that there isn’t value to the part. That SiS is able to produce the chip on a 0.13-micron process is impressive in itself, and I’m happy to see that the new drivers have fixed all the compatibility problems. With the improved compatibility of the latest drivers, SiS at least has a DirectX 8-class graphics chip with the Xabre600, even if it’s not the most competitive one. 

Latest News

What is Disney+
Streaming News & Events

Reliance and Disney Agree On $8.5 Billion Merger to Create an India Entertainment Powerhouse

Statistics

Romance Scam Statistics – Common Types & How To Stay Safe

Online dating is the most common way to meet people, which unfortunately paves the way for scams and frauds, as the latest romance scam statistics show. The devastating figures and...

Meta Accused Of Running A Massive Data Processing Business
News

Meta Accused Of Running A Massive Data Processing Business; Violating the GDPR

Meta has been accused of running a data processing business in the facade of its privacy paywall; both of which violate European privacy laws. Consumer activist groups for 8 countries...

ChatGPT Sued Over Unauthorized Use Of A Newspaper Articles
News

ChatGPT Sued Over Unauthorized Use Of A Newspaper Articles

companies
Statistics

27 Landscaping Industry Statistics (2024 Added Data)

Top Crypto Gainers on 29 February - BONK and DOGE
Crypto News

Top Crypto Gainers on 29 February – BONK and DOGE

Dogecoin (DOGE) Price To Mirror 2021 Parabolic Move and Surge 10X - But When?
Crypto News

Dogecoin (DOGE) Price To Mirror 2021 Parabolic Move and Surge 10X – But When?