The SiS 315 graphics processor

WHEN CONVERTING FROM from one language to another, something invariably gets lost in the translation. Looking at the name “Evil Sam” emblazoned across the box holding SiS’s latest graphics chip, I have to wonder why the marketing folks didn’t catch this one. The more I look at the name, the more it could actually be “Evil 5am.” Not that that makes sense either. Thankfully, the name of SiS’ reference card isn’t really of much concern. Instead, it’s SiS’ new 315 graphics chipset that’s the object of our attention.

SiS touts its 315 as having a “high performance 256-bit 3D Graphic engine” with “Impressive 2D/3D Performance.” Let’s be realistic, though. When was the last time you saw a video card advertised as having an “adequate” graphics engine with “mediocre” performance? In a world with GeForce3 Titanium 500 cards, it seems unlikely that absolute performance will make the SiS 315 an impressive offering. Still, when available for a scant $50, even decent performance can turn a few heads in the value segment.

The goods
The first thing that caught my eye on the 315’s spec sheet is the fact it has its own transform and Lighting (T&L) engine. Chips from market leaders NVIDIA and ATI have T&L engines, but other graphics companies, like STMicro and Matrox, haven’t incorporated T&L yet. The fact SiS was able to incorporate T&L support is impressive—at least on paper. The 315 supports relatively advanced 3D effects like environmental bump mapping, as well.

Among the 315’s other notable features is the ability to support a second monitor at a different refresh rate, color depth, and resolution than the primary display. This used to be a realm occupied solely by Matrox, but it seems that SiS has figured out the tricks necessary to get this feature working with the 315. Our test card, however, was the edition with a TV-out port instead of a second VGA port, so we weren’t able to test SiS’s multi-monitor support in action.

Beyond that, the 315 is standard fare for a budget video solution. Our reference card came with DVI and S-Video connectors, and 32MB of plain old SDRAM. The 315, which supports up to 128MB of DDR SDRAM, is capable of better performance than we’ll be able to show you today. However, I have to think that 128MB of RAM is quite beyond overkill for a card of this calibre.

The 315 chip is built on a 0.15 micron process, so it needs little more than a passive heat sink to keep things cool at its core clock speed of 166MHz. Despite a rather weak thermal compound application, our 315 didn’t have any stability or heat problems. I suppose I should be happy that the heat sink wasn’t simply glued on, like it is on some cards.

Though the box and spec sheet for the 315 claim the chip has antialaising support, it’s nowhere to be found in the latest WinXP drivers from SIS. However, antialiasing on low-end cards is fairly useless anyway. Without the bandwidth and fill rate to render scenes in higher resolutions, value cards are generally reduced to a crawl with even 2X antialiasing.

Theoretical performance—by the specs
Before we get into the graphs, let’s take a look at some charts to see what exactly the 315’s theoretical performance bottlenecks are. Up first: fill rate.

Like the GeForce2 MX, the 315 has two pixel pipelines and can apply two textures per pixel. At least theoretically, only its slower 166MHz clock speed keeps the 315 behind the GF2 MX in fill rate. ATI’s low-end Radeon VE and STMicro’s Kyro both lag in the fill rate department.

Theoretical fill rate isn’t everything, though. Memory bandwidth is far more important when it comes to real-world fill rate—and thus performance.

The Radeon VE is the only card with DDR memory in our comparison, but it’s hampered by a 64-bit bus width. All in all, the theoretical memory bandwidth limits are very similar here.

This is all just theoretical, though. The real test of a card lies with benchmarks rather than comparisons of architectural limitations.

Our testing methods
Our test system was built using the following components:

With Windows XP officially out the door, it’s time to put Windows 2000 to pasture as a test platform. As good as Windows 2000 is, the benefits of XP can’t be denied.

Before you start moaning about the lack of an Athlon test platform in this review, pause. The 315 is likely going to see a lot of action on the OEM front, and that means it’ll invariably be paired with low-end Intel systems on the Intel’s Brookdale SDR platform. (Plus, I only have one Athlon test rig, and it’s doing something more exciting at the moment, the results of which will be graphed for you shortly.)

We used the following versions of our test applications:

• MadOnion 3DMark 2000 Ver 1.1
• MadOnion 3DMark 2001 Build 200
• Vulpine GLMark
• Quake III: Team Arena
• Wolfenstein MP Test Demo
• Serious Sam v1.02
• Max Payne v1.02
• SPECviewperf 6.1.2
• MDK2 Demo

The test system’s Windows desktop was set at 1024×768 in 32-bit colour at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Unless otherwise indicated, all tests were performed with all the visual options turned to their highest levels.

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.

3DMark 2000
The release of 3DMark 2001 seems to have killed off the use of its predecessor, 3DMark 2000. However, since we’re dealing with budget cards which tend to choke on 3DMark 2001’s DirectX 8 routines, let’s kick it old school for nostalgia’s sake. Well, and for the sake of DirectX 7, which plenty of games still use.

While the 315 can hang with ATI’s low-end Radeon VE, there’s simply no catching the MX in this benchmark.

The Kyro locks up here—a sign of things to come. The Kyro card suffered driver problems throughout our testing.

Depending on how you look at it, there’s either a striking penalty for running 32-bit color, or a hefty speed boost to be had by dropping to 16-bit color.

3DMark 2001
Now that we’ve seen these budget cards chew through an older benchmark, let’s watch them try to keep pace with something a little more challenging: DirectX 8.

The 315 again lags just behind the Radeon VE, as the MX continues to outclass its competition in the value sector. This is a straight GeForce2 MX we’re dealing with here, not a faster MX400 or a slower MX200. The MX200 is considerably slower than our our GF2 MX card, because the MX200 has only half the memory bandwidth. The Kyro completed this test properly, but it comes in last.

Moving to 3DMark 2001’s individual tests, we’ll see how the cards stack up in various environments. First up is the car chase demo.

All of the cards see a big drop in performance when we switch over to high detail. The 315 shadows the Radeon VE again. T&L makes a real difference, as the T&L-deprived Kyro takes a huge dive when we move to the high-detail scene. SiS deserves credit here not only for having a T&L unit, but having one that actually works.

Game 2’s dragon ride is next.

The Radeon VE jumps ahead in this test; in the high-detail test, it even bests the GF2 MX. Our 315 hangs in, as the Kyro’s performance makes it obvious T&L can have a very big impact on performance.

Game 3’s Max Payne shootout scene is next.

The Kyro sprints ahead at low detail, while the 315 lags appreciably behind the field. The 315 manages third place in high detail, though. The Kyro’s frame rate plummets with more polygons to draw.

Synthetic fill rate tests are next up.

With their similar specs, the 315 and GF2 MX perform very similarly in our first synthetic fill rate test. This result actually speaks rather well for the 315, since the GF2 MX leads it in core clock speed by 9MHz. The Radeon’s single pixel pipe hurts it here; it turns in a less-than-impressive performance.

So what happens with multi-texturing?

Both the 315 and the GF2 MX can lay down two textures per pixel, so again their scores are similar here. Because it can lay down three textures per pixel in a single pass, the Radeon VE rejoins the pack. Still, the Radeon VE falls well short of the surprising 315.

Now for a T&L-oriented test.

With only one light source in the scene, the MX trounces the competition. Though the SiS card is able to beat the Radeon VE by a sliver, the 315 doesn’t even come close to competing with the MX’s T&L engine here. Without T&L, the Kyro doesn’t have a chance.

Next, we turn on a few more lights.

Eight lights make the scene much more complex. The MX’s massive lead evaporates, and all the T&L engines struggle with this one.

Now let’s look at some bump maps.

Though the 315 supports environmental bump mapping, the only type of bump mapping supported by all the cards is dot-product bump mapping. All the cards do relatively well here, with the 315 and Radeon VE again switching places. Next, we move to vertex shaders.

As if on cue, the Radeon VE leapfrogs ahead of the 315, and takes up position just behind the MX. The Kyro’s outclassed here. Our 315 starts to look more and more like the little video card that could.

The final 3DMark 2001 test measures performance with particles, also called point sprites.

The graph doesn’t tell the entire story here. While the 315 jumps into the lead, the actual image displayed shows far too many missed sprites. This issue is being addressed by MadOnion, and they’ve acknowledged that the problem isn’t related to SiS’s drivers. Whether the 315’s performance will drop with MadOnion’s fix remains to be seen.

Vulpine GLMark
Now that we’ve spent some time testing Direct3D performance, we’ll look at Vulpine GLMark to see how well SiS has done on the OpenGL front.

Vulpine’s results are a little odd, as there’s not much of a drop in performance from 16 to 32-bit color; the bottleneck in this case is elsewhere. The GF2 MX card wins this one in a walk. ATI’s Radeon is only able to take a slim lead over the 315, as the competition between the two heats up.

Quake III: Team Arena
In our ongoing quest to keep it real, we’ll fire up Team Arena for our first true game benchmark. We start off with Team Arena’s “Fastest” setting, which is about as ugly as you can get; it compromises eye candy for performance.

On the “Fastest” setting, three of the cards offer acceptable performance. The fourth wouldn’t complete the test.

While the MX makes a habit of winning, the 315 is locked in with the Kyro in a battle for second place. It seems the Kyro drivers agree with anything above 640×480, and the card performs quite well. The Radeon VE slips behind the surprising 315. To stress things a little more, let’s turn up the resolution to 1024×768 and activate all the visual goodies.

The GeForce2 MX wins again, as the Kyro and SiS cards battle it out for second. The Radeon VE brings up the rear.

Wolfenstein MP Demo
Anand did such a nice job with his Wolfenstein demos that I’m going to use them here again. The first, atdemo6, punishes the video card with a wide open beach landing that’s a cross between Saving Private Ryan and Hamburger Hill.

SiS’ prospects with the 315 start looking better and better as it squeaks out a win over the Radeon VE in our most recent of benchmarks. Still, the 315 is far from catching the MX, and the Kyro will wake up once we turn up the resolution.

At a higher resolution, the 315 begins to show more weakness. None of the cards really pumps out a playable frame rate here, but the 315 comes in last with a measly 10 fps. Next we move to atdemo8, an indoor scene free of the massive polygon count that comes along with the beach landing.

With the resolution lowered, the 315 can again squeak past the Radeon VE. The MX remains well ahead, though. Will the 315 recede as the resolution is turned up again?

Yes. The 315 just doesn’t have the fill rate to deal with higher resolutions; it drops to last place. In a rare victory, the Kyro emerges to prove that tile-based rendering works especially well in fill rate-limited situations.

Serious Sam
Serious Sam’s benchmark results are up next, and they look a little funny, because they’re plotted over time. Being able to see a chip’s performance over time lets us compare performance peaks and valleys, so we can see whether cards are able to sustain a playable frame rate throughout the demo. The thing to watch out for here is the valleys, which can really kill playability.

Serious Sam shows us that Evil Sam could very well be a Radeon VE-killer. The two cards’ performance is very similar, even on our second-by-second plot. Still, both are outclassed by the MX, whose lowest valley is 5 fps faster than their highest peak. Sadly, in low resolutions, the Kyro’s AWOL again. Let’s see if the 315 loses ground again as we move to 32-bit color, where fill rate matters more.

The similarities in performance between the Radeon VE and SiS 315 really are quite striking, as the two shadowbox once again. The GeForce2 MX is still well ahead, though.

With the resolution higher and the Kyro in the mix, things get crowded. The 315 shows a little more weakness here as its frame rate oscillates more wildly than the others. Consistency isn’t the 315’s forte at 1024×768, and it falls to last in performance. Will turning to 32-bit color change things?

Increasing the color depth makes the Kyro more of a contender. Meanwhile, the 315’s frame rate again bounces all over the map. The more games we test, the more the 315’s fill rate limitations show.

Max Payne
Max Payne brings us back to Direct3D territory. This test gives the cards a workout some of the game’s more demanding cut scenes. We’ll start things off at Max Payne’s medium detail settings with bilinear filtering.

In what comes as a bit of a surprise, the 315 tops all challengers. In both 16 and 32-bit color, the 315 beats the competition with a solid margin of victory. Time to turn up the resolution.

The 315 falters at higher resolutions again, but this time it’s only suffering in 32-bit color. Otherwise, it pulls ahead of the pack, which now includes the Kyro. Can the 315 keep it up with trilinear filtering at the high detail setting?

It most certainly can, at least in lower resolutions. Turning up the detail and filtering doesn’t slow the 315, at least not in relation to its competition. Let’s see if the 315’s streak continues to our last Max Payne test.

The 315 retains its lead, even in 32-bit color. SiS certainly has its DirectX drivers in order here.

MDK2
We’ll round out or gaming benchmarks with another OpenGL test, MDK2.

The 315’s MDK2 results are embarrassing at low resolutions; even the Radeon VE takes a big lead. Can SiS pull its socks up when we turn up the resolution?

Sort of, but the 315 still comes in last. Perhaps SiS spent most of its resources on Direct3D driver development.

SPECviewperf
We’ll move on to a workstation-class 3D benchmark with SPECviewperf. These budget graphics cards were never intended to run workstation-class applications, but it’s interesting to see how they fare, regardless.

As you might have guessed, the Kyro couldn’t complete the SPEC tests without locking up.

Starting things off with Awadvs, the MX takes its customary lead. The 315 jumps past the Radeon VE here, as the Kyro still can’t complete the SPEC tests without locking up.

For DRV, the tables turn on the 315, as the Radeon VE comes out ahead. Still, both lag behind the MX, which is a far better, erm, ultra-low-end workstation chip.

SPEC’s Light tests have the Radeon VE out in front of the 315 again.

It’s beginning to look like the 315’s Awadvs performance was a fluke, as SiS trails again with the MedMCAD test. The Radeon VE comes in well ahead of the 315 as it chases the MX.

Concluding the tests, the Radeon VE gets a surprise victory over the MX, as our 315 languishes well out of contention in last place. It’s not as if you’d be shopping to SiS for a 3D workstation card, though.

3D image quality
Let’s take a look at some Quake III screenshots to get an idea of the 315’s 3D image quality. First up are mip maps, where we can see if the 315 is up to the task with trilinear filtering. We’ve turned on Quake III’s color mip maps feature in order to see whether the 315 properly handles trilinear. When it’s done right, trilinear filtering produces smooth transitions (on the Z axis) between mip maps, which are shown here as bands of color.

Unfortunately, the 315 doesn’t manage trilinear filtering correctly. Oddly, the color bands on the gun show proper trilinear color gradients, but the rest of the scene doesn’t. The 315 chip appears capable of trilinear filtering, but it’s somehow not implemented right here.

Next, we look to the skies, and Q3’s notorious clouds. Let’s check out how the cards draw them. In the past, we’ve used this test to demonstrate how texture compression affects image quality. This time around, we’ve turned texture compression off. The results are enlightening.

Something’s going on here with the 315, and it’s not good. For whatever reason, the 315 isn’t displaying 32-bit textures properly. Texture compression is turned off. Quake III’s texture size slider is maxed out, and everything’s supposed to be in 32-bit color. But something’s very wrong with those clouds—and with all the other textures in the scene. It looks like the SiS drivers may be really only drawing 16-bit textures here, regardless of the fact that both Quake III and the desktop are running in 32-bit color. This is a big blow for the 315, because Quake III looks rather ugly when compared with the competition. I can only hope this is a driver issue that will soon be resolved.

Video signal quality
Unfortunately, assessing the 315’s 2D image quality isn’t as easy as snapping a few screenshots. To assess video signal quality and sharpness, I compared the 315 to its competitors from this review, and also to a GeForce3 and Matrox G550 (our review is coming up shortly). To my surprise, the 315’s video signal quality is quite good; it’s right up there with the GeForce3 and Radeon VE.

It’s really hard to tell the difference between these cards with my monitors and eyes, though. To me, the G550 takes the cake with its gorgeous text. However, the 315 is at least as good as any of the other competitors.

While I wasn’t able to test the 315’s DVD playback, video playback using all manner of formats was smooth. The Radeon VE, which is known for its DVD playback, didn’t have an apparent advantage over the 315 across multiple video formats.

Conclusions
Initially, I thought the 315 was going to be a bit of a dog when it came to performance. I’m pleasantly surprised that it held its own against more established competition. Instead of spending all of its time at the bottom of our pack of budget cards, the 315 took a few opportunities to flex its muscles and take the lead. As one of the select few graphics processors on the market with a really useful T&L engine, the 315 certainly bodes well for SiS’s future graphics parts, at least on the low-end.

Additionally, our reference sample came with only SDRAM; it’s possible upcoming cards could offer faster DDR SDRAM, which would give the 315 more memory bandwidth. More than 32 MB of memory might help, too, in higher resolutions and with larger textures.

In terms of performance, the 315 can’t keep pace with NVIDIA’s GeForce2 MX, though it might have more luck against the crippled MX200. The MX200 is more in the 315’s price range, anyway. A more appropriate competitor for the 315 is ATI’s Radeon VE, or Radeon 7000, as it’s now known. Both are priced around $50, and the performance and video signal quality between the two is quite similar. I have to give the edge to ATI here, as the 3D image quality is much better. Also, ATI has a proven track record with video products.

Still, the 315 is a solid product, and SiS’s future graphics efforts will be something to keep our eyes on. NVIDIA’s nForce is just around the corner, and ATI is working on its own integrated chipset. I wouldn’t be surprised to see SiS slide some 315 technology into an integrated chipset of its own. If they get a few bugs ironed out, the proposition of a chipset featuring the 315’s graphics core certainly is intriguing.

In the end, the 315 is a good first step for SiS. If SiS can improve on the 315, we might just have another viable player in the 3D game. Who said graphics would get boring after 3dfx’s demise?