S3 Tries to Get Back Into PC Graphics

mikemuch writes "ExtremeTech has a review of S3's attempt to get some traction in the lower-end graphics card market, the Chrome S27. Though its specs look great--256MB memory, 700MHz core clock rate, 1.4GHz memory clock, and 22.4 GB/sec memory throughput, it still manages to underperform similarly priced video cards from the red and green graphics companies."

Anyone else nostalgic for the S3/Virge?

[Gothmolly]

I remember the day when my PC was finally faster than the processor on the Virge, but boy, Descent looked kickass in special 'S3' mode. Of course that was also 1996.

Go S3!

New name, same old story.

[Anonymous Coward]

This is what, the third or fourth time S3 has tried to 'make a comeback' in the graphics market? It always seems to happen the same way. Big price features in a small price card! While the hardware might be ok, and compete on the low end.. The drivers never seem to catch up. This is what the big players have. An established codebase for drivers that gets tweaked each time a new chip comes out.

Predictions for new S3-super-thinamagig:

1. Early previews from hardware review sites- "Shows promise!" "Should compete at entry level" "Good for casual gamers." Drivers will be buggy.

2. Card released many months after initial previews. What was mid grade is now low end, and card doesn't look so hot against current competitors. Drivers still buggy. S3 promises bug fixes and performance improvements.

3. Several off brand Taiwanese manufactures will make cards featureing new S3 chip. Cards will quickly be relegated to bargain basement prices in retail and online shops. Mobile versions of chips will be found in cheap low-end laptops and versions of the core will be seen integrated in to via chipsets for cheap onboard video. Drivers still buggy.

4. S3 continues product line and no longer updates drivers. (Drivers still buggy.)

With any luck S3 will do better than their previous attempts, but they've got a lot to prove. In all likelihood, this will go the way of the S3 savage, S3 chrome, trident cyberblade, XGI volari, powervr2, and powervr KYRO.

See the fanboys review!

[jmorris42]

First they compare a $115 card to cards costing $125 and $129. Then the price drops to $99 and they 'stand by their review' against those more capable boards because they didn't pan it for performance, but for basic flaws? Uh huh. That would be because SLI mode doesn't work? What sort of idiot would buy a $99 card for SLI work? Ok, AA doesn't appear to work for GL, that is bad but will almost certainly get fixed in the drivers pretty soon.

It looks like S3 is trying something interesting, throw high speed but dumb hardware at the problem of 3D instead of trying to put more compute power than a P4 on a board. But they are going to discover that the drivers are a big part of the equation, it was clear that their drivers probably what was holding their scores down on several of the tests. Since they obviously don't have a lot invested in them yet perhaps they are the ones we should be pushing to support open source. Despite what that PR moron at Nvidia said I suspect the Open Source crowd could whip those drivers into shape in short order, Use the right license (MPL or BSD) and they could roll those improvements back to Windows and carry the fight to ATI and Nvidia.

I know I'd certainly switch from ATI Radeon 9250 (most current 3D with Open drivers) to this new S3 tech if it had an open driver.

This is why graphics drivers are closed-source

[RzUpAnmsCwrds]

This is exactly why NVIDIA and ATI keep their drivers closed-source.

If you look at S3's product, you see a device that has great hardware specs (looks great on paper) but fails to deliver because of buggy/incomplete drivers. S3 isn't alone in this - XGI faces similar problems.

The truth is that a lot of the performance of modern GPUs comes not from the hardware but from the drivers which supply it with data. NVIDIA and ATI keep their drivers closed-source because they don't want a company like S3 to benefit from their software - NV and ATI love the fact that everyone else has buggy, slow, incomplete 3D drivers, and that's the way they want to keep it.

Re:Good

[non0score]

What you're saying is actually a generalizing statement. It's similar to saying that because (so far) the Conroe is better than the K8 (this is the general overview), the integrated memory controller on the K8 is worthless and Intel shouldn't copy it (then you specify that the specifics aren't worthwhile because of the general overview argument, and this is flawed). There's a good portion of the DeltaChrome system that performs better than its ATi or nVidia counterparts. And revealing the driver source will reveal how they designed these parts of the system. If S3 Graphics did open source their drivers, they'd lose every bit of advantage they can use to "level" the playing field.

Open source: good for some things, not for others

[typical]

I'm no longer entirely convinced that open source has a huge benefit for video card drivers.

Okay, yes, it's probably not a bad idea, but I read an article from some guy at Precision Insight about this.

The argument was that video card drivers are among the most complex of drivers. It's kind of like Mozilla -- there is a significant barrier to just dropping in and hacking away -- so you don't accrue programmers left and right.

Granted, I'd still buy any card that has good open-source driver support and is more modern than my Radeon 9250, but I'm part of a limited set of people.

Open source works very well for a number of things, but there are definitely systems that it's less effective at:

* Systems that require a significant understanding of a large deal of code before commits can be made. People get scared away from Open Office and Mozilla. Emacs has done very well, but even though it's a sizeable codebase, it's also extremely modular on a by-feature basis. You don't need to learn much to be able to write a new useful feature or understand how a particular feature works.

Anything that requires non-general knowledge has a significant cost in an open-source package. Custom code that replaces standard code (string classes, replacement data structure code), knowledge of the software package's threading model, strange conventions, you name it. Having a codebase that is easy to drop into and start working is always nice, but it is especially important for open source projects.

* Systems that cannot be used by the developer. Okay, this probably isn't an issue for graphics card drivers. Really, really esoteric stuff or things that few programmers have interest in (like simplified interfaces for existing software) don't tend to attract many volunteer developers.

Games with limited replay value are another example. Successful open-source games, almost without exception, have very high replayability and make relatively little use of elements like plot twists, eye candy, or anything else where the enjoyment only comes through once or a few times. Such games tend to either be multiplayer or involve a significant random element. If games can't be played over and over by the developers, the developer doesn't get much enjoyment out of playing them and drifts away.