S3 Graphics Comes out of Hiding with Chrome20

Steve from Hexus writes "S3 Graphics, having been quiet for a while, has today announced a new graphics solution, Chrome20, with which they intend to take some market share away from ATI and Nvidia. From the article: 'We were offered a chance for some hands on play with a mid-range Chrome20 series desktop board - the machine was loaded with over 40 top games. A quick run of Half Life2 , Far Cry , Halo and a couple of other titles demonstrated that S3G's new 90nm mainstream card was working without any visual problems and with very playable frame rates.'"

Comments

Re:The Obligatory Question

[LnxAddct]

Last I checked, Linux's desktop share was higher than Apple's which puts Linux over the 3% mark. Desktop numbers are highly biased againt linux simply because a) Most linux machines were previously windows, and b) Windows machines tend to be replaced more often, i.e. if i buy a windows computer today and another one in 2 years, both will be considered to be active and the nuber will be twice what it really is.
Regards,
Steve

Re:My Take

[adisakp]

Most people who understand CPU architecture will tell you x86 is an old inefficient design

Actually, x86 is a very inefficient instruction set. However, the efficiency of the instruction set has been sidestepped mostly by on-the-fly hardware translation to a more efficient instruction set, large virtual register sets, out-of-order execution, and speculative execution. Neither AMD nor Intel CPU's operate on the x86 instruction set internally. Both of them translate x86 instructions into micro-ops internally and execute those instead -- believe it or not, they're doing in hardware much of what Transmeta was doing in software. The Pentium 4 doesn't even have a true L1 cache for instructions but rather uses an "execution trace cache" which has pre-translated micro-ops.

Furthermore, it's a chicken-and-egg problem when it comes to CPUs. A lot of optimization for X86 occurs because of the vast amount of software (Windows, etc) that runs only on X86. This software is often less than efficient and the manufacturers (Intel and AMD) optimize for the software inefficiencies with things like branch prediction, dynamic fetching, out-of-order execution, etc. Unfortunately, the optimization units to deal with x86 inefficiencies end up costing nearly as many transistors as the units that actually do the work. Other architectures that are more efficient or ship less volume will get less optimization simply because there isn't a reason to throw more $$$ at these optimization units if the core architecture and Instruction Set (IS) are already efficient.

Video cards are not bound to a particular architecture. You can have a radically different video card programmed with a similar API (Direct X or Open GL). Perhaps this can be considered similar to the CPU markets where AMD and INTEL have different internal micro-architectures that interpret and execute the same API (of x86 instructions). However, if one architecture is much less effecient than another, it's easier to switch to the more efficient architecture with an intervening well-designed software abstraction layer in-between (DirectX/OpenGL) than to do the hardware-level translation (x86 procs). Video cards don't have to worry about the software compatibility as long as they can support a minimum number of DirectX/OpenGL features. And it seems like add-on (PCIx/AGP/etc) video cards *ALWAYS* have to worry about performance and price more than CPU's. There's a market for slower cheaper CPU's like the Semprom and Celeron but the only market for cheap video cards is in the MB/integrated category. People aren't going to get excited about an add-on video card that's slow.

Re:A tiny market, but a loyal one?

[borg]

Quoth the poster:

"now you could strip all the patented code, and fix it into a working driver, and provide source for it, but ATI already has been doing that for years, yet all I see from the /. community is a bunch of Nvidia fanboy ravings of how good the closed source Nvidia drivers are."

Correction: there is an open source radeon driver that only supports 3D acceleration for cards up to and including the 9200 models. Newer models are only have 3D acceleration with the closed source 3D driver.

Up until ATI stopped releasing 3D programming information to the community, ATI-based cards were all I bought and recommended. The reason is pragmatic: I didn't have to worry about the card working with a new kernel version or the latest -mm patchset. This was my choice, in _spite_ of occasionally incomplete GL implementations (I seem to remember problems with Scorched3D on my radeon).

The last ATI card I bought was a 9200. Now, I buy nvidia. I may be stuck with a closed source driver, but at least it is a _good_ closed source driver. The latest version can do 3D acceleration over multiple cards (xinerama) if all GPUs are similar, which makes for a stunning game of quake on my triple-head system.

If S3 came up with an open source driver that was included in the kernel sources and a marginally competent 3D implementation, I would use them for future purchases in a heartbeat.