Slashdot Mirror
VIA Quits Motherboard Chipset Business
arcticstoat writes "Following the media hit that was VIA's Nano processor, VIA says that it's now quitting the motherboard chipset business that used to be its bread and butter product for years. VIA's vice president of corporate marketing in Taiwan, Richard Brown, explained that: 'Intel provides the vast majority of chipsets for its processors and, following its purchase of ATI, AMD is also moving very quickly in the same direction.' VIA will still be developing chipsets for integrated motherboards featuring the Nano CPU, but will no longer produce chipsets for Intel and AMD CPUs. Was this the right decision, and where does this leave other third-party chipset manufacturers such as SiS?" Seems like this is a tough business to stick around in.
I have one older PC that had this sort of problem - until a driver update that brought a workaround.
AFAIK the VIA chipset had a fundamental flaw in the first place (data loss on the PCI bus under high load) but such flaws happen to other vendors too and a workaround in the driver is usually acceptable. In this case, the problem showed up in the field and VIA only fixed it after getting bad publicity.
C - the footgun of programming languages
The low power CPU market, where Via is quite strong, is currently expanding a lot, in contrast. It makes sense to leave a shrinking market for a growing one. Their main competitors are likely to be people like TI and Samsung with Cortex A8/A9 MPcore based chips containing everything you need for a computer.
I am TheRaven on Soylent News
Why is there even a chip set business? The stuff on the chipset isn't much compared to whats on the CPU and in 5 years I don't expect there to be much of any market for chipsets at all. Your motherboard will consist of a CPU and line drivers for things like the audio circuit and loads of static control resistors.
To survive it's sometimes best to be good at what you are doing. But being too specialized isn't good either because then you will become extinct as a dinosaur because you suddenly find yourself left behind in a swamp of old technology.
VIA are good at producing low-power devices with reasonable performance for general use. There is competition from Intel now, but since the environmental concerns are growing over time VIA has a place in the server room for some applications that doesn't require a lot of computing power. Many of their processors actually have a lot of the chipset functionality built in instead of in an external chip. So they aren't leaving the chipset knowledge behind completely.
Intel and AMD are working on the high-performance end of processors and chipsets. But so far they have mostly been into standard machines with relatively few CPU:s. And I suspect that there will be an upcoming step when a lot of the chipset functionality moves into the processor itself.
SiS and others have their main area at just chipsets, but if they doesn't have alternate product lines or get into specialized solutions where they are filling out gaps that AMD and Intel are missing they will get a hard time.
The point is that if you are small you may have to provide a more complete product than if you are big. In the case of VIA you can get a complete working motherboard (add memory and go), but many of the other competing manufacturers are still depending on other key components like processors from Intel or AMD.
But then - AMD also need something that can distinguish them from Intel if they are going to survive. The 64-bit instruction set was a good step when it came, but now they need something else to get an advantage again. It's no idea to try to beat Intel at their own game, you must be different in some way and do something unexpected or something that makes you special.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
In response to this fiasco, Intel engaged more directly with the chipset vendors; at the time, VLSI Technology was the leading one. Intel was in the process of coming out with the original Pentium, and VLSI needed detailed specifications so that they could have chipsets available when the processor debuted. Intel promised VLSI information as quickly as Intel's own engineers had it.
Since VLSI had an operation in Chandler, very near Intel's own chipset design operations, VLSI inevitably heard when Intel started up their own chipset team. VLSI was understandably concerned that they were becoming dependent on cooperation from a company that had gone into competition with them, and approached Intel. Intel reassured VLSI that Intel's team would not have any "unfair" advantage over VLSI's engineers, and reiterated that VLSI would have processor specifications as soon as Intel's engineers did.
So, VLSI worked away at their design. Intel released the final Pentium specs, and the Intel chipset engineers accomplished an unheard-of feat: they finished their design, streamed out the chip, fabricated it, packaged it, tested it, and released samples the same day!
Later, Intel found other ways to make life difficult for chipset companies, such as suing chipset vendors for using their bus designs or pricing the processor plus chipset at the same price as the processor alone. This has periodically led to chipset vendors deciding that the business isn't worth it, followed by Intel screwing the pooch with a chipset design, followed by Intel realizing that having more than one chipset provider is good for the processor business, followed by Intel making nice to the chipset vendors, lather, rinse, repeat.
Here we go again. This could be the last time around the merry-go-round, or maybe not.
Lacking <sarcasm> tags,
Uhg. Personally, I've never had anything but trouble from Via chipsets. I'm pretty sure I even had KT133A and KT266 chipset boards at one point. Endless Safe Mode reinstalls of the IRQ routing drivers, occasional Windows flakiness from said drivers, the USB filter drivers, weird voltage/clock frequency stuff... once I spent the little extra for an Intel chipset for my P3, I never went back to Via. A friend had given me a KT-7A RAID and after a little fighting, I gave it back and bought a nForce chipset board. I realize they need drivers too, but at least they don't break the system utterly when not installed and don't require weekly reinstalls. I can't say I'm sorry to see VIA out of the game.
AMD are making chipsets (having consumed ATI who were making them before), haven't you read the reviews? Look at the AMD 790GX chipset, or the 780G ...
Sure it makes sense. Via's constantly having to chase Intel and AMD in order to license the interconnect bus changes. On several occasions Via's had to reverse engineer the bus without a license; and, as a result, run into trouble. It's in Intel's best interest not to keep advance specification updates flowing to the 3rd party chipset manufacturer's because it gives them a head start with their own chipset. Same applies to AMD. Intel and AMD have seen the light and are now producing chipsets targetting enthusiast. They should be able to optimize the chipset solution better than a 3rd party due to inside engineering knowledge. Via will never be able to keep up.
Add to that the obvious pairings:
Intel chipset optimized for Intel processor
ATI/AMD chipset optimized for AMD processor
Nvidia chipset for best performance with Nvidia GPU.
What does VIA offer? Nothing.
Honestly, I think this is the right move for them -- it allows VIA to concentrate limited resources on VIA's embedded processors and the chipset optimized for use with the VIA processors.
That trend has been happening for the past 40 years. If you looked at a 1990's graphics accelerator card (Hercules Graphics Station Card or a Voodoo 5000/6000, you would see that all the different components (RAMDAC, graphics processor, memory controllers) were all on different parts of the circuit board. Now, most of that logic is within a single chip Geforce 9800GTX
Memory chips keep changing as rapidly as the CPU's do. Assuming that a CPU manufacturer wanted to enter the memory chip market, by the time they had caught up with current state-of-the-art in memory technology, bus communication and got the product onto market, the memory chip manufacturers would already be designing, producing and marketing the next generation.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Yeah, nVidia makes integrated video chips chips.
Actualy, nVidia, VIA and SiS each shared one advantage over AMD and Intel - namely they made chipsets for both major platforms. AMD and Intel only make chipsets for their own.
Of course, of the three, I've only like nVidia (their onboard video had passable 3D unlike SiS and VIA, and I found they had better stability too).
An nVidia CPU...
That thought made me very happy. It'd be nice to have a 3rd CPU in the performance market, to compete with AMD and Intel.
Self proclaimed typo king, and inventor of the bear destroying coffee table (patent not pending).
Complete opposite experience here. I admit I haven't used the old slot A KX1xx athlon chipsets KT133/A or the original KT266, but I've run several motherboards on the Apollo Pro 133, KT266A, KT333 and KT400, and haven't had any problems. I even ran several of them overclocked, on XP, 2000, and Linux. In fact, the machine I'm typing this on is a KT400 with 1GB RAM and a 2GHz Athlon XP running vista, with no stability issues whatsoever, it's a bit slow but I put a lot of load on it and there are server apps running in the background too.
The KT266A board that I had (Epox 8KHA+) was one of the fastest boards I ever owned, for its time. And it never had any problems, even overclocked.
I can understand that people have had issues with several VIA chipset revisions. But they were in many instances a lot better than the alternatives. They were much better than intel during the i820 fiasco and have always been somewhat better than AMD's native chipsets (until the K8 chipsets that is).
In fact, until nVidia came along with the nforce, they really were the only option for athlons. I'll admit that the nForce/2 offered some stiff competition and was good, and that nVidia eventually did usurp via with the nForce3 Ultra and beyond.
You speak as someone who has limited anecdotal experience with a few via chipsets. Well, here I'm offering mine, with a few facts to back it up, as well as the experiences and opinions of many I've met over the years.
VIA definitely played an important role in the game. For one, they were partially responsible for the Athlon's ascendancy. And second, they provided competition for Intel's chipsets when those were lacking. It is sad to see them exit the business.
Do not downmod posts "overrated" simply because you disagree with them.
Speaking as someone who has PC's dating back to the very beginning of the 90s (Possibly even '89 for the 386) I think *HALF* of all my motherboards are VIA based. And y'know what? I've NEVER had any of the major problems that were reported on them, other than the AGP FastWrite bug (And quite frankly, I've had *THAT* problem on every mobo that was supposed to support it, be it Intel VIA OR SiS). Furthermore my one current nice gaming board, which I got secondhand from my cousin is an ABIT VT7 with a Via Dual Channel chipset on it. It spanks all but one board I've got, is rock solid and has great onboard sensor support. You think Abit would put out 150+ dollar enthusiast boards with a Via chipset if they didn't consider them stable/high-performance enough?
Granted Via has produced lots of cheap or buggy parts over the years, but if you look at their range of market segments they were the only player to try and cover all their bases. Otherwise you had SiS on the ultra-low end, or Intel/Amd up mid-high range (Plus back in the day the various other nb/sb makers, but really, we're talking p2 era up here, there were what, maybe 5 manufacturers?)
So yes, Via's importance in this market cannot be understated enough, if for no other reason than to give us alternatives to an otherwise dangerous duoculture.
Mind you in their defense I think much like nVidia's licensing snafus for intel p2p bus-link stuff, the simple reality of the matter is that the patent licensing to manufacture new chipsets is prohibitive to their profit margins, given the dwindling RoS available in the low end of the market. Furthermore the only reason they were doing well for so many years was because they were infringing a lot of IP to make Socket 370 and earlier pin-compatible (cpus? Boards?). After intel sued them.. what 5-8 years ago? They worked out an agreement to license the patents from intel in exchange for discontinuing use of pin compatible 370 cpus and not making ANY pin compatible 423/478 cpus.
As such the background of the how's and why's here may take a bit of digging to uncover.
Only kinda. I'm yet to see a cheap Mini-ITX, Nano-ITX, Pico-ITX board from Via. They're always very expensive. Especially when you compare them to the Atom options today. The cheapest Via I can find is their EPIA ML8000AG with an 800 MHz C3 processor costing almost twice as much as Intel's D945GCLF with a 1.6 GHz Atom or Intel's D201GLY2 with a 1.2 GHz Celeron.
Back when Via were the only ones with Mini-ITX boards the premium was somewhat okay, but not any more.
We do not live in the 21st century. We live in the 20 second century.
Well, just looking at market capitalization: VIA == $25.26 Billion New Taiwan Dollars == $810 Million US Dollars AMD == $3.13 Billion US Dollars I think it would be quite challenging for VIA to buy AMD. Even less likely, given that there is typically a premium required for a buyout.
Wow, way to miss the point. I've worked with plenty of Intel-based systems and in the post-Windows 2000 world, they're generally every bit as stable. Say what you will about security and usability (there's plenty to complain about there), but I don't hear a lot of people complaining about stability much anymore. That simply wasn't the case with Via's 4in1 trash or SiS's, well, anything they built.
You're right in that Linux sometimes survived on the same box; after garbage chipsets had been on the market long enough, the kernel developers had figured out which features would and wouldn't cause problems. Kudos to the developers for having the time and drive to write proper drivers when Via never could be bothered to do so in the first place. Windows did, at points, have patches to fix issues with Cyrix processors (for example), but it's a little ridiculous to expect Microsoft to go write workarounds for sub-par gear. Likewise, it was a little unfair to blame them for what was really the fault of uber-trash drivers and physically faulty hardware.
Perhaps we'll see nvidia entering the CPU business some time soon
I guess you missed the memo. nVidia are an ARM Cortex A8 licensee. Their platform contains a multicore ARM CPU, an nVidia GPU and a few other things on a single chip. This is the kind of product you see in the fastest-growing part of the computing market.
I am TheRaven on Soylent News
For a while, north and south bridges from a manufacturer like Via communicated with the same interface, so you could make boards for AMD and Intel chips (post Slot 1 and Slot A - before that the motherboards were the same) with only the wiring between the north bridge and the CPU needing to be different. In some cases, you would combine an Intel or AMD north bridge, with a third party south bridge that supported more peripheral interfaces.
Now, the features of the south bridge are pretty fixed. AMD moved memory controllers off the north bridge, so it doesn't do much, and most chipsets integrate the north and south bridges on a single chip. In the embedded market, it's common to integrate a CPU core or two on the same bit of silicon, and I expect to see this trend continue into what's left of the PC market.
I am TheRaven on Soylent News
Nvidia *IS* in the CPU business. We call their products GPUs, and we try to limit their use to display adapters, but GPUs are really slightly specialized CPUs. Go on, split the hairs, but it's way more true than false. There is even clustering and app s/w for GPUs.
No, NVidia is in the stream-processor-specialized-for-3d business.
We don't try to limit their use to display adapters. Rather, they're designed for use as display adapters, and they're not as useful outside of that niche. Current GPUs are actually incapable of being used as CPUs -- there is literally no possible method by which one could configure a GPU to boot and run an operating system independent of a CPU. (If you've never looked into it, the programs which run on a GPU are very limited in size, and support for branching and general purpose logic is limited.)
Besides, even in the areas where they might have some application outside 3D, they aren't doing so well. GPGPU (the use of GPUs for general-purpose FP number crunching) has failed to gain much of a foothold in scientific computing. In practice, they're very hard to program, the GPUs have been optimized for the wrong thing (fixed point & 16/32-bit FP instead of 64-bit double precision FP, since DP wasn't necessary for game graphics), and branchy code is difficult to adapt to them. Worst of all, there are major questions about the reliability of the results. GPU designers are able to get away with less data integrity protection structures in their circuits than CPU designers, since it is not a critical error if somebody playing a game sees a 1-frame glitch because one vertex calculation was wrong. That isn't an option for science & engineering applications.
The GPU guys are trying to rectify some of these shortcomings. By the time they fix all of them, what they have will be... a CPU. And Intel's cutting to the chase by making a GPU which is an array of x86 CPUs with stream processing oriented vector extensions.
This is not a new phenomenon in graphics. Read some of the standard textbooks and you'll find out that decades ago the same cycle occurred, even in the context of 2D graphics. Early graphics workstations used CPU cycles to do all rasterization, then people started designing fixed-function accelerators to help out with the parts which were hardest to do with the CPU, then people said 'hm, if I put some programmability in these accelerators, they could easily cover more of the graphics pipeline', then the programmable functions expanded, and the accelerators started looking more and more like general purpose CPUs, and then people started realizing they could just design a general purpose CPU powerful enough to render the accelerator irrelevant.
(And then the cycle starts again.)