Slashdot Mirror
Intel moving on VIA Technologies?
One of our readers sent us an interesting Forbes article detailing some of the "odd" movements that Intel has taken in reference to one of their partners, VIA Technologies. VIA makes a 133 mhz chipset, competition to Intel's 100 mhz chipset. You may remember VIA as the company that Intel accidently sued in mid-April, but withdrew the suit, saying it was a clerical mistake.
Man I don't know how I could look at myself in the mirror in the morning if I worked for a company as sleezy as Intel.
Oh wait. I do work for Intel.
Ra ra, Pentium III.
The development time for 2.2 should be looked at
as being in the upper extreme (i.e. longest
5 percentiles).
I'll take linux, as long as it isn't MSLinux 2005
This sounds like an insidious new technique for throwing your weight around: "We don't have grounds for a lawsuit, we'd get thrown out of court and forced to pay legal expenses if we sued... how about if we just create enough FUD about a possible lawsuit that nobody in their right mind would design this part into their product?"
Who needs design engineers when you've got lawyers?
There is no need for Windows to support this chipset. And there would be no benefit to Win supporting it anyhow. It's low enough down in the system of the machine, that there is likely no benefit at all to Win supporting it. Just like Win didn't have to be designed to support EDO RAM, when it became available.
You really do have to go point-to-point to get decent speeds off a processor-memory bus. Otherwise, you're going to get nuked by "turn around" cycles, arbitration delays, capacitive slowdown effects from multiple loads in an SMP, and crap like that.
Not to mention that you get delay through drivers and receivers so the data going cross the bus should be latched on both ends..
It's going to take a while to get to a GHz bus that isn't a serial link. For the time being, shared address buses with separate data buses to the memory controller probably is the way to go in an SMP deal.
You are correct about #2.
Two of the really nice things about VIA is that they have always supported ECC on the low end and large L2 caches. I have a VP1 with 1MB (2 COAS modules) with a K5/133 (originally a P75, I think) and 128MB 60ns EDO parity and a VP2 with 2MB L2 and 256MB ECC cache with a K5/166 and both are very, very quick. Friends are always surprised by the speed with which they run and always assume a much faster processor. The cache matters and the parity/ECC is just common sense. My new addition is an MVP3 FIX 2012 (?I will have to look) with 2MB L3 cache and a K6-3/400 and it is so fast that it is startling with compiles and so forth. 384MB ECC PC-133 does the trick, with DPT Ultra2. I think that I will be keeping this for quite a few years as well without feeling the need to upgrade -- the first two are, respectively, my web server and my DB2 testbed (soon to be SAP and DB2).
Thank you VIA. Keep it up.
Damn it. 2MB L2 cache and 256MB ECC RAM. I don't have 256MB ECC cache.
... less Oracle ...
Need more coffee
The difference between the two is an "I", not much
more. If you look at some benchmarks then it's
clear that the internals haven't changed at all.
Latency of RAM or harddiscs is much more important than bandwith since it makes the CPU/application to wait. Bandwith can easily be increased by interleaving or widening of data paths, but you can't do much to decrease latency (that's why we have cacheing).
Such techniques has been done a lot in the past, eg. [56]86 class IA32 CPUs are 32 bit CPUs but have an 64 bit external Databus, Alphas are 64bit, but you can see them with 128, 256 (and maybe even higher) bit organized RAMs. Well, I except some K7 Boards with 128Bit as well (the 200MHz EV6 bus makes it possible), so let us all dump RAMBUS.
i could get a
dual motherbaord and two PII 450s for the same price and have 900mhz
Well, gee, couldn't you overclock 2 Celeron 300A processors and get 900Mhz for $160 instead?
I for one will buy it. I am running a small computer network at a behavioral research lab where we work with digitized videos. I dont really care about the processor speed but I need as fast bus as I can get to get as good video streams as possible. After my expirience with VIA's chipsets I will most defenetly buy the 133 mhz boards to maximise our investment in PII's which speed is hinderend by the BX chipsets. The speed difference on our old humble AMD's and Pentiums when I got rid of the TX chipset and installed my old hardware in motherboards with VIA chipset (Aopen A59-Pro)will be my beacon in getting rid of that dreaded Ixxx inside logo. The worst thing about the behavior lab is that the potential users of the behavior software are using windoze as no one has told them that there are other platfroms (hardware and software) that are useable. I would love to be able to use alpha's for our work but WinTels monopoly on the desktop market does not make that an options.
Ahem. Say it with me, M-O-O-R-S--L-A-W.
550Mhz-1100-2200-4400-8800-17600Mhz
CPU speed doubles every 18 months.
Thus, about 8 years.
I suppose I might be wrong about this. But I think that the journalist made one "minor" error in the piece. The FSB is the interface between the hipset, and the processor. So the fact that VIA has a 133 FSB doesn't mean that their chipset will work 33% better with an existing chip, UNLESS, someone overclocks the processor, so that it is running at a multiple of 133, instead of the current 60/66/100 mulitples used by Intel parts. And this makes the Celeron comparison even more absurd, since in order to get any benefit from this 133 FSB on a celeron, you'd be running the chips FSB at double the rated speed. Sure it's possible. And most likely it can be done stably. But all Intel really has to do to slow the adoption of this part, is not realease processors with 133 FSB ratings. By and large, most people don't want to overclock their systems, or run their stuff outside of the rated specs.
If Intel is legally a monopoly, it sounds like the Justice Department has some work to do here. If Intel is refusing to lisence GTL+, then VIA can't produce a chipset which works with Slot 1. I'm not a lawyer, but I think that would be considered using monopoly power to gain a monopoly in the chipset market. Intel also own's a large share of Rambus, and PC133 is the biggest threat to Rambus's royalty attached ram design being widely accepted.
I hate being forced to buy expensive, propriety RAM, which hasn't shown significant performance benefits, just so Intel can make more money. I'd much rather buy relatively cheap PC133 SDRAM, or let competition decide what's the best technology for the next generation of RAM.
http://www.theregister.co.uk/990610-000031.html
http://www.theregister.co.uk/990521-000004.html
http://www.theregister.co.uk/990503-000005.html
That should get you started, you darned Canadian.
Of course I am posting as an AC -- I am afraid of Canadian Revenge-Spammers, as any sensible person would be. Eh?!
NVRAM harddrives have been available for a while. The CRAY T932 has 800GB/s memory bandwidth, I'd say it's safe to say that that bus is running at at least a GHz. I haven't heard of a 10GHz CPU, unless you are talking in the aggregate (eg. 10 processors running at 1GHz each), in which case the CRAY T932 almost certainly qualifies. You should go out and get one, and install an NVRAM harddrive in it.
----
Open mind, insert foot.
IIRC, it ends up that your average performance will be about the same with Rambus as with SDRAM. For large block memory reads, Rambus is faster due to its higher bandwidth, but for small noncontiguous reads it's slower, due to its higher latency. For average use (whatever that is), they supposedly end up being about the same.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
These chipsets nowadays have a lot of stuff in them. Most importantly, the IDE controller. The basic interface is the same, but to use DMA IDE, you need a chipset-specific driver. (My biggest gripe with my VIA-based mobo is that the driver's don't like single-channel IDE, they expect both channels on the controller to be enabled. To save interrupts, I do not.)
USB is usually handled by the core logic, too.
retrorocket.o not found, launch anyway?
Unfortunately I don't have the specs handy anymore, nor am I going to look for them on the big Cray machines. While Cray's 800MB/sec bandwidth over their Craylink stuff is good, it's not really that fast--furthermore, it is designed for aggregate use by several processors. The SGI Origin 2000 machines use the Craylink interconnect, but the funny part is that the interconnect has more bandwidth than the pathetically slow memory on each node board. This means that no single node board (2 cpus) can saturate the interconnect (a good thing if you have a sufficiently parallelizable application, a bad thing if you don't).
On the other hand, check out some of IBM's servers, like the S70A for instance. Or, try the Compaq 264DP mobo that we're running in our lab. _Each_ Alpha 21264 has 2.6 GB/sec bandwidth to main memory. To make this much bandwidth doesn't require that the bus (or point to point connnection(s)) be operating in the GHz range, though. You have to take into account how wide the pipe is. If you pump 8 bytes per cycle for 2.6GB/sec, you can run much slower than 1GHz.
-Paul Komarek
Actually AMD is just now starting to beat out Intel in terms of CPU power. They still lag behind in influence and economical terms. They aren't nearly as wealthy as Intel. They are struggling because Intel is able to keep dropping prices on Celerons since those aren't their main moneymakers. AMD is still around because they make a good product for the money and the DOJ and FTC have been rather active lately. I don't think Intel wants to provoke them anymore.
It's not enough to bash in heads, you've got to bash in minds. - Captain Hammer
Talk about going off half cocked. Can somebody please point me to the story?
The next Cmdr Taco duplicate will be ready soon, but subscribers can beat the rush and see it early!
This is more a patent issue than anything else. If there were no patents involved, why would VIA neeed a GTL+ license at all? I don't know the precise rules here, is there some sort of obligation on the part of a patent holder to license the patent to other manufacturers at reasonable conditions? I feel there should be, but I don't know. A clause that more or less insists that VIA can't use the license to make a chipset that is better than Intel's doesn't sound reasonable.
Whatever your view on patents (and hardware patents obviously have a lot less damaging effects than software patents) they are a sort of government-granted monopoly, so you cannot be surprised when they are used to maintain monopolies.
The point is that other companies aren't doing what intel is telling them to. And intel really doesn't like that.
Indeed. Add to this that Intel has a large stake in Rambus and that every single Rambus module sold will result in a royalty payment to Rambus and you can see why they don't want PC133 and PC266 to succeed.
The K7 may have a 200 mhz bus between the chipset and the processor
According to this great article on K7 rumours Slot A will be able to run up to about 250MHz, but Slot B will go up to 400MHz. Yum yum!
VIA made K7 chipsets [will support PC133]
Here in time for Christmas by the looks of things. This sabre rattling by Intel might even make takeup faster. If people are worried about whether VIA has the rights to the GTL+ bus VIA might advise them to use the EV6 stuff for the 21264 and K7 instead. If only AMD would second-source the K7 so people could really believe that supplies will be reliable. You don't piss off Intel unless you are very sure you won't have to come crawling back. Actually I did see some rumours of a second source for the K7. IBM and Samsung would be the obvious candidates.
While future K7 chipsets will support RAMBUS
It would be ironic if high end K7 chipsets were delayed because they decided to invest a lot of effort getting Rambus to work, and then the RAM modules don't turn up. I think for the high end, with huge 2nd level caches and enormous bandwidth requirements Rambus may have the edge if the caches take the top off the latency problems, and AMD may have thought the same way. And who would have guessed that an Intel-sponsored technology could fail in the PC space?
I've got a pretty hard time lookin' in the mirror as well - I work at MS!
InThane
Intel chipsets use the Rambus memory standard while the VIA chipset uses the PC-133 Synchronous DRAM, which is nearly 35% cheaper than Rambus memory.
Sneakin' it in the back door. Thats like attaching a handgun bill to a public transit proposition.
Bastards. Go VIA.
Four-digit slashdot ID. Recognize.
I'd take that bet, on the basis that I think 2.4 will arrive before that -- and, as far as I know, nobody's even squeaking about a 3.0 yet.
In WinTel there's the word Windows and Intel, if Windows decide to don't support the VIA chipset, VIA users will become like the amiga/atari gang, 0.001%... with better machines than 50% of the 99.999% remaining! :o)
i'm wondering if BeOS will support the VIA chipset too? ish, i'll be forced to have a BeTel machine
--
"Science will win because it works." - Stephen Hawking
When I see Intel expanding their reach into other parts of the system I see them closing doors on competition... basically ripping pages right out of the Microsoft Tactics handbook.
:) antics. I wonder how favorably AMD and Cyrix's reverse-engineering expenses compare against Intel's basic R&D expenses?
:-/
It's great that Red Hat is just giving away money to BeOS, VA Research, Red Hat and all the rest. And Intel can only build better CPU's if they integrate the FPU (remember superior non-Intel 386 FPU's?), bully their way into the networking biz, hijack the OPEN VBL2.0 spec with PCI, bully Intergraph and other chipmakers into "license your stuff to us or no early access to our next CPU + motherboard. *Deliberately* build more expensive "Slot One" CPU's if that's the ONLY way to isolate AMD and Cyrix (it worked).
Intel doesn't even have good technology. They just produced Yugo's on a much higher scale than PowerPC, MIPS, and Alpha. Big deal - Intel's entitled to their success... BUT when they start acting like Microsoft and actually SUPPRESS innovation that really rankles me.
People talk about Microsoft this and that, well, the difference between INTEL and MICROSOFT cannot be summed up as simply as "Intel has clones therefore they are not as bad [as MS]". AMD builds its ENTIRE business on reverse-engineering... everything from scratch and twice documented to cover your ass from lawsuits. If you think about it, AMD could sell their chips CHEAPER if it weren't for Intel's strong-arm (no pun intended..
Intel can do all sorts of nasty things to Linux too. Sure Linux is GPL but how many people care? To reshape the question, how many people refuse to use Microsoft Internet Explorer or Hotmail because it is Microsoft??
Linux has a lot of shortcomings that could be exploited without violating the GPL... just offer enticements seperately via downloads. If Linux achieves world domination *without* open source fixing the shortcomings of the Linux platform it will be vulnerable. A lot of folks would take closed-source binaries from Intel, particularly if they perform better than open source initiatives. Want consistent printer support (especially for non-Postscript models), X displays hand-tuned with MMX assembly language tailored only for Intel display chipsets, co-porting Microsoft Media Player along with the Indio and i268 multimedia codecs, maybe even Linux drivers for WinModems, WinPrinters, and the eventual WinMotherboard.
A few years ago, did anyone think Microsoft could stitch things up so quickly? I still remember running DooM on a Windows-less computer running IBM PC DOS..
We need robust support for Linux on non-Intel CPU's if there's ever to be any real competition, as opposed to AMD playing catch up. AMD might someday exceed Intel's best performing CPU, but it looks like they will ALWAYS be playing by Intel's book of rules.
No, that's a 384-bit bus (64 x log2 6 = 384).
Ooops. 64 x log2 64 = 64 x 6 = 384.
1) VIA isn't THAT tiny. Intel's chipset sales were 1.2 billion. VIA's was 200 million. That's still 1/6th of Intel's sales, which isn't too shabby. Also VIA sells chipsets at bargin rates and Intel doesn't. That means that VIA has shipped a lot more chipsets than Intel has, per dollar amount.
Also, VIA has to compete with ALI, SIS, and a few others for AMD/Cyrix sales. Intel has no such competition for the Celeron/PIII sales.
2) VIA is very close to AMD and I believe that AMD licensed VIA's chipsets for AMD's 640 Socket 7 chipset. VIA will also be one of the main manufacturer's making K7 chipsets.
3) VIA is trying to do an endrun around Intel by using National Semiconductor's license with Intel for the basis of their chipset. This was also done by AMD and Cyrix back in the 486 days, and it worked. However, it's almost a natural that Intel should sue.
4) VIA is mostly owned by FIC, who is a major OEM manufacturer in Taiwan. If they, and other Taiwan manufacturers could purchase Celeron and PIII chipsets locally, they would in a hearbeat. VIA already has very good relationships with all of them (most use the MVP3 chipset already).
Could be interesting...
jf
Linux 2.6 ?? 5 years .. naaaahh, make it Linux 3 :-)
ah, I'd say that the pace of Linux development is increasing exponentially and that in 5 years time we'll hit version 5, released less than a year after version 4.
Life's a bitch but somebody's gotta do it.
Intel's lawsuit is going to fizzle out, I think they are just throwing some FUD into the fan. Personally i'm not impressed by a 133mhz bus or even a 200mhz bus. I wont upgrade to a 133mhz intel board using RAMBUS because it's more expensive then SDRAM for not very much of a performance gain. Why would I want a 533 or 566 PIII for close to 1000$ when i could get a dual motherbaord and two PII 450s for the same price and have 900mhz total (which is what I'm building right now). I'd like to see the DDR-SDRAM become more popular because it's alot faster than RAMBUs and is an open technology, so I wont pay insane amounts of money for 128 megs of ram.
I'm a loner Dottie, a Rebel.
It is very hard to comment on Intel's actions without knowing what VIA's license states. I would assume that they have a license to the P6 bus architecture, in which case what does it matter what speed it runs at? It is still the same bus.
Either way I hope that VIA succeeds in introducing this chipset, competition is good.
Q.
According to Tom's Hardware the 133mhz bus thing only provides incremental improvements over the 100mhz bus, and at astronomically higher prices. BTW, when is the 200mhz bus/K7 combo supposed to come out?
Okay, so 100 to 133 mhz is incremental. So is 66 to 100 mhz, but that doesn't mean gadget junkies and stupid people won't flock to the retailers and snap them up like the "miracle" tonics of the 1800's. I can't tell a darn bit of difference, even playing games, between my K6-2 350/128 MB/Banshee AGP system at 66 or 100 mhz.
You might want to review the prior post and link.
The point is that other companies aren't doing what intel is telling them to. And intel really doesn't like that.
You see, the real issue (although the original article doesn't seem to mention it) is memory. VIA and others have been arguing for supporting a PC133 memory standard. Intel doesn't like this because they want everyone to move to DRDRAM (Direct RAMDBUS DRAM), while VIA and others are more interesting in DDR-SDRAM (Double Data Rate SDRAM). Intel says RAMBUS is great because it runs at 800 mhz. The slight problem is that it's only 16 bit. That means that 800 mhz RAMBUS has the same bandwidth as 200 mhz SDRAM. It's not that great an improvement. Plus, RAMBUS has some serious problems with high latencies.
PC133 is actually just the first step towards PC266, using DDR-SDRAM at 133 mhz. DDR-SDRAM is like conventional SDRAM only it transfers data on both the rising and falling edges of the signal, giving twice the bandwidth at the same speed. Plus, DDR-SDRAM doesn't have the latency problems of RAMBUS. In addition, it costs less and is more similar to SDRAM so the memory manufacturers can switch to making it more easily. If that's not enough, RAMBUS is a proprietary standard controlled by one company, and DDR-SDRAM is an open standard. Here's even more: DRDRAM has some serious yeild problems and intel is still working on some major bugs while DDR-SDRAM is almost ready for widespread adoption.
But intel has absolutely no plans to support DDR-SDRAM or even PC133 (Heck, there are rumors that even when they come out with a 133 mhz chipset they will run the memory at 100 mhz.). Perhaps it's because they've invested a lot in RAMBUS, or maybe because they are stubborn bastards who subscribe to the NIH (not invented here) philosophy. It's probably both, actually. But there are companies like VIA who are saying, "Gee, DDR-SDRAM is a LOT better. Why don't we use it?". This is making intel furious. They are used to having everyone in the hardware world do exactly what they tell them no matter how stupid it is. Increased competition has started to change this, and it's definately for the better.
Back to your K7 comments, this is very much an issue. The K7 may have a 200 mhz bus between the chipset and the processor, but the memory bus is only 100 mhz. This is simply because there is no 200 mhz memory. While future K7 chipsets will support RAMBUS (just in case intel wins and it's adopted), they will also support DDR-SDRAM. VIA made K7 chipsets certainly will, as will VIA made P6 chipsets. There is even a chance DDR-SDRAM will be supported on the old socket 7 platform if VIA releases a new chipset for it (It's not that unlikely. The K6-III will still be made and should go at least as high as 600 mhz in the future). Maybe common sense will prevail and the superior technology will actually win. Then again, with intel's marketing muscle, maybe not.
I am not an idiot. Please use my name to email me.
"That's right, I'm quoting myself."
-Upsilon
well I hardly think Intel has a monopoly.. they had *Less* then 50% of the "sub $1000" PC market last year.. AMD is kicking there ass as Economical, if not in terms of CPU power or influnce :)
---------------
Chad Okere
ReadThe ReflectionEngine, a cyberpunk style n
Various intel leakages have implied there were or
as it appears now should have been 133MHz FSB
parts. Notably the documentation that escaped on
the 0x2A MSR.
I wonder what the real story is
Alan
If we're entirely honest, the only reason to run any chip much faster than (say) a Celeron-333 is to play games.
Outside of the corporate markets, where whatever is fitted in the latest Dell/Compaq/IBM will ship (I know - I've put PII/350s on secretaries' desks), the one thing that drives most power-user types to upgrade is the 3D game. Quake/Unreal/WHY...
The fact is, most of these games have an inner loop that's tight enough to fit (to a large extent) into the on-chip cache: most 3D gaming benchmarks show Celerons based on the Mendocino core (with 128K core-speed L2 cache) are faster clock-for-clock than Pentium-IIs (with 512K half-core L2).
133MHz FSB is a great concept for scientific computing. Intel architecture machines are notoriously lacking in memory bandwidth. But I don't seriously think that VIA is chasing the scientific market (even if they were, Intel's relatively poor FPU performance would probably sink them). They're chasing the people who clock their Celeron-300As at 504MHz: these are predominantly 3D power-gamers.
The increase in performance that you're going to see on a Quake II benchmark is not going to be spectacular - we saw maybe 10% going from 66->100MHz. Figure on half that by going to 133.
133MHz FSB means PC-133 RAM, which is going to be quite a lot more expensive than the PC-100 variety.
Amdahl's law indicates that optimizing the most frequently used parts of a system yield the best performance returns... I think 133MHz FSB for Wintel is missing the point for those who are performance led, rather than technology driven.
Another point that's worth bearing in mind, is that I hear that VIA's AGP implementation is less than stellar in performance. Like: a 133MHz pre-release system benchmarks slower than a system based on the (now-almost-venerable) BX AGPset, because of it.
So, tell me, who's going to buy into 133MHz FSB now?
-- O improbe amor, quid non mortalia pectora cogis!
And Intel can only build better CPU's if they integrate the FPU (remember superior non-Intel 386 FPU's?)
To be fair, anyone who knows anything about architecture knows that putting the FPU onto the same die as the ALU is good sense. I mean, MIPS and most others did it a long time before the 486.
-- O improbe amor, quid non mortalia pectora cogis!
I don't see any of this coming to fruition.
I am not buying another computer until it's upgradeable to a non-volatile RAM hardrive,
With a GHZ bus and 10 GHZ CPU.
How long do you think I'll have to wait?
Cheers
I have to disagree with your phrases about the "failure of capitalism" here. In the case of monopolies, there is almost always government force backing it, making monopolies pretty inconsistent with a free market. Of course, you go on to make a similar point -- my quibble is just with your choice of the subject and the opening phrase. Perhaps "Failure of Intellectual Property Law" would be closer.
Geeky modern art T-shirts
What's the big deal of the 133 Mhz bus ? the K7 is beeeing released with the 200 Mhz bus (that Digital developed), more intel really needs to do something, the PIII is NOT that much of a inprovemnt over the PII, I just hope the k7 has good floating point performance, since in the past that's what kept allot of people away from them.
It doesn't matter what your market share is. No company, no matter what size, should be able to exercise monopolistic practices. And Intel is exercising those practices.
--------
Capitalism has failed when an entity achieves a monopoly, and is able to smother any smaller one. What's worse is that capitalism is not failing by itself; our "Intellectual Property" laws which were designed to promote it are choking it to death.
Capitalism, like anything, works only when you know how to use it.
--------
PC133 does result in a very small increase in performance, mainly because most x86 chips sold nowadays have high-powered on-chip caches - either lots of cache memory (as with the PII and PIII series), or very fast cache memory (as with the K6-III and Celeron). Main memory (and L3 cache, for the AMD chip) simply isn't accessed often enough during normal operation for a 33% increase in speed to be noticed.
But I doubt the price difference will be so great as to be called "astronomical". PC133 is a new spec supported by few motherboards, and all things PC133 are priced at a premium right now. But prices will soon drop as PC133 becomes more common. Wasn't it Micron which recently announced that PC133 prices would be similar to PC100 prices now once high volume production commenced?
As a side note, among the three benchmarks Tom used were two floating-point intensive benchmarks (Expendible and Naturally Speaking). Floating-point performance is dependant on CPU speed moreso than memory/cache speed. And the third benckmark (WinBench 99) is known to be more favorable to chips with on-chip caches, suggesting that main memory speed is less significant in these benchmarks. Then again, Naturally Speaking probably involves accessing/manipulating a large amount of data in main memory (and sure enough, notice the 8% performance increase...). My point being that these benchmarks alone may not tell the whole story. I bet the K6-2, for instance, would benefit quite a bit from PC133 (and notice that there are rumors of AMD releasing a "K6-2 Pro" with PC133 support).