"Flaws in Via chipsets hit ATA/133, SCSI performance"

[AGodspeed]

Coming from The Register, eh. I guess a lot of people are ***confirming*** this one, since everyone is taking it as fact. I'm not effected, and my other home machine isn't effected. The fact of the matter is, even if there is a flaw, is this really going to effect enough people to even mention, or is it just one of those usual very concentrated flaws with very specific components. Lol, I remember reading an article that said that if you had ANY VIA chipset, a GeForce 3 series card, and Win95/98/ME, that you would suffer from a 30% performance decrease. Lol, that obviously was true [hint, sarcasm].

Move along, nothing to see here...

 

[oldfart]

It didn't originate from The Register. It came from Tecchannel.

 

[Diable]

NicColt, the High Point RAID controller on the AK35GTR uses the PCI bus so its going to be effected by this problem to. Also if you go to this link you will see that TecChannel tested using a High Point RocketRAID controller which uses the same RAID chip as the AK35GTR and the VIA limited its performance.

 

[sMashPiranha]

So why is it that it should affect PCI SCSI/IDE boards but not, for e.g., PCI video cards? I mean doesn't a PCI IDE controller just xfer data through the PCI bus to the North bridge and back just like a video card, sound card, firewire card, or any other card would? Why only ATA-133 and SCSI?

 

[Bovinicus]

It's definitely something that needs to be resolved, but not at this moment. For most users, harddisk performance is pretty inconsiquential. Most users don't really stress the bandwidth of their buses with one or even two harddisks. The way things are right now, I don't think it will really matter. As devices get faster... it could pose a problem. However, I do agree that those who use RAID setups and are in dire need of good disk performance shouldn't really look at VIA's boards. I have the 8KHA+, and I love it. This problem hasn't seemed to affect my needs. I hope it gets resolved, but I'm not going to make a big deal out of it in the meantime.

 

[Damascus]

<< Damascus,

uhh no.

WD has a special-edition HD with 8-MB cache, and we're also talking about SCSI. Try 16 MB+ there >>



That's why it's called a "special-edition". Most IDE drives do not have more than 2MB
SDRAM. Woohoo I knew about those 16MB+ X15's and stuff too. But if you have enough
of those drives to stress the PCI bus (ie RAID5 array) what are you doing playing with
a VIA board? U160 works best with 64-bit PCI any how.

 

[Insane3D]

"Not true. It affects data transfer rates and processor utilization from both on-board IDE devices, as well as PCI add-in controllers."


Quoted from here.

"Note that later VIA chipsets such as the P4X266A and KT266A no longer have their IDE controllers mapped to the PCI bus, so using the on-board controllers should get you around this difficulty."

Just the facts, man

 

[oldfart]

Onboard is a bit confusing. By onboard, they mean an onboard SCSI or IDE Raid controller such as a Highpoint or Promise chip. They use the PCI bus and will have the same issue as an add in PCI IDE/SCSI controller. The KT266A IDE controller does not use the PCI bus.

 

[Pariah]

"Sorry but burst speed has little to no effect on performance."

Compare performance between the WD WD1000BB and WD1000BB SE and tell us that burst speed has no effect on performance. The only difference between those two drives is the size of the disc cache which means more opportunities to burst data.

"Dude, largest cache on most IDE drives is like 2MB. "

True, but take that tiny 2MB buffer off and tell us again that burst speed doesn't matter. No buffer to little buffer will yield more performance than little buffer to very large buffer.

"The issue won't affect you unless youy're running at least 3 drives in RAID. Use some common sense, most drives don't come close to utilizing ATA/66."

The newest WD drives hit 50MB/s. 2 of those will begin to put a hurt on the PCI bus, 3 will be well beyond the limits. It won't be long before other IDE drives are surpassing 50MB/s themselves. 2 X15-36LP's already surpass the real world limit of PCI, with a few others right on the border.

There is no 16MB X15 either. Seagate originally planned one, but it was cancelled. All of them have 8MB. Some of the AV Cheetah drives have them and I believe the Maxtor Atlas 10K III has a 16MB cache as well.

 

[ST4RCUTTER]

Not true. It affects data transfer rates and processor utilization from both on-board IDE devices, as well as PCI add-in controllers.

It's no surprise, really. Each and every VIA chipset has suffered from laughably low IDE performance. It's pretty disappointing when the lackluster ALi chips even surpass VIA in these areas


Since I own both KT266A and SiS 735 boards I'll chime in on this one. IDE performance isn't laughable at all...in a word I'd call it "excellent", I have two 7200 Maxtors, and after dropping one into the SiS box, they both bench and perform roughly the same. If there is a difference, it certainly isn't noticeable.


It should also be noted that the lowered IDE performance occurs when using an ATA133 card so it's no surprise the onboard ATA100 controller on my AK31 is just as fast as the ATA100 on the SiS K7S5A. As pointed out by Insane3d, the Southbridge no longer maps to the PCI bus on KT266A so unless the user has a PCI IDE card they'll never see a problem. I'm not saying VIA didn't screw this one up, just that it's not as bad as some would make it sound.

 

[MadRat]

I think your IDE controller is ported through the PCI bus.

 

[Leon]

It's all Nvidia fault!

 

[AGodspeed]

It's all Nvidia fault!

Yes it is.

Anyway, when I get my K7N420 board (maybe tomorrow), I'll hopefully be getting the most stable Socket A board on the market.

 

[VBboy]

AGodspeed,

I bet tomorrow you'll be complaining about smoke coming out of your non-Via PC

 

[andrey]

I'm even more glad now that I decided to stay with Intel chipset / Intel CPU. There are no 4-in-1 stuff which I need to update on weekly basis and there are no chipset / nVidia issues. Everything just works the way it is suppose to work.

Cheers!

 

[pyrrhus]

Well, in light of this finding, I just busted out HDTach and Sandra - both were bad. Sandra scored my RAID array (using PCI Promise FastTrak100 controller) 10000 points below their standard for an ATA100 RAID 0 array. HDTach's graph spiked all over the damn place.

Will they be able to fix this with software or is the flaw in the hardware? If its hardware I'm gonna RMA the thing... I have no idea what I'd get to replace it, though.

 

[Diable]

Pyrrhus, its a flaw in the hardware that dates back to the MVP3 chipset for K6 processors.

 

[SuperSix]

<< What do I think?

I think I'm damn glad there aren't any VIA chipsets in my system.
>>



Amen brother. ALL 440BX in this house, (Well, ALI in my son's machine, and 430TX in the proxy server)

NO VIA!!!

 

[Raspewtin]

<< I think I'm damn glad there aren't any VIA chipsets in my system. >>




me too

 

[LostHiWay]

No more VIA for me. I just got rid of my MB with the kt266a chipset. I replaced it with the IWill Xp333-R with the ALi chipset. I wasn't sure about the ALi chipset when I ordered but it's damn stable as a rock. After fine tuning it beats the kt266a in everything except memory benchmarks, and that should be fixed in later BIOS releases. I also get really good RAID performance!! Finally a board that works!!

 

[VBboy]

Guys, you all want to switch from Via to another chipset maker, yet tend to overlook one thing.

ANY company can have a defective/underperforming product. It's just that Via's chipsets (namely KT-266) have been out long enough for everyone to benchmark, examine and discect them. The good old Pentium had an Fdiv problem, Trident used a video chipset that wasn't Direct-X compatible in their "gaming" videocard, and Creative had a lot of compatibility problems.

Who knows what bugs other chipsets have? Only time can tell.

 

[Doomguy]

<< No more VIA for me. I just got rid of my MB with the kt266a chipset. I replaced it with the IWill Xp333-R with the ALi chipset. I wasn't sure about the ALi chipset when I ordered but it's damn stable as a rock. After fine tuning it beats the kt266a in everything except memory benchmarks, and that should be fixed in later BIOS releases. I also get really good RAID performance!! Finally a board that works!! >>



Real-world benchmarks still show the KT266A in the lead.

 

[Damascus]

<< There are no 4-in-1 stuff which I need to update on weekly basis and there are no chipset / nVidia issues. Everything just works the way it is suppose to work. >>



They only time I have ever touched the 4-in-1's was when I installed Win2K for
the first time on my machine. Haven't needed them since. And nope, I didn't need
them for 'basic stability' as some would claim. My system did not crash a dozen
times between the first boot-up and the installation of the drivers.

Hell I didn't even need them when I installed Linux. It should be clear that the
inherent flaw is within the monolithic Microsoft Windows operating system.

About the cache issue, I've seen only a handful of benchmarks, but the 8MB WD
drive only wins by a few % in most of the ones I've seen, taking a 10% win in
only one or two instances (sound like another red herring we know about? )



<< Real-world benchmarks still show the KT266A in the lead. >>



Yes, it's quite saddening. People are perfectly willing to throw away boards based
on the (poor) results from synthetic benchmarks when most real world ones show the
boards are actually in the lead.

 

[Pabster]

Damascus wrote:

"Yes, it's quite saddening. People are perfectly willing to throw away boards based
on the (poor) results from synthetic benchmarks when most real world ones show the
boards are actually in the lead."

Quite to the contrary, most people recommend KT266A because it is "leading" -- in synthetic benchmarks.

Anyone who has used, for example, SiS 735, knows well there is zero "real world" performance difference between these chipsets. Zero.

Some even offer that there is zero "real world" performance difference between AMD760 and KT266A. That's debatable.

Anotherwards, like I've said previously -- if you play your games @ 640x480, and can honestly discern the difference between 202 and 220fps, by all means ...

 

[rahvin]

Pabster, the SIS 735 has "issues" just like the VIA chipset does. We were all spoiled by the BX, intel should get an award for the stability/compatability of that chipset.