VIA PCI bandwidth issues confirmed

[Dr Mordrid]

TecChannel.de put VIA and other chipsets on a hardware analyzer and found the root of the problems. Read their excellent analysis here;

http://www.tecchannel.de/hardware/817/index.html

Summary:

VIA chipsets only transfer 24 packets/burst before having to ask for another address. This limits them to 96 packets/block.

In contrast: Intels chipsets can transfer 1024 packets/burst and 4096 packets/block.

Not a trivial difference

This expecially affects RAID adapters and other high bandwidth devices.

VIA's "RAID patches" do help some by patching registers in the northbridge, but only with chipsets that have V-Link.

Even then the VIA systems cannot achieve the PCI bandwidth of other chipsets.

Dr. Mordrid

[Ryu Connor]

The article in question is a few months old.

The 8235 Southbridge was their answer to getting caught.

[Dr Mordrid]

Yes it is a few months old, but the VIA issues are years old....dating all the way back to the MVP3 K6 chipset. This makes one wonder if VIA's engineers are deaf, illiterate, stupid or just plain ignorant

As for the KT8235; I'll believe it when I see it.

Meaning; when I plug both an RT.X100 and a RAID card into a system based on it and see it work I'll change my mind, and not before. Been burned by VIA's promises once too often.

Dr. Mordrid

[Elie]

Doesn't matter, even the newer chipsets by VIA are crashing like there's no tomorrow, and the bandwidth issue has not been resolved.

Regards,
Elie

[Dr Mordrid]

Yeah...they did make a big "splash" (or was it crash_BOOM?) at QuakeCon, didn't they?

Dr. Mordrid

[Ryu Connor]

Yes, I already beat this one like a dead horse.

http://www.tech-report.com/onearticle.x/3244

http://www.tech-report.com/onearticle.x/3280

As for the 8235 RAID-0 test have shown an immediate improvement over the 8233A (the last broken bridge).

http://www.hardtecs4u.com/reviews/2002/msi_kt3ultra2-br/index14.php

[Dr Mordrid]

A maximum read burst of 83.8 mb/s. That's basically what you get by applying the RAID patch to the KT8233.

A bit faster than the KT8233 (what isn't?), but still not up to the standards set by other chipsets.

Dr. Mordrid

[Ryu Connor]

It's also HDTach, which doesn't give the maximum burst, but only the maximum sustained throughput. In other words, it doesn't stress the PCI bus.

The 20MB/s gain on the other hand does show (as noted) the sort of difference gained between a patched and non-patched VIA solution. So obviously something was changed for the positive.

[Greebe]

Please wrap this up Terry, it's getting on my nerves in a big way and we all know how ugly that can get

[Dr Mordrid]

Ryu: so...VIA users should wait for or buy a KT8235 chipped board, who's main advantage seems to be that it codifies the RAID patch?

UhHuh...

Greebe: no kidding!!

Dr. Mordrid

[Ryu Connor]

Ryu: so...VIA users should wait for or buy a KT8235 chipped board, who's main advantage seems to be that it codifies the RAID patch?

If it was that simple the KT333 + 8233A boards and older could have implemented a firmware fix, much like was seen with the SBLive! fix. The SBLive! fix was also made part of the 4in1 drivers.

The RAID patch never saw implementation into firmware because the needed changes would have disqualified boards based from being WHQL certified. This is also why the RAID patch they developed was never made part of the 4in1 drivers. VIA's patch also had the larger tendency to adjust other people's hardware versus modifying the chipset.

There was also the problem of compatiblity. I'm not sure how much you've kept up with George Breese's efforts, but it's been extremely difficult to fix the performance of the bus without impacting the compatibility. The adjustment of a single register did, could, and would have the impact of preventing PCI based modem users from being able to use their modems. That is but one example of the problems and it alone is an unaccetable trade off.

http://www.tecchannel.de/hardware/893/0.html

TecChannel served up Wen Chen Chi on the matter.

And as noted the bus has been redesigned.

http://www.tech-report.com/reviews/2002q2/via-p4x333/index.x?pg=1

If one needed to a reason to purchase a VIA board over other brands I would assume they would do so because the SiS board is [sic] "slow."

http://www.tech-report.com/reviews/2002q1/via-kt333/index.x?pg=1

[Flying dutchman]

I just regret the fact that AMD processors become the victim of this. The only reason I refuse to buy an Athlon is because of the lousy chipsets.

My Asus CUSL2-C with Celeron-900 has never let me down with video capturing, even full-frame Huffyuv under Windows 2000, there simply seem to be "no issues" with this hardware.

[Ryu Connor]

The only reason I refuse to buy an Athlon is because of the lousy chipsets.

They have been rather slow to mature.

there simply seem to be "no issues" with this hardware.

There are a few issues:

MCH:

http://developer.intel.com/design/chipsets/specupdt/290695.htm?iid=ipp_815epchpst+prod_spec_mch&

ICH2:

http://developer.intel.com/design/chipsets/specupdt/298242.htm?iid=ipp_815epchpst+prod_spec_ich&

But I know what you mean. My Asus CUSL2-CB has been rather hard to kill.

[The PIT]

Yes Via have got their problems my main problem with them is that they first deny it and then say it effects all chipsets. They then admit it and then deny it again just as quick. Later they say the fix is in the 4 in 1's and forget about the problem.

[Greebe]

Ryu doesn't it feel like you're preachin to the choir here?

Yes the SiS chipset isn't as FAST as VIA, but then again they also don't have any of the headaches nor do they need special attention to specific patches and even then are only a tad slower.

I pick simplicity over half assed patches designed to cover up hardware cludge and a small speed loss and stability over VIA's instability anyday.

Question is why do you feel the need to counter every truth we state by shoving VIA down our throats? We've had the ill pleasure of using them all for eons now, so your comments are both unneeded and unneccessary. Oh just wait they'll fix this and they'll fix that... just give them time... Bahhhhhhhh how many years or their denials and forced to goto the extreme to prove that for once and for all that we were correct before they get off their lazy arses and actually fix it... far to long IMHO and many users in the know agree.