"Colonel Blip" said:
Hello,
All of the help around memory and ASUS board raises another question in my
mind about motherboards from ASUS. I've had several, all around the VIA
chips. I'm wondering what the pros/cons of VIA based vs. nVidia based
chipset motherboards. Is there some inherent advantage of one vs. the other?
I generally try to o/c but just modestly trying to make sure the system gets
the most performance at absolute hardware stability.
Thanks.
Colonel Blip.
E-mail: (e-mail address removed)
To answer that question, you need to look at a block diagram of
a typical S939 system. See the figure at the bottom of this page:
http://www.via.com.tw/en/products/chipsets/k8-series/k8t800pro/
First of all, memory connects directly to the processor. That
eliminates Nvidia and VIA, from any discussions about memory
stability. At one time, the Northbridge interfaced to memory,
and some chipsets were better at driving memory than others.
Now, you are dependent on the S939 processor having a good memory
controller design.
The next thing to see, is the HT bus that goes from the processor
to the Northbridge. HT has an upper limit, and it could be 800 or
1000MB/sec. Since HT is a full duplex bus, there can be 800 up
and 800 down, 1000 up and 1000 down. That is why sometimes the
diagram will show 1000 or 2000, depending on how the artist is
counting the bandwidth.
So, you want a chipset that support 1000 on HT if you can get it.
When using the board, FSB x HT_mult = HT_rate. If you run the FSB
at 300, and set HT_mult to 3, then the resulting 900 is less than
the 1000 max. If you had a limit of 800 to work with, the HT_mult
would have to be set to 2.
Does HT matter ? Well, all the I/O is going through there.
Including sending commands to the video card.
After that, we are comparing internal features. For example,
Nvidia has a firewall in one of its chips. And depending who
you talk to, that can be either a feature or a bug.
The next thing to consider, is clocking and issues related.
Does the chipset have a "PCI/AGP lock" ? The lock is another
way of saying, "Is the AGP bus rate independent of the other
clock settings ?". By independent, that means a separate clock
signal comes from the clockgen to the AGP interface circuitry,
and when you dial up the FSB clock, the AGP clock can remain
a nice steady 66MHz. Similarly, PCi remains a nice steady
33MHz.
Clocks that need to be stable, are PCI clock, AGP clock,
IDE disk drive clock, SATA disk drive clock, and any interface
circuitry that glues those blocks to the system busses. If you
want an obscure bug, for example, there are some chipsets where
the AC97 onboard sound goes for a **** if you overclock, which
means the designers did not do a good job of gluing the sound
logic to the Southbridge.
If you read up on overclocking experiences, like Anandtech
reviews, you may find some hints there as to which chipsets
have SATA clocking problems, and at what frequency they happen.
Sometimes, the workaround is to use whatever separate RAID
controller the motherboard has, as the home of the boot drive.
There is less chance of such a controller locking the SATA clock
to the PCI bus clock, or making a silly mistake gluing the
SATA interface to the PCI bus. So a chipset SATA clocking problem
is not the end of the world.
The bus that runs from the Northbridge to the Southbridge is
also important. If has to run faster than the busses that
feed off of it. VIA chipsets use 266/533/1066MB/sec busses,
and it is a good idea to check the chipset, to see which is
being used. Intel uses 266MB/sec on theirs. I don't know what
Nvidia does (as Nvidia provides no documentation on their
chipsets). As long as the North/South bus is faster than a
reasonable mix of bus traffic for devices off the Southbridge,
you should be in good shape. There was a time when a PCI
bus joined the Northbridge to the Southbridge, and with
only 133MB/sec to offer (half duplex too), I/O was slower
than it had to be. Fortunately, those days are over.
HTH,
Paul
