In article <dnsuse$ivh$(E-Mail Removed)>, "Bill
Morgan" <(E-Mail Removed)> wrote:
> Anyone experience any problems with this barebones system? I added a
> Sempron 64 2600+ CPU, 512MB Kingston memory, Seagate 80GB SATA HDD, Sony
> Combo DVD/CDRW drive and was hoping that the system would run silently and
> reasonably fast, using the on-board video.
>
> It was very unstable, crashing on playing DVDs and even on using quite
> ordinary applications.
>
> I then inserted a spare AGP 128MB video card and it hasn't crashed since.
>
> ************************
> Bill Morgan
> (E-Mail Removed)
> Crickhowell, Powys, UK
> ************************
I would describe your results as "normal". There have been
a number of different chipsets, where use of the built-in
graphics core, was enough to make things unstable. In some
products, the answer was for the BIOS to prevent the fastest
RAM clock modes from being used. In other cases, performance
settings were altered from BIOS release to BIOS release, until
a measure of stability was achieved.
The way this works, is there are well known rules for package
design for ICs. If you have high current I/O on an IC, you might
need one power/ground pin pair, for two I/O signals. For
a slightly less aggressive I/O technology, perhaps four I/O
signals can be powered by one power/ground pin pair. These
rules are usually dictated by the fab, and if you follow the
rules, then most of the time your design will work. A chip
with 600 I/O on it, might end up with 900 pins on the bottom,
to make it stable.
Chipsets are very cost sensitive. In particular, microATX
motherboards, with built-in graphics, are intended for bulk
deployment in corporate situations. That means they have to
be the cheapest-of-the-cheap. The more pins are placed on an
IC, the more it costs. Much of the cost of making an IC, is
the package and the pin count. That means there is cost pressure
on designing the package properly - it must take a lot of
empirical lab work, to make chipset that are on the edge of
stability, the way that some of these micro ATX products are.
What can you do as an end-user ?
1) Turn down the memory clock. That is applicable to Northbridges
with integrated graphics, where the memory controller is in
the Northbridge. If the Northbridge is dual channel,
sometimes running in single channel mode (i.e. leaving 64
I/O pins turned off) can help as well.
2) Use a separate graphics card and get to keep your memory
running at full speed.
Instability like this is not limited to integrated graphics chips.
My experiments with my Nforce2 chipset board revealed the
same kind of behavior. The chipset in that case is near its
performance limits, in terms of internal electrical noise.
The solutions mentioned in (1) above work for that chipset.
Can you blame the motherboard maker ? Yes. There have certainly
been a number of boards that should never have shipped. But
the competition in the industry, means that you cannot
"play unless you ship". So, marginal designs leave the factory,
and the motherboard companies try to fix the mess via BIOS
releases.
This is one reason why BIOS adjustments are so important to
any motherboard. The simple minded BIOS, that have no ability
to adjust memory clock, memory timing, voltages for Vcore,
Vdimm, Vagp, and so on, leave the customer stuck when such a
problem strikes. Always download the manual, and make sure
the BIOS has options like that, in case the chipset is a dog.
Paul
Similar Threads
