tHatDudeUK said:
It's a Nebula DigiTV which cost me just over £100 at the start of this year.
I think the problem is that it's PCI. I think the USB2 version should
technically be ok (at least that's my understanding.)
You didn't describe your problem, but start here:
http://nforcershq.com/forum/viewtopic.php?t=50182&start=15
About a third of the way down the page, someone advocates using
wpcredit. Perhaps you can make some sense of what they are
doing.
There are two parameters on the PCI bus that are used for
tuning performance and fairness, There is a "PCI Latency Timer",
and it is a policeman on the bus. Many BIOS expose the setting,
so you can adjust it, and modern motherboards set it to 32 or
64. The numbers are times, possibly measured in PCI bus clock
cycles, as to how long a PCI card can have bus mastership. If
a card runs for too long, the latency timer kicks the card
off the bus, and then the next card has a chance to do any
pending transfer. AFAIK PCI arbitration is round robin, so
if any card does extraordinarily long transfers, the other
cards can get "starved" for service opportunities.
A second parameter, is the burst size set by the card itself.
At one time, the recommendation for burst size, would have
been a cache-line-sized chunk. Whether that is a factor
any more or not, would depend on whether the PCI scatter
gather DMA space is cacheable or not. I would think a card
would do a single burst, and release the bus to the next
card, so if the burst on a card was set to less than the
global pci latency timer, then the cards should play nice
without needing to be thumped by the latency timer. (Of
course, if one card has its burst size set to a ridiculous
value like a burst of 1 word, it will starve itself. I
think the policy used on a proprietary system I worked on
years ago, was to always set all the cards to exactly the
same burst size, in lieu of wasting time trying to tune
them.)
Adjusting those parameters, could correct your problem.
Storage controllers are notorious for having their transfer
size set really high, to give exaggerated benchmarks.
Reducing the burst size used by the SIL3112 might help, or
reducing the PCI Latency Timer will force all PCI devices
to a common max value. (My A7N8X-E doesn't have a PCI
Latency Timer setting in the BIOS, so I guess I would need
WPCREDIT to adjust it.)
As to whether the AGP max burst needs to be adjusted, I
would only bother with that if the other adjustments aren't
helping.
Good luck,
Paul
)