Greg said:
If I had a PCI Express spec, it might be possible to tell from the
state diagrams in there, whether the lanes function independently,
or work as a group. My comment about training is based on other
high speed serial interfaces I've worked with.
The idea that a video slot would have an unbalanced set of TX and
RX doesn't make much sense. The chipset comes with them in
equal numbers. The coupling caps are on the direction from chipset
to video slot, which the NI article claims is the direction they
would not shrink down. In the opposite direction, the video card
would have the caps (so no reason for fewer lanes on the motherboard,
as no component cost would be saved by the motherboard maker).
The only resource this would save, is routing resources. Making an
asymmetric TX/RX would mean less diff pairs to route. But the board
is full of stuff like that, so I fail to see a really big saving.
The premise is a bit skewed as well, when you consider the low
end video cards with Hypermemory and Turbocache. Those two methods
are used on low end video cards, to allow some texture memory to
be located in system memory, rather than video card memory. During
an operation that required movement of textures into or out of
the video card, information could be travelling in both directions,
according to whatever they use for a cache management algorithm.
Now, being low end cards, I don't know if you could tell whether the
slowness of the card was caused by the concept they're based on,
or on an imbalanced TX/RX lane configuration.
I really haven't seen mention of any anomalies in postings here,
to suggest a problem like this. There was one case of an SLI
board (might have been A8N-SLI family), where an Areca RAID
card placed in the second x16 slot, didn't work. A BIOS upgrade
fixed that, and then the card worked as far as I know. Not
too many people try stuff like that, so perhaps there just
aren't enough reports to uncover something like this.
I don't recollect any benchmark articles where there was a
strange difference between read and write to RAID arrays.
In terms of practices, HP is the last company I would suspect
of doing something like this. I can see a company like Asrock
perhaps, because they do all kinds of crazy stunts like that.
("AGI slot"). When Asus pulls a stunt, usually the user manual
contains an admission of guilt (like when a x4 slot only has
x2 lane wiring - another good reason to read the manual before
buying). Some of the Asus Intel boards, that have a couple x16
slots, will have x16 for one, and x4 wiring for the other,
but we can kind of figure that out from the known lane limits
of the chipsets being used (few chipsets have enough lanes to
do x16/x16, so no surprise there - the X38 might be the first
to do it). Some of the high end Nvidia SLI boards rely on both
the Northbridge and the Southbridge, having a video card interface
each. So that is how they manage it.
To answer your original question, I don't see an easy way
to detect this situation. If you had a high resolution picture
of the motherboard (and those are hard to find), you could
look for differential pairs on the B side of the slot, and
see if they look deficient in number. (The A side has all the
cap pairs, for the signals coming from the Northbridge towards
the slot.) The B side diff pairs would be returning to the Northbridge,
to give some idea of the direction they would be moving in.
I'll keep an eye out for this, but right now, I suspect at
least some of the entries in that table are caused by BIOS
issues. And that is based on the one Asus board I've heard
of, that wouldn't work with an Areca RAID card, until the
BIOS was updated.
OK, have a look at the Areca compatibility list. This needs
to be updated, but it has one hopeful entry.
http://www.areca.us//support/download/RaidCards/Documents/Hardware/MBCompatibilityList_011606.zip
"HP XW6200*1 INTEL7525 Tested by customer PCI-E X8 Intel Xeon
*1. Update latest BIOS xw9300 to (1.29),xw8200 to (?) and
xw6300 to(?) can solve the problem . Update controller
firmware to 1.39 also solve it."
The XW6200 was in the NI table as only doing x1.
Pages 4 and 5 of that document, cover desktop boards with
SLI slot configs. Even the A8N-SLI family got honorable mention.
On the Asus site, the A8N-SLI Deluxe 1013 2005/08/10 BIOS lists:
"Fixed system cannot detect ARC 12xx Serial ATA RAID Host adapter."
And if you do have a problem with whatever you purchase,
don't forget to post about it
So other people get
properly warned.
Paul
