[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Fw: PCI Clock
Hi Clark,
As you said you found your PCs crashes often, you wrote about
spec-violations could the cause on this.
That could be, but even more the drivers may be buggy. I just came back from
Microsofts WinHEC ( which means Hardware Engineering Conforence ). The
biggest track on this conforence was about driver programming, safe
programming, and driver quality. As you know, there is a crashdump analysis
in Windows XP. This crashdumps were send to microsoft. And MS analyses the
errors with automated tools.
And there are many failures out there, even in mayor brand products. Also in
audiocards drivers for example.
Also this bugs, could be very hard found, while this bugs are only seen,
when the system stress is very high ( system runs out of memory and has to
page-in and out rappitly ), its a nightmare to debug such errors.
So I think you will hardly find an hardware problem, even if there is any.
More then this you may find out an problem between the carddriver and your
system.
As you talked about soundcards and pcispec violations:
Soundcard are used for dataoutput ( from main-memory to PCI-card). And
hazard would cause in corrupted data ( sound would fail, or in PCI-parity
error which causes NMIs ). If your system donīt stall every time by an
bluescreen coming from NMI ( many mainboard bioses ignore PCI-parity
errors... ), this would be a very care condition to crash an system by
audiocards ( Adress of next dma-transfer must be corrupted, to cause an dma
transfer from an wrong adress, which cause an page-fault. Even very,very
rare...... ) It seems you may see such one error in thousend of houres
runtime....
Best regards
Andreas Bergmann
----- Original Message -----
From: "Clark, Raymond J (WBST)" <Raymond.Clark@usa.xerox.com>
To: <JohanHZ@Sycron-IT.com>; "Clark, Raymond J (WBST)"
<Raymond.Clark@usa.xerox.com>
Cc: <pci-sig@znyx.com>
Sent: Thursday, April 11, 2002 7:29 PM
Subject: RE: Fw: PCI Clock
> Without reviewing it, I expect that the PCI-SIG process does something
like
> you propose. They have this all covered. The issue is vendors not
caring.
>
> Microsoft will not care either I would guess, as they need to support any
> and all hardware so they look like the ultimate OS. Although they should
> care, since otherwise THEY get blamed for the crashes.
>
> There probably is not a solution short of customers demanding reliable
> machines, and most customers are too ... um... unsophisticated to demand
> anything. Another driver could be published reviews of compliance or
> stability. But who would fund them? And who would read them? Not the
> masses that buy PCs from the Sunday paper. The only way I can think of
that
> headway might be made would be for a major vendor to decide that this was
a
> good idea and hype it on their box. Others would have to follow suit to
> stay in business. I ASSUME that the PCI-SIG logo is sufficient
technically.
> It could become impossible to gain significant market share without the
> logo. Cease-and-desist suits would have to be funded by PCI-SIG members
> against violators, as well as audits. The way to start would be for
> designed-to-spec cards to make a demonstratable difference and have one
> vendor decide to push the issue in an attempt to take over market share.
> One marginal design in a big market should do it - honest and undeniable
> demos could be created and plastered all over. Open a case of brand X and
> see crashes. In every instance install brand Y and see it work reliably.
>
> A big step. Probably never happen. We are doomed to running on marginal
> machines.
>
> -----Original Message-----
> From: JohanHZ@Sycron-IT.com [mailto:JohanHZ@Sycron-IT.com]
> Sent: Thursday, April 11, 2002 11:34 AM
> To: Raymond.Clark@usa.xerox.com
> Subject: RE: Fw: PCI Clock
>
>
> Ray,
>
> Isn't that the task of PCI-SIG :
> Verifying if a card (or system, component) is PCI compliant?
>
> Why not giving a particular card vendor a "temporary" Vendor-ID for
> developping and testing.
> Once the card has been validated at the PCI-SIG workshops, the card can
have
> its definitive vendor ID.
>
> And why not couple the PCI-SIG compliancy with the Microsoft HCL list (I
> know, it's microsoft, but at least microsoft has a HCL list) : if you
didn't
> pass the workshop, you can't have the microsoft logo.
>
> What's your opinion.
>
> JOHan.
>
>
> > -----Original Message-----
> > From: Clark, Raymond J (WBST) [SMTP:Raymond.Clark@usa.xerox.com]
> > Sent: donderdag 11 april 2002 16:45
> > To: Venkateshwarlu V; dinesh@cosystems.com
> > Cc: pci-sig@znyx.com
> > Subject: RE: Fw: PCI Clock
> >
> > Anecdotal information regarding cheating on the spec.
> >
> > My PC at home had a sound card by a major vendor who was gaining fast on
> > the
> > leader. The PC randomly crashes, some months alot, some less. I had
> > tried
> > everything. Following the discussion on this subject a few months ago,
I
> > pulled out all my cards and found (1) The clock trace was too long, (2)
> > there were almost NO decoupling caps on the board (perhaps 1 or 2).
> >
> > I put in the same model, plug compatable card made by a MORE major
vendor
> > who bought the first vendor about 2-3 years ago. It was a new PWBA
> > design,
> > and minor changes to other components. It had very few decoupling
> > capacitors, but at least you ran out of thumbs counting them (>2). The
> > clock trace is closer to being in spec (but still too long), but there
is
> > a
> > series resistor about 1.5 inches up, and a ground trace paired with it.
> > There has been a 10x reduction in crashing. Given that my "problem"
waxed
> > and waned, I cannot be sure that this was the cause. But I am
suspicious.
> >
> > I wish vendors would follow the spec. Also a few months ago someone
> > posted
> > data on the cost differential for a cheap 4 layer board, I think it was
$2
> > for a small PCI board. Of course decoupling caps are pretty cheap. I
> > would
> > gladly pay $5 or $10 more to get something that works vs. something that
> > makes my system unreliable. That might add up to another $50 for the
> > whole
> > system, a small price to pay. What good is a computer that crashes all
> > the
> > time? Vendors could make it a sales point - "Our boards meet the spec,
> > don't crash your system like those OTHER guys..."
> >
> > My perspective as a frustrated consumer who knows better.
> >
> > --Ray Clark
> >
> >
> > -----Original Message-----
> > From: Venkateshwarlu V [mailto:venkateshwarluv@myw.ltindia.com]
> > Sent: Thursday, April 11, 2002 4:30 AM
> > To: dinesh@cosystems.com
> > Cc: pci-sig@znyx.com
> > Subject: Re: Fw: PCI Clock
> >
> >
> > ** Proprietary **
> >
> > Hi,
> > PCI spec says SKEW should not be more than 1ns.It is required to provide
> > good shielding(guard) for clock to reduce the skew.
> > Series Resistor increases the rise time, inturn affects skew.It's
> > better to provide end termination resistor(equal to characteristic
> > impedance) .
> > Experts comments pls.
> >
> > Thanks,
> > Venkat.V
> > Design Engineer(FPGA),
> > Larsen & Toubro Limited,
> > MYSORE
> > INDIA
> >
> > >>> "dinesh" <dinesh@cosystems.com> 04/11/02 01:56PM >>>
> > Hi,
> >
> > Putting a serice resistor of appropriate value at a distance of 2.5"
may
> > help in handling reflection related issues on clock line while extending
> > it
> > beyond 2.5".
> >
> > Any comments on this point from Exp. Group ?
> >
> > Din
> >
> > ----- Original Message -----
> > From: arvind v <v_arv@lycos.com>
> > To: <pci-sig@znyx.com>
> > Sent: Thursday, April 11, 2002 1:06 PM
> > Subject: PCI Clock
> >
> >
> > > Hi All,
> > >
> > > In one of my designs in Processor pmc form factor, I need to run pci
> > clock(66MHz) to processor for trace length more than 7 inches
(limitation
> > of
> > the design).
> > >
> > > As per PCI Specs , on an expansion board I can run only 2.5 inches.
> > >
> > > What are the implications of this long length , any ideas to over
come?
> > >
> > > Thanks in advance
> > > Arvind.
> > >
> > >
> > > See Dave Matthews Band live or win a signed guitar
> > >
> >
http://r.lycos.com/r/bmgfly_mail_dmb/http://win.ipromotions.com/lycos_0202
> > 01
> >
> > /splash.asp
> > >
> > >
>
>