Conference noted::ibmpc-95

>	(b) more likely simply divide the 75MHz clock by 3
>		to get a 25Mhz PCI bus.  The PCI spec allows
>		any clock rate from 25 to 33 MHz.

Actually, any clock rate from 0 to 33 MHz.  (Early versions of the spec had
a lower limit, long gone.)


Cyrix's web site does indeed say this (in bold text) about FIC's PA2005
motherboard:

	NOTE: The PCI bus runs "out of spec" (37.5 MHz) for the PR200+.
	Contact FIC for the recommended list of PCI peripherals.

Well, it's really not much different than pushing CPU clock speeds, is it?

I agree with Bill Lin (in .27) that 37.5 vs. 33 MHz is not a small change. 
Some of the PCI signals allow only 4 nanoseconds for interconnect delay,
and shaving 3.33 ns off of that is no small peanuts.  Yeah, maybe it'll
work ... or maybe not.

See note 2891.7 for a reply that probably was intended for this note
thread.

I'm baffled by their implication that older cards don't support the
37.5 MHz bus frequency but newer PCI cards do support it (as if they
were designed for it or something).

If they support 37.5 MHz at all, it must be by sheer luck, since PCI
is never supposed to run that fast!

    These clowns are driving the PCI bus at 37.5 Mhz instead of the
    standard 33 Mhz.  They are covering their ass with this BS by falsly
    claiming that only "old" PCI devices can't handle their bus.
    
    The PCI spec is very clear on this.  These folks are full of it.
    
    The issue is that these folks want to run the Cyrix x86 clone faster
    and that Cyrix chipset only supports certain divisors for the processor
    clock to PCI clock ratio.
    
    The 66 Mhz option of PCI does not come into play here because:
    
    (1) almost no 66 Mhz options exist and a card specifically has to
    indicate support of 66 Mhz,
    
    (2) 66 Mhz requires 3.3 volt signalling instead of the more usual 5
    volt (per PCI spec) and many interesting PCI option cards are currently
    5 volt only and building a 3.3 volt system would eliminate them from
    your system (universal options would work in either voltage),
    
    (3) when running in a 66 Mhz system, clock frequencies between 66 and
    33 Mhz are special in that the clock may not change except in
    conjunction with a PCI reset (the assumption being that slowing the
    clock all the time is done for power reasons, 66Mhz systems are not as
    concerned with power, and PLL technology may be required to make 66 Mhz
    work but PLLs don't work with changing clock frequencies).
    
    (4) A 66 Mhz system starts out running at 33 Mhz (or lower) until
    software/BIOS/whatever has decided to shift into high gear. Software
    does this after verifying that all interested PCI devices are 66 Mhz
    capable. It then sets the 66 Mhz enable bit in all the appropriate PCI
    devices and switches the clock.  The Cyrix stuff can't run slower than
    37.5 Mhz (at least not in software) so it can't do this.
    
    (5) when scaling up to 66 Mhz, the PCI spec writers changed the timing
    budget around.  It's not just a simple divide (multiply) by two.  This
    was needed to make things (barely) doable.
    
    Steveg

>     If you use an older type peripheral that doesn't support the 
>     37.5MHz bus frequency you may not get optimum performance from 
>     it.

	What a lovely way of putting a positive spin on "If you use an older
	type peripheral that doesn't support the 37.5MHz bus frequency you
	may experience random hangs, crashes, and possible data corruption."

Larry

You may want to take a looking at http://sysdoc.pair.com/cpu.html.  There is
quite a discussion of overclocking the PCI bus, what works and doesn't work,
including a survey of people that have tried it.  One of the great comments in
the article is that the author has finally found a use for the 150 Mhz pentium,
i.e., setting the external clock to 75Mhz and jumpering for 2X.  Anyway, a good
read about the results of overclocking the PCI.

ac

    I read .37 twice and I couldn't come up with any other
    thought other than what kind of snake oil are these guys
    trying to propagate. Their answer seems like such utter
    denial and spin on a situation and shows such an arrogance
    that I'd never want to do business with them, that's for sure!
    
    Bob

    
    Change of plans:
    
    I'm buying an AMD K5/100 because they are cheap.
    this will hold me over until the Cyrix MMX chips are
    out in the next 6 months. 
    
    Now to decide on the PA2005, also cheap, supports
    what I want, with a slight twist which is solved by
    buying the right cards.
    
    But can I live with 3 ISA slots? Ah, that is the question.
    
    Doug.
    

    Didn't Intel just dropped the prices?  Pentium 133 is now $120 or
    something.
    
    - Vikas

    
    That's still twice the price of the AMD, and why pay extra for a chip
    I'll be replacing in 6 months?
    
    The deed is done. Board, chip and video card will arrive next week.
    
    It will still be a boost over the the 486-120, but I'm really iching
    for that 166/200 ....
    
    Doug.

    Does the motherboard take 5.0V supply?  I have old 486-33, so I am sure
    my system does not provide 3.3 PS.
    
    How much is the board+chip?
    
    thanks,
    - Vikas

    
    Installed and running.
    
    Comments on the PA2005:
    
    You must by it with the size L2 cache you want. It 
    is not upgradeable. I could not find a vender that
    carried boards with anything but 256K of L2.
    
    1 ISA slot is blocked by the cpu so mid and full size
    ISA cards won't fit in that slot.
    
    Other than that, It seems to be well built unit.
    Well worth the $99.
    
    Doug.

  >  These clowns are driving the PCI bus at 37.5 Mhz instead of the
  >  standard 33 Mhz.  They are covering their ass with this BS by falsly
  >  claiming that only "old" PCI devices can't handle their bus.
  >  
  >  The PCI spec is very clear on this.  These folks are full of it.
  >  
  >  The issue is that these folks want to run the Cyrix x86 clone faster
  >  and that Cyrix chipset only supports certain divisors for the processor
  >  clock to PCI clock ratio.
 
  Apparently, ASUS has a few clowns working for them as well. Their board
  will run 37.5 and 41.5MHz PCI buss speeds (undocumented).

  The new PCI spec calls for PCI buss speeds right up to 66mhz.

  Doug.

>  The new PCI spec calls for PCI buss speeds right up to 66mhz.

Well, yes, but it's a "different" PCI bus.  66 MHz PCI goes by a completely
different set of rules.

That spec has been official for 20 months now, but from what I've read,
nobody has yet to make either a 66 MHz PCI card, or a motherboard with a
*legit* 66 MHz slot.  For that matter, nobody in the PC space seems to be
doing 3.3V PCI busses either, which is a prerequisite for 66 MHz PCI.

    Umm, the previous comment may need a bit of clarification, as "3.3V"
    has two different meanings on the PCI bus.
    
    The first is 3.3V power.  The PCI bus specifies +12V, +5V, and +3.3V. 
    These are all assigned to specific, unique pins.  Since most PCI cards
    don't require +3.3V power, a number of systems "cost engineered" this
    out (there are times I think I'm responsible for the first PCI
    peripheral on the planet using +3.3V!).
    
    The second is 3.3V logic for signalling.  The PCI spec calls out both
    5V and 3.3V logic -- with different pinouts and different connector
    keys.  The newer components -- basically everything build in .35 micron
    CMOS -- are all 3.3V (or less).  Thus, we are starting to see a lot of
    3.3V parts on PCI cards.  Most of these 3.3V parts use 5V signalling,
    but some are beginning to use 3.3V signalling.  I haven't seen any PCI
    cards (yet) using the 3.3V signalling.
    
    I need to double check, but I believe the 66 MHz PCI bus is just that
    -- 66 MHz; not "up to" 66 MHz.  66 MHz is extremely fast for an I/O
    bus, and requires a lot of signal integrity work, very short total bus
    lengths, and small numbers of connections to the bus.
    
    Probably the first widespread use will be with AGP, which is
    effectively an extended special purpose 66 MHz PCI derivative graphics
    bus.  The AGP is single slot, and doesn't use standard PCI connectors. 
    It is also a direct point to point connection onto the system/memory
    bus.

>    I need to double check, but I believe the 66 MHz PCI bus is just that
>    -- 66 MHz; not "up to" 66 MHz.  
    
    It's been awhile, but I believe the 2.1 spec defines this as "0 to
    66mhz" to allow for power-saving/"sleep" modes, etc...
    
    /dave

Thanks for clarifying my note.

Indeed, many people don't realize this, but there are two completely
different, *incompatible* PCI busses defined by the PCI spec, which differ
according to the signal levels they use.

The "5V signaling" PCI bus uses essentially TTL levels, as have been used
by both TTL and most CMOS chips for decades.  So far, this bus has 100% of
the PCI marketplace in PCs.

The "3.3V signaling" PCI bus uses levels that are tailored for 3.3V CMOS. 
They are similar to the TTL levels, but all the thresholds, min. and max.
levels, etc. differ.  One very big difference is that signals on a "3.3V
signaling" PCI bus MUST be clamped above 3.3V and CANNOT be driven to 5V.

All PCI busses in PCs today that I know of, use "5V signaling" levels, EVEN
IF THEY EMPLOY 3.3V CMOS chips.  Such chips are designed to tolerate 5V
levels even though they themselves may be powered by 3.3V.

>                                                   I haven't seen any PCI
>   cards (yet) using the 3.3V signalling.

Actually, there are cards (note I said cards, not busses) that can use the
"3.3V signaling".  There are so-called "universal" PCI cards that can be
plugged into either kind of PCI bus.  Digital makes some.

66 MHz PCI uses the "3.3V signaling" PCI bus, which nobody does yet.

66 MHz PCI is pretty much limited to a maximum of one slot.  You won't find
motherboards with three 66 MHz PCI connectors, unless each one of them sits
behind its own bridge.  This is because of signal integrity and timing.

The 66 MHz bus spec covers anything faster than 33 MHz.  A 66 MHz bus isn't
required to operate at 66 MHz (it can be run anywhere from 0 to 66 MHz, the
timing characteristics change at 33 MHz); but a PCI card designed for the
66 MHz bus specs must be capable of working that fast should it be plugged
into one.

    I've read that K6 chip from AMD is pin compatible with the Pentium. Can
    I just buy a K6 chip and drop it in to my motherboard? Any issues
    with BIOS or DC Voltage?
    
    Of course, there's pin compatible and almost pin compatible.  
    
    Gim
    

    The K6 is supposed to be Socket-7 compatible and will be using a lower
    core supply voltage.  So, to a first order, any board that will take an
    intel MMX *should* take the K6.  BUT, nobody is yet advertising their 
    boards as K6 compatible.  The K6 has been sampling since November, so 
    I guess there will be more board info when the chip starts shipping,
    supposedy later this month.

    /j