Welcome, Guest. Please login or register.

Author Topic: This Amiga 4000D has a secret:-D  (Read 11466 times)

Description:

0 Members and 1 Guest are viewing this topic.

Offline tjaoz

  • Full Member
  • ***
  • Join Date: Feb 2003
  • Posts: 165
    • Show all replies
Re: This Amiga 4000D has a secret:-D
« on: July 03, 2004, 03:06:35 PM »
Funny hack! ;-)

It is like driving a car at 250km/h when your car has the E-speed tires (up to 70km/h) :-)

Do not complain with your PCI card seller if this card does not work in this case or is not stable. The PCI specification does not allow passive PCI expanders! E.g. the PCI Address&Data trace lengths from the busboard's (Prometheus, in this case) PCI connector to the pins of the PCI chip on the PCI card (the pins of the Avenger chipset in Voodoo 3) are limited to 1.5 inch!

In case of regular Voodoo 3 PCI card, these traces (Address&Data) have about 1 inch each. If you add about 3 inches of a cable expander, you break the PCI specification by the factor of 2.7! In your case these traces have (with the expander) 4 inches instead of 1.5 inch that is allowed.

In case of the PCI CLK signal, you break the PCI spec by the factor of 30! :-) (You have added 3 inches when the PCI spec allows adding 0.1 inch at the most.)

Here is the relevant part of the PCI Specification:

4.4. Expansion Board Specification
(...)

4.4.3 Physical Requirments
4.4.3.1 Trace Lenght Limits

Trace lengths from the top of the expansion board's edge connector to the PCI device are as follows:

i. The maximum trace lengths for all 32-bit interface signals are limited to 1.5 inches for 32-bit and 64-bit expansion boards. This includes all signals groups except those listed as "System Pins," "Interrupt pins," and "JTAG Pins."

ii. The trace length of the additional signals used in 64-bit expansion are limited to 2 inches on all 64-bit expansion boards.

iii. The trace length for the PCI CLK signal is 2.5 inches +/- 0.1 inches for 32-bit and 64-bit expansion boards and must be routed to only one load.
 

Offline tjaoz

  • Full Member
  • ***
  • Join Date: Feb 2003
  • Posts: 165
    • Show all replies
Re: This Amiga 4000D has a secret:-D
« Reply #1 on: July 03, 2004, 04:48:11 PM »
@redrumloa

Quote
Welcome back Tojaz!
Welcome back Jim!

Quote
It's been a while, I was worried about
you.
Sorry for a brake. During the last weeks I hadn't access to the internet.
I was spending my holiday sailing on the yacht. :-)

Quote
How are things in Krakow?
Well! Nice weather, many interesting thing is happening here.
I just go to see the final concert of the Jewish Culture Festival "Shalom in Szeroka Street".

Bye for now,

Jacek
 

Offline tjaoz

  • Full Member
  • ***
  • Join Date: Feb 2003
  • Posts: 165
    • Show all replies
Re: This Amiga 4000D has a secret:-D
« Reply #2 on: July 03, 2004, 04:48:37 PM »
@Piru,

Sorry for disappoint you, I'm not from Elbox.
 

Offline tjaoz

  • Full Member
  • ***
  • Join Date: Feb 2003
  • Posts: 165
    • Show all replies
Re: This Amiga 4000D has a secret:-D
« Reply #3 on: July 06, 2004, 09:37:05 AM »
@mboehmer_e3b


Quote
Ah. I smell a rat... nice to hear something from Elbox again...

Poor Michael,

You think that anyone who is familiar with electronics is related to Elbox. :-)

Quote
Not correct. The PCI spec 2.1 says:

"5.2.2.: [Planar implementation] Two types of planar implementation are supported by the PCI
expansion card design: ... and expansion connectors mounted on a riser card."

These riser cards are also commonly used in small size systems.

The risers mentioned in the PCI spec are only for motherboards designed to use the 'extender' or 'riser' cards. When you have such a motherboard you can treat the 'riser' (extender) card as the part of this motherboard. The point is that such motherboards have shorter signal lengths. In this kind of systems, the sum of the motherboard and 'riser' card clock length is always 2.5".

It is surely not the case of the Prometheus board.

The attempt to install the Voodoo card in the passive expander will cause violation of input setup times for signals driven by the Voodoo card plugged into the extender. It means that reading from the Voodoo card memory will be disturbed. Without reading the Voodoo card memory the dream about any future support for any PCI busmaster card (sound, TV, Fast Ethernet, USB,...)in Prometheus is ended...

Quote
So. Quoting on PCI specs:

"3.7.4.: [Generating configuration cycles]... Configuration mechanism #2 is defined for backward
compatibility and must not be used by new designs [footnote 21: This mechanism adds a significant
software burden and impacts performance when used in a multiprocessor system]."

There is not such a statement in the current PCI specification.

Quote
Hm. How does Mediator handle this? It uses only the config mechanism #2, which is forbidden according to 2.1 specs for new designs... and what does this mean for SharkPPC cards, which sit behind a bridge (which *needs* mechanism #1, btw)?

Your problem has nothing to do with this tread. Well, apart from the fact that the docs for the Prometheus show that Prometheus does not support config mechanism Type 1, which you named #2.

AFAIK Mediators support both kind of configuration mechanisms described in the PCI spec: Type 0 and Type 1.
But you should rather contact Elbox if you want to learn something more about it.

Quote
So, if quoting, Tjaoz, be fair.

I'm fair. It is you who quoted PCI specs parts not related to this thread.