Been thinking about the possibility to "replace" the AGP slot with PCIe using a PCI-PCIe bridge (e.g. PEX8112) in reverse mode, considering 440BX's AGP uses 3.3V which the bridge chip can just use it.

The questions would be
- Whether 440BX's AGP bus can run reliably operate at overclocked rates like 89MHz, 100MHz or even 133MHz.
- How to provide adequate power for cards that draw a bit more but without a dedicated power connector (usually those between 25W-50W range).
- Whether the PCI(X)-PCIe bridge in reverse mode can handle overclocked PCI host (higher than 66MHz). Couldn't find any info from datasheets but in this thread some tests done by sdz might provide some clues.

If 440BX's AGP bus can really operate at overclocked rates without issues, and that PCI(X)-PCIe bridges in reverse mode won't mind, then the issue with AGP when using FSB133 CPU with 440BX would become irrelevant if the clock generator can be configured for 133/33.

He said somewhere it worked up to at least 95mhz.
I have thought that a potential board should have a 1x slot between the agp and board edge, that's directly wired to a 16x slot(1x actual)

This would allow someone to build an agp to pcie adapter, on one custom, 2 finger board.

Ie, it takes agp from the agp slot, feeds it into the 1x slot, and it "pops out" on the 16x slot. Alternatively, many of the mining risers just used a usb cable, so something that crude is possible too.

As far as I've looked, there's no reason in the specification that the chipset must supply the clock. Even Intel's drawings show a 3rd source, and later chipsets don't supply it.
I'm guessing they were just being cautious, in the early days ensuring that the clock was of the right "shape" and timing.

I'm pretty sure that if someone modded a bx with a separate clock as we discussed a page or so ago, it'll work.

See pic

As for power.. Well, if you're building a board, you'll be making sure a pcie slot can provide... ?75w? And agp whatever the highest was

myne wrote on 2024-12-02, 13:58:

As for power.. Well, if you're building a board, you'll be making sure a pcie slot can provide... ?75w? And agp whatever the highest was

It seems standard AGP is rated at 25W just like PCI. There was an extension to AGP called AGP Pro which could be found on some boards, that had additional connections to allow drawing up to 50W/110W.

EDIT: From the AGP Pro specification, it seems there were no signals in the pins added by AGP Pro -- just VCC3.3, VCC12 and GND. So nothing's required on the chipset side to implement this.

PCIe slot itself allows upt to 75W, it's given by numbers of assigned power pins. you just route 12V and 3V3 from ATX PSU to slot. For more you need to use extra power connector(s) on PCIe VGA...
In age of AGP cards most of them had power consumption below AGP slot limit...

I think I found first real bug in Intel reference design 😀 Q5 Q6 Q8 are marked 2N7002 but use 2N7000 pinout.
Pages 12 13 PIIX4E, symbols PIIX4E, 74F07, 74HC10, 74LVC14, 74HC112, 2N7000.

RayeR wrote on 2024-12-02, 17:50:

PCIe slot itself allows upt to 75W, it's given by numbers of assigned power pins. you just route 12V and 3V3 from ATX PSU to slot. For more you need to use extra power connector(s) on PCIe VGA...
In age of AGP cards most of them had power consumption below AGP slot limit...

PCI and standard AGP were only rated 25W. AGP Pro existed in a few boards to allow extra power than standard AGP, but it eventually became irrelevant as video cards requiring more power included dedicated Molex or 12V connectors instead.

Still, there existed a few PCI video cards (actually PCIe GPUs behind a reverse PCI-PCIe bridge onboard) with TDP rated above 25W, so it's possible this 25W is merely a rated value and how much the bus can really handle depends on external factors, like board design (regarding power lines), and perhaps the presence of a card with its own power connector (e.g. some X-Fi models like SB0550, SB0460).

EDIT: The question was mainly about concerns I had with use cases of existing PCI-PCIe reverse adapters for plugging PCIe cards to PCI slots. Those adapters usually do not have an external power connector, and PCIe video cards rated between 25W-50W (which is above PCI power rating while below PCIe power rating) do not have power connectors either. Yet the video card in question does work okay on my system for the time being.

For new designs, this question may not be as relevant with proper power layout.

Either way, there's no point delivering more than any real would board delivered.
Any card that needed more was built with power plugs.

rasz_pl wrote on 2024-12-02, 20:53:

I think I found first real bug in Intel reference design 😀 Q5 Q6 Q8 are marked 2N7002 but use 2N7000 pinout.
Pages 12 13 PIIX4E, symbols PIIX4E, 74F07, 74HC10, 74LVC14, 74HC112, 2N7000.

Interesting. Mistake or trap?
Historically, map makers would put in fake roads or towns to find copiers.

Reference design is the stuff you give your clients, you want it copied. Mistake is in suspend circuit and even the intel-se440bx-seattle didnt implement it like in this diagram.

https://www.alldatasheet.com/html-pdf/170639/ … 3/1/2N7000.html

That shows them both on the same datasheet

Yes. If you look closely 2N7002 is a SOT23 package while 2N7000 is TO92. Someone at Intel must have put 2N7000 at first, but then most likely someone else realized they should spec SMD part and just swapped part number without correcting diagram symbol 😀 Gate-Source swapped around.

Yes, quite a common mistake. I was 'happy' to make a such bug when replaced miniHDMI connector by big HDMI. It looks the same single row of pads so i just swapped it 1:1 but the pinout is different! The biggest fun was that we got image on monitor so nobody noticed a problem for several months. Then someone tried 4k output and it didn't work on _some monitors. We all wonder why. It just run via SE instead of diff pair. Fixing a bad transistor package is much easy but depends on how many boards 😀

Another mistake, first PCI pin B14 (Reserved) pulled hard to ground. Reading PCI spec board build like this violates PCI spec and loses its PCI certification 😀

Done pages 15 agp, 16 17 PCI. PCI, AGP, 74AS07 symbols.

I did see some confusing pins in the p2b boardview.

That, and the fact I couldn't at the time get component values was why I did this.

I found them in the bom file. I'm slowly working on a powershell script that will hopefully turn a fabmaster asc boardview +Asus bom backup into a pcb.
So far the ASCs are all in tables. The bom is a bit more quirky.

Then it's theoretically straightforward task of writing all the table data into kicad files. I already tested it with the pcb edge, and that worked.
I'm sure there will be problems though. Writing 900 footprints and aligning them is bound to be problematic.

In theory it will work. I tested the pcb edge and that worked - albeit kicad and asc work differently. Asc uses 5 points to make a rectangle, but kicad uses 2.

Still, I think the reference is important even if it's fairly unique. The script will never write a good schematic.

If anyone's interested in the progress of that script.
It currently works on "ASUS_Motherboard_P2B_1.12" and gets as far as a bunch of mostly correct tables.

I'll leave it to you to find the files.

You dont want to upgrade Kicad and I have disabled PowerShell (for security reasons) 😁 Will try on another computer later.

Heh.
There's nothing special about it yet.
It's not even at the point where I'd call it a useful list of CSVs.
It's close to that. But not quite.

This:

1                             華碩電腦-備份資料                   
2                                                                     FROM:P2225
3                               單階材料用量明細表                      
4
5製表日期:02/11/28 17:07:46   有效日期:02/11/28  列印版本:               頁次:1
6================================================================================
7主件料號:60-M6P260-17         目前版本:   來源:M 發料單位:PCS  生產單位:PCS 
8品名規格:P2B/URM2/ICS                            主件數量:1
9          R1.12                         
10 項                       燒 來 發料  組成用量   生效日期 圖號         特 保 Dis
11 次  元件料號/品名規格       源 單位    底數     失效日期 插件位置/ECN 性 稅 app
12---- ----------------------- -- ---- ----------- -------- ----------------------
13  10 02-010137154            P  PCS       1.0000 99/03/25                  N  Y
14     INTEL FW82371EB PIIX4E IC                           
15     F-II A0 SL2MY                                        U9,                 
16  20 02-010144356            P  PCS       1.0000 99/03/25               S  Y  N
17     INTEL FW82443BX       IC                             ECN:PE-8-2-T053
18     440BX C1 SL2VH                                       U7,                 
19     MARKING:INTEL/AGPset/FW82443Bx/xxxxxxxx/SL2VH/INTELxxxxxx/xxxxx
20     QVL:INTEL,
21    (02-010144326        )                1.0000 99/03/10                     Y
22     INTEL FW82443BX       IC                            
23     440BX    SL2T5                 
24  30 02-230278100            P  PCS       1.0000 99/03/25                  Y  N
25     WINBOND W83781D  PQFP48                             
26     H/W STATUS MONITORING IC                             U6,                 
27     MARKING:W/W83781D/xxxxx
28     QVL:WINBOND,
29  40 02-230397721            P  PCS       1.0000 00/01/18               S  N  N
30     WINBOND W83977EF-AW P128                            
31     VERSION E                                            U4,                 
32     MARKING:WINBOND/W83977EF-AW/xxxxxxxxxxxxxx/xxxxxExxxxxxxxxxx
33     QVL:華邦,
34    (02-230397714        )                1.0000 00/01/18                     N
35     WINBOND W83977TF-AW P128                            
36     VERSION D                      
37     MARKING:WINBOND/W83977TF-AW/../xxxxD..
38     QVL:華邦,

Continued for several pages with footers

To this:

1UnitID ListNumber ComponentType                  UnitDetails                   
2------ ---------- -------------                  -----------                   
3U9     10         INTEL FW82371EB PIIX4E IC      F-II A0 SL2MY                 
4U7     20         INTEL FW82443BX       IC       440BX C1 SL2VH                
5U6     30         WINBOND W83781D  PQFP48        H/W STATUS MONITORING IC      
6U4     40         WINBOND W83977EF-AW P128       VERSION E   

Was not easy.
I'm no powershell guru, so I'm sure someone can do it all in 3 lines, but I couldn't. I'd probably do some of it differently now I know how, but whatever. It's fast enough.
Onward.

Ah.
Boardviews contain pad and net info, should be enough for https://forum.kicad.info/t/how-to-import-a-ki … schema/13178/10
Painful output, but still better than doing everything from scratch 😀

Getting closer!

100% generated from the files.

Buggy, but enough for one day.