I just attached a piece of copper wire as a bridge over the DIP switch pins as suggested. Didn't have time for extensive testing yet, but the problem persists or got even worse.
On the FIC SD11, the 600 MHz setting results in 300 MHz. On the EP-7KXA, I couldn't find any combination of settings (multiplier and VCore) that leads to a successful POST. The system would lock up during POST, most of the time showing a C1 error (memory), but sometimes other error codes as well.

I'll go through the updated schematics again and see which other pins might need to be bridged...

Weird. Unless there are hidden traces under the switches on the front of the original.

Anyhow, here's the Ninjamicro Freespeed schematic. I just have to route the PCB.

@wiretap you wanted the diode information: they are both forward bias with a voltage drop of approx 0.7v using my meters diode testing mode.

The positive (annode) is upper side (closest to the ‘solder side’ screen printing) on both of them.

Will you use through hole for the components on the ninja remake? It will make assembly, for me at least, much easier

I have now tried pulling pins 7, 29 and 33 to GND with jumper wires as well, but to no avail. Neither board will POST in any of the switch positions I tried.
Still, no hardware destroyed due to shorts etc. 😁
I still don't have a 2.1k Ohms resistor at hand but the 2.2k is still in place, but that's a mere 4% difference, I can't really imagine that this would make the difference between a working and a completely non-working device.
I'd really love to make the Northwind clone work... any ideas are appreciated. I'll go through the schematics once again myself...

Regarding the Ninjamicro clone: Through-hole components would probably make it much easier to assemble for most of us, so I'd prefer that as well.

I am about 90% done with the Ninja. It uses 1206 size SMD, so it is just about as easy as through-hole stuff. I wanted to make it in the exact look/layout. I'll post the KiCAD stuff tomorrow. It could easily be converted to through hole. Maybe I'll do two versions.

About the Northwind. Yea there must be other traces under the lower DIP to the J1 header to a different pin than ground.

Ninjamicros Freespeed Pro -- KiCAD, Gerbers, and BOM attached. Please review everything before using on your system to make sure it matches up with the original. Keep in mind I'm doing these off of pictures 🤣. Despite being mostly surface-mount, you can see that these are 1206 size SMD components. The pads are as large as a standard through-hole component and it is just as easy to solder. If you want dead simple -- order from JLCPCB with the stencil (sometimes free add-on to your order), then just use solder paste, place the components, and heat it up with a hot air pencil to flow the solder. Example video: https://youtu.be/sOZ_Bry957k?t=114

Mouser Cart: https://www.mouser.com/ProjectManager/Project … ssID=c42aa0e678

Schematic:

Traces:

3D View:
(Ignore the 3d-rendered switches that show upside down in this model. Mount them per the original design.)

I think it looks good comparing the traces to the actual one. The only area I worry about is attached. I can’t tell if you have all those pins attached it should only be the two you see with the multimeter.

On solder side SW2 bottom 3rd in from right only goes top left most on connector

Edit: nevermind I looked at the gerbers and it looks ok

This design I feel confident in because I checked it over along with the one I have. I’ve ordered some boards to test.

I haven’t given up on the free speed design and in fact prefer that one so would like to get that working.

I have some parts still arriving so I’ll keep posting in this thread.

It's funny you mentioned that pin.. when I was reviewing it in KiCAD this morning, I looked at it as well.

I am currently looking into the SMT assembly service at JLCPCB and they don't have the exact diode FDLL4148 available, but only FDLL4150 which is similar, but not identical.
Would that one work as well?

Yea it should work. However, I did not design the board for SMT assembly so they will probably reject it. I believe it would need to have tooling mount holes and abide by a few other standards that aren't met.

gex85 wrote on 2021-03-04, 10:34:

I am currently looking into the SMT assembly service at JLCPCB and they don't have the exact diode FDLL4148 available, but only FDLL4150 which is similar, but not identical.
Would that one work as well?

We think alike! I used a Semtech LL4148.

I did have to adjust the placement of the parts. I believe they all had to be rotated 90 degrees from their current x/y coordinates.

My order is in processing, the PCBs at least, no errors so far for the pick and place, I’ll see once it gets to that stage.

I actually got around to making a very minimal design based on the Innovatek Goldfinger. Sadly I damaged the original in the process. 🙁
Due to its small size (36*25mm) I dubbed my design the Golden Pinkie. It only requires the PCB, the edge connector and a single row of DIP switches while offering 1.30V up to 2.05V voltage settings in 0.5V increments and supposedly 3x-10.5x multipliers in .5x steps. I have never tried the lower multipliers with the original though, so no gurantees.
I am going to order a prototype batch and test it before releasing the gerber files.
I'm a little excited since I've never made a PCB before but if we all cross our (golden 😉 ) fingers it may actually turn out to be functional. Wish me luck! 😁
Until then here's a 3D rendering. (In which the switches are just a grey block for some reason.)

Awesome. 😀 If you need any help with it, let me know. I think that design could be shrunk down even smaller. :p

wiretap wrote on 2021-03-05, 11:46:

Awesome. 😀 If you need any help with it, let me know. I think that design could be shrunk down even smaller. :p

Thank you! 😃👍 Originally I had planned on taking some good pictures of the PCB with the components desoldered and leaving it up to the pros to make a clone but then I decided that I could use a little success right now and this would be a great and easy project to learn some KiCad basics.
I also thought about shrinking it down further but I'll wait and see if it works at all and how much clearance is left on CPUs that just have a bit of the case cut out for the GFD instead of the whole back removed.
If the footprint outlines are correct there's still a 3mm gap between the DIP switches and the edge connector that could be removed. There's plenty room left for routing the traces and I suspect the switches I'll be using are even smaller than the footprint outline so there's definitively potential for an even smaller final revision. 😁
Idea: Maybe somebody could design a replacement cover for Slot A Athlons that factors in the final (maximally shrunken) Golden Pinkie and adds some extra support and gives the whole assembly a flush look?

Oh boy, after all the reverse engineering is done guess what I found while looking for those cache OC tools on my HDD:
A complete build guide for a GFD with scematics(!!!) that I saved from Tom's Hardware a long time ago.
Yep. Sorry.

Apparently that's not even a hack but a design made by AMD. The notes say the 56Ohm resistors should be changed to 220Ohm like on the FreeSpeed Pro for enhanced longevity. They also suggest using 5V and two 1N4001 diodes to generate 3.2V instead of using a real 3.3V source, again similar to the FreeSpeed Pro as far as I can tell.

For copyright reasons I don't think I can just post the whole article here but I guess this picture of the schematic is fine (i.e. covered by fair use). If not, please let me know mods, and I'll make my own version.

I've found the cache tools too btw. There's one for editing an ASUS K7V's BIOS to permanently use a fixed cache divider, another tool to set the cache divider in Windows (don't know if it'll work under NT) and one to set it in DOS. I'll check wether they're freeware and if they are I'll post them later.

Hit here:
https://www.tomshardware.com/uk/reviews/make- … vice,142-2.html

That's awesome. 😁 I was looking for that specific article, but it wasn't on the Wayback Machine. But it looks like the UK site still has it. What needs to be done is to make a pinout table for mapping the AMD CPU finger header to the TE 1734037-4 slot connector footprint in KiCAD. That should only take a few minutes to generate.

PARKE wrote on 2021-03-05, 14:37:

Hit here:
https://www.tomshardware.com/uk/reviews/make- … vice,142-2.html

The German version of the article improves on that one by explaining how to get 3.2V with two diodes and suggests the use of 220Ohm resistors instead of 56Ohm as AMD suggests but apart from that it's mostly the same. I just can't find it online anymore.

Funny how that info was avaliable for 20+ years and nobody has made an open source design until recently.😅

The diode and 220-ohm note is on page 4 at the bottom of the UK article.

I guess we can easily make a nice design now with the proper AMD schematic that better explains some of the unknowns about the AMD side of what connects to what.