I only suggested that you edit your 'store' comment. If you do not want to edit your comment, perhaps because you feel it was an attack on your authority, character, or whatever, so let it be. I feel that I have done my job in pointing it out, and that not doing so could imply agreement by complacency. Based on the comments from others thus far, I do not feel I am alone in my original observation. I've also had a private message confirming my original thought, stating that galanopu's comment seemed rather 'slimy' and out of place.

Nobody is or was intending to attack your character or sense of self worth; it is pretty clear based on the point-by-point tries at vindication that this is how you have taken the comment. If I see something out of place in the context of the forum's dialogue, which I felt your 'store' comment was, I will point it out. I firmly believe that your post was an innuendo for advertising your website and was placed in poor taste with the rest of the thread. I have always been in strong agreement with the forum rules that any form of advertising is banned, however there seems to be some quiet acceptance of open sourced projects, or projects originating from the forum.

I've tried my best to formulate my sentences in the least combative approach while still conveying the intended message, and if I have failed in this regard, I apoligise. If there has been any misunderstanding, I deeply apologise for that disruption as well.

Now that you have tested your unit, please feel free to share your Vin and Vout noise levels with a 100 mV voltage scale and 100 us time scale. For a closer comparison, could you run the SXL at 40 MHz in 1x mode, sitting at the DOS prompt (not at or before POST, nor while booting), and with the cyrix.exe flags, -e -i1 -f -r and -m- but not -cd. This will provide a more telling comparison.

I say 1x/40 MHz only, because after clock doubling, noise gets cleaned up a smidge. From the three SXL2-66 chips I have tested, they would run OK at 66 MHz and 3.3 V on the replica interposer that sphere has so generously worked on for the past few months. I did not try lower voltages. 75 MHz needs 3.6 V. You will want to, at minimum, let Quake run in look for an hour, and ensure doom completes. I let Quake run for 3 hours because I walked away and forgot about it.

80 MHz gets more interesting and depends on the CPU. I have one which will do 80 Mhz at 3.75 V, while others need 4.0 V. At 90 MHz, going up to 4.6 V still did not help. If you have a device which fits into the DIP-14 machine pin socket on 386 boards, and allows for 1 Mhz increment adjustment in the 80-90 Mhz range, we may be able to further optimise the operating frequency.

Sphere, I recall that I routed the trimmer so that a clockwise turn increases voltage, not decreases it. I have a few motherboards I modded and didn't think carefully about the direction and am always frustrated having to turn counter-clockwise to increase the voltage. I don't recommend swapping the turn direction.

I have tested with 8x caps in the centre and showed it can improve the noise some. Yes, use the 1210 MLCC size. If you just cannot make 8 caps fit, do 4 or 6. I spent some hours playing with combinations yesterday, but kept coming back to my original configuration. But it will not be detrimental to use 4 or 6 centre caps. Yes, make Cin1 1210. if you want more space, 1206, but it is the same length as 1210, just less deep. Alternately, I showed previously that there was some benefit in using 10 uF at the PGA pins, but I did not try removing all the 100 nF caps and replacing them with 10 uF.

Yesterday, I tried one more time with all 21 PGA caps, but system didn't boot and I got a very curious waveform on the scope. Went back to 8 and it was fine.

Just let it go guys lets move on future people trying to read this in 2o years are going to have to read through that..

okay, damn I was hoping you were going to say I didn't have to put the center back

I'm considering that the new design is electrically different than the old design. it may be that the center is no longer needed and we just don't know it.

that said I am going to see if there is a easy way that I can add it back, but I don't think it will be a organized square like last time. (sorry)

good news is that my flood and mask idea worked. the pad shapes are only on bottom now. rest are round.

I gotta run I'm half way through redoing the trim pot and capacitor. here is my place save. I circled the work area. 🤣

Using 1206 in the centre might help if you are tight for space.

Yeah I did 80Mhz at 3.6V
90MHz crushes once I enable clock doubling even with 4.1V
I suppose I will know more once I get the sockets and more CPUs.
But this sounds in line with your observations.

PS: Let me just say that I have zero need to directly advertise here, so I do not do it.
I did not do it also here as this is not available yet, and I did not post a product link.
In the end I do not even know if I will sell this, it really needs so much time to assemble.
I have my Youtube, anyone who checks my final results will also see if he can get or not and how.
Finally the reality is that there are many commercial and highly praised by the community projects here.

And yeah you just it right this felt like and was a personal attack.
Looks like the OS community acts like a cult sometimes.
The truth is I posted this a bit in a hurry and did not payed a lot attention to every word.
That why the short message too. But I really hate to be told what to say and to self-censor. So yeah.

feipoa wrote on 2022-12-09, 02:34:

Using 1206 in the centre might help if you are tight for space.

I had a thought, the pins on the backside of the PGA 168 socket are so easy to use for capacitors what if you used different value capacitors there could you get the same benefit as the center section with different values using those pads? can you do some experiments?

That would allow me to keep the central ground plane intact and we need that more than ever now with the shrink. So I really don’t want to disturb the middle unless we have to.

I know it’s going to mess up your symmetry and kill the cool factor of the center caps and symmetrical decoupling caps but looks really are not as important as making the best electrical device we can, so see if you can accomplish the same effect with different value capacitors on the PGA 168 pins and let me know please. Or have you tested this? (Hard to remember what tests you have run.)

The ceramic for cin will be great actually. It will allow me to pull that trim pot over. 😀

Sphere, I was ready to call it quits with testing. We already have a product with performance that matches that of a commercial unit and is smaller. I share your desire for maximum optimisation, but there is a point where you got to say good enough. I can do one last test with 8x 10 uF 0805 caps on the PGA pins, but I don't think it will clean up Vcc5 like the central caps do (albeit small).

Sphere, attached are some results. The original configuration is still slightly better.

Here is the scope image with 8x 100 nF caps on PGA pins, plus 8x central caps (10 uF, 47 nF, 220 nF, 10 uF). Screenshot shows 48 mV, but average is around 52 mV.

It became too time consuming to adjust the cursor for the hundreds of photos taken, so I started to rely on the Vpp value, averaging, and the waveform view. Then I try to pause data acquisition when the Vpp value matches the average I manually tabulated. Ignore the Vrms value - at this 100 mV scale and the offset zero position, my crappy scope messes up. It also messes up Vmax, Vmin when I pause the run. lol. However, the paused Vpp value agrees with running Vpp and with manual cursor measurements.

Now I remove the 8x100 nF caps and replace them with 8x10uF caps, but leave the central caps in place. Vout average was around 56 mV.

Last, I remove the central caps, but leave the 8x10 uF caps in place. Sceenshot shows 64 mV, but average was closer to 60 mV.

It is kind of splitting hairs at this point, but look at the waveforms and judge for yourself to the extent that you can. I still feel that the original configuration is the best. If possible, keep the central caps. To save space, maybe 1206 size? I recall pentiumspeed saying that the physically larger MLCC caps would be more beneficial? If you are intent on removing the central caps, I will need to test the 8x 10 uF PGA caps with the MIC regulator as well. I have only tested with LP here. You will recall that swapping the reg is a little problematic.

I also wanted to point out that those periodic looking spikes in the waveforms tend to be caused by the a setting which disables caching the first 64 KB of each 1 MB boundary (-m- setting). Some motherboards need this set for it to work properly in Windows, particularly with a SCSI card. I think the AHA-154X was the most problematic in this regard. This is why I mention that if testing and wanting to compare apples to apples, to use the same cyrix settings that I am. cyrix.exe -e -i1 -i2 -i3 -i4 -f -r and -m-

Okay, I’ll try a few things but if we can avoid doing the center again we should because there aren’t enough clear spaces devoid of traces to do them by vias. And running tracks through the now present full ground plane there is something we should try and avoid especially now that we basically lost the outer ground ring around the outside of the socket.

There may be a way to do it without much disturbance. Maybe using latge traces to pins.

I was curious if the remaining 13 spaces on the pins could be used to replace the benefit you saw in middle

So we really only need more filtering on vcc5 if I am understanding? We don’t need anything on vcc3 in middle? This may help me make something work.

Sphere478 wrote on 2022-12-09, 02:31:

I gotta run I'm half way through redoing the trim pot and capacitor. here is my place save. I circled the work area. 🤣

-you can totally remove that pcb notch near trimmer and everything will fit just fine, trimmer case can slightly overhang
-according to https://jlcpcb.com/help/newsdetail/42-6-Layer … on-PCBs-for-$20 "$2 limited-time special offer before 5th January 2023, that the board within 5cm*5cm (5pcs with ENIG and via-in-pad) will be reduced from over $100 to a special price of $2." you can do via in pad for free on jlcpcb now, might hurt solderability of Cio Cout, but better electrically
-decoupling cap pads are truly symbolic, they dont do anything for soldering caps there 😀 they dont extend at all. G16 L17 dont have clearance and short to ground

I just have to ask - whats with your aversion to clearance? :0)

galanopu wrote on 2022-12-09, 03:16:

90MHz crushes once I enable clock doubling even with 4.1V

worse than Sphere478 one? weird

galanopu wrote on 2022-12-09, 03:16:

PS: Let me just say that I have zero need to directly advertise here, so I do not do it.

yet you did

galanopu wrote on 2022-12-09, 03:16:

In the end I do not even know if I will sell this

you knew two days ago "and then to the store"

galanopu wrote on 2022-12-09, 03:16:

And yeah you just it right this felt like and was a personal attack.
Looks like the OS community acts like a cult sometimes.

Holy shit man. You posted in a thread dedicated to developing open source solution:
-picture of a PRODUCT
-information about future availability of pictured product in your STORE

and _you_ feel attacked because someone pointed that out? and call it "toxicity"? and to defend yourself the best course of action was to reply with "you do not want to hear my critique on the design choices here"?

galanopu wrote on 2022-12-09, 03:16:

That why the short message too. But I really hate to be told what to say and to self-censor. So yeah.

do you realize this is a You problem then? and not toxic people attacking you?

Some people really never quit... I wonder if attacking someone like that is also in the rules of this forum too.

rasz_pl wrote on 2022-12-09, 20:55:

galanopu wrote on 2022-12-09, 03:16:

90MHz crushes once I enable clock doubling even with 4.1V

Worse than Sphere478 one? weird

The same people are also so assuming and gaslighting that can not even get the Numbers right.

rasz_pl wrote on 2022-12-09, 20:55:

-you can totally remove that pcb notch near trimmer and everything will fit just fine, trimmer case can slightly overhang -accor […]
Show full quote

Sphere478 wrote on 2022-12-09, 02:31:

I gotta run I'm half way through redoing the trim pot and capacitor. here is my place save. I circled the work area. 🤣

-you can totally remove that pcb notch near trimmer and everything will fit just fine, trimmer case can slightly overhang
-according to https://jlcpcb.com/help/newsdetail/42-6-Layer … on-PCBs-for-$20 "$2 limited-time special offer before 5th January 2023, that the board within 5cm*5cm (5pcs with ENIG and via-in-pad) will be reduced from over $100 to a special price of $2." you can do via in pad for free on jlcpcb now, might hurt solderability of Cio Cout, but better electrically
-decoupling cap pads are truly symbolic, they dont do anything for soldering caps there 😀 they dont extend at all. G16 L17 dont have clearance and short to ground

I just have to ask - whats with your aversion to clearance? :0)

galanopu wrote on 2022-12-09, 03:16:

90MHz crushes once I enable clock doubling even with 4.1V

worse than Sphere478 one? weird

galanopu wrote on 2022-12-09, 03:16:

PS: Let me just say that I have zero need to directly advertise here, so I do not do it.

yet you did

galanopu wrote on 2022-12-09, 03:16:

In the end I do not even know if I will sell this

you knew two days ago "and then to the store"

galanopu wrote on 2022-12-09, 03:16:

And yeah you just it right this felt like and was a personal attack.
Looks like the OS community acts like a cult sometimes.

Holy shit man. You posted in a thread dedicated to developing open source solution:
-picture of a PRODUCT
-information about future availability of pictured product in your STORE

and _you_ feel attacked because someone pointed that out? and call it "toxicity"? and to defend yourself the best course of action was to reply with "you do not want to hear my critique on the design choices here"?

galanopu wrote on 2022-12-09, 03:16:

That why the short message too. But I really hate to be told what to say and to self-censor. So yeah.

do you realize this is a You problem then? and not toxic people attacking you?

With the new cap I can move the trimmer over enough to make it rectangular again.

Honestly I only modded the decoupling pads out of boredom 🤣. The round ones will work just fine haha. Silk should help with shorting issues

But I think we need to look hard into using all 21 spaces with different capacities. Because if we can avoid needing caps in the middle that would be best. These decoupling spots are basically free cap pads without any negative impact to design. Read that again slowly.

Sphere478 wrote on 2022-12-09, 18:40:

So we really only need more filtering on vcc5 if I am understanding? We don’t need anything on vcc3 in middle? This may help me make something work.

haha, don't think I said that. Probably a case of reading what you want to hear. I did make a comment, before I got started with testing that having 10 uF caps on the PGA pins only will not help Vin. A day or so ago I mentioned that both central VCC3 and VCC5, helped with the VCC3/VCC5 noise. I spared everyone a few dozen photos, but I ran the tests in series VCC3 only, VCC5 only, VCC3+VCC5, and the waveform kept clearing up every so slightly. I mentioned it in writing. I then reaffirmed this determination with my latest post.

galanopu wrote on 2022-12-09, 21:14:

Some people really never quit... I wonder if attacking someone like that is also in the rules of this forum too.

rasz_pl wrote on 2022-12-09, 20:55:

galanopu wrote on 2022-12-09, 03:16:

90MHz crushes once I enable clock doubling even with 4.1V

Worse than Sphere478 one? weird

The same people are also so assuming and gaslighting that can not even get the Numbers right.

I was able to clock double at 90 MHz and run CHKCPU, but that was it. I think this is what rasz_pl is referring to. From what you write, it sounded like once you clock doubled, it crashed.

galanopu wrote on 2022-12-08, 17:23:

Lets keep the destitution in topic and to the technical stuff. What I will do next only depends on feipoa's will.

I would like to take you up on your offer. My will is for you to please stop posting. Please honour it. I've already served my time working with arrogant, self-glorifying, and self-righteous engineers. That era in my life is over and I do not wish to relive it on this forum. I apologise in advance if this sounds terse.

Feipoa you are the creator of the thread, so you decision is accepted.
I tried so much to help her with recommendations etc.
It is funny how some follow what I recommencement and still attack after that.
The personal attacks and insults never stop, including from you.

I am sorry Sphere478 I just can not do this any more. I am un-subbing from this topic.
It was never my intent to advertise here and I realy do not need to do so.
In the end this is just a prototype, I just wanted to share early results here.
I will make my own topic once this is finished with a Video, as usual.

I play around too to find out, in art of experimentations.

Cheers,

Sphere, yes I can try a mix of 10 uF and 100 nF. I just recall that adding 21x 100 nF cause blank screen. I checked carefully for shorts and measured, but none. I will try some selection of 10 uF with 100 nF, perhaps 8x 10uF and 8x 100nF, or some combination thereof.

Sphere478 wrote on 2022-12-09, 21:25:

But I think we need to look hard into using all 21 spaces with different capacities. Because if we can avoid needing caps in the middle that would be best. These decoupling spots are basically free cap pads without any negative impact to design.

afaik there is a good reason for putting bulk caps at the source and decoupling at the chip pins

galanopu wrote on 2022-12-09, 22:47:

I tried so much to help
"you do not want to hear my critique on the design choices here"

galanopu wrote on 2022-12-09, 22:47:

It was never my intent to advertise here
"Coming soon... to the store"

Im speechless

Okay, this is what I can do in the center without sacrificing the design much. These are all 5v tied.

To add 3.3v tied will require sacrifices.

A 3.3 can probably be added under the cin by the trim pot again.

Can you make these constraints work? Or do you definitely need 3.3v caps in middle?

These are low resistance connections directly to 5v mobo pins btw. (In middle)

I don't think adding another cap under the trimmer is a good idea. When I briefly tried with two Cin1 caps piggy backed, I am pretty sure I noticed those periodic noise spikes increased slightly in amplitude. Scope image are saved on my HDD, but I didn't present them. I can only comment on the caps added to the centre region because I have tested for this and provided scope images. If you can only do 7, that's fine. We are already at the point of splitting rabbit hair.

You were able to make the overhang smaller? CONGRATS!

Cost aside, would there be any benefit to producing the PCBs in 2 mm height so that the PGA pins won't stick as far down under the PCB? And if there are signal traces stacked on one another without any grounding between, could this extra thickness help at all?

Looking at this closer, I can add more on the underside without much sacrificing also. Again, on the 5v though

Do you want more 5v caps?

Thicker pcb will mean less interference and suck pins in more.

It will also increase via resistance.

Give and take.

I think 1.6 makes sense unless 2.0 is same price and even at that, a toss up

Edit: managed to fit 10 5v 1210 caps above and below without much sacrificing of design.

Want all of em? Or take some out?