I asked a EE friend of mine about the caps, what ones he thinks to try.

Here was his response:

“ Not really, he just needs to futz with it. Add both larger and small input and output caps, use higher ESR on output, etc.”

Sounds like F around and find out. Use the scope to see.

He might had lighthearted about studying on EE.

ESR, less is best since this allows least resistance to keep noise small due to able to supply high current spikes that processor imposes on the voltage rails.
And paralleling many same capacitors goes around the impedance issue at high frequency (noise) in effect broadens the conductors larger allows for lower overall resistance will make the noisy CPU more quieter.

For now, get eight MLCC 33uF or 47uF 6.3V capacitors and install them on the center pads.

GPU and processors and any very low resistance on VCC rails for chipsets needs this.

800Hz is too low frequency to be true from a processor and can be safely ignored and also impossible because the power supply is switching type in region of around 20KHz to 50KHz switching frequency through the power supply's transformer. The true noise is white noise on the processor's power rails is over 100MHz or more at less than 1mV ideallly.

Yes, you do need to put one each 10uF MLCC capacitors right at input and output legs of the regulator IC to strong ground, always as per the datasheet to prevent oscillation.

I read datasheets and have thick book called "The Art of Electronics" by Paul Horowiitz and Winfield Hill, on doing designing electronics properly.

PS: When measuring using oscilloscope, *always* keep the ground clip original short length and always measure with ground clip attached close to the DUT's grounds under study.

Cheers,

feipoa wrote on 2022-11-27, 22:26:

ChrisK: I have definitely observed a good variance in how hot different linear regulators get under identical load conditions.

its pure ohms law, all linear regulators act as resistors radiating required voltage drop in form of heat. Difference between different models is in how precise (accuracy vs changing load, vs changing input voltage) and stable (thermal stability, load transient response, ripple rejection, internal noise) they are, how much margin (Vin-Vout delta) they require, max current rating, quiescent/standby current (how much power is wasted with zero load/how much power is wasted in off state. Important for battery powered devices). Then you get more fancy stuff like over current protection, under voltage cutout, thermal limit shutdown, inrush current limiter.

TLDR: The only factor influencing LDO efficiency is quiescent current, and this is measured in microamps nowadays 😀 Even granddaddy 7805 is barely ~5mA meaning it will waste "burn" whole 60 milliwatt.

feipoa wrote on 2022-11-27, 22:26:

I plan on trying a variety of PSU's. The 800 Hz may even be my crummy scope. 800 Hz is awfully close to the 1 KHz reference option it employs.

The unit is functional, but I'm trying to get more MHz out of it, and this is why I'm focusing on the seemingly unimportant 200 mV waveform. If someone has a cap. value to recommend for an 800 Hz ripple, it would speed things up some.

Another probe in the input will tell you if its from psu. The source of 800Hz might be LDO oscillation, might be your supply, highly unlikely its you scope unless you clipped ground lead somewhere away from measured circuit.

Sphere478 wrote on 2022-11-28, 00:23:

I asked a EE friend of mine about the caps, what ones he thinks to try. Here was his response: “ Not really, he just needs to fu […]
Show full quote

I asked a EE friend of mine about the caps, what ones he thinks to try.
Here was his response:
“ Not really, he just needs to futz with it. Add both larger and small input and output caps, use higher ESR on output, etc.”
Sounds like F around and find out. Use the scope to see.

There are different specializations. Im an EE and couldnt tell off the top of my head either, I never designed linear power supplies. I just plop in whatever datasheets tells me to.

pentiumspeed wrote on 2022-11-28, 00:53:

ESR, less is best

Nooo. This isnt a switching regulator, and this isnt a high transient load.
https://www.edn.com/pcb-and-esr-subtleties-in … nd-ldo-designs/ TLDR its all control theory, transfer function. https://www.youtube.com/@katkimshow has great video series on control Control Systems.

pentiumspeed wrote on 2022-11-28, 00:53:

GPU and processors and any very low resistance on VCC rails for chipsets needs this.

modern ones powered by buck regulators

Maybe it’s the ram refreshing

I'll reply with photos later. In short, the issue with 200 mV ripple at low freq. (700-800 MHz) does not occur with the commercial product (Improve It 5V PGA168 to 5V PGA132 adaptor + another Gainbery voltage interposer) on the same testbed. There is also a measurable high freq. ripple at the FSB freq of 25 MHz, also around 200 mV. The amplitude of the both the low and high freq. ripples appear to reduce once clock double is enabled. The reduction is from around 300 mV to 200 mV. The low and high freq. ripple witnessed from the commercial product is much smaller, like 20 mV.

It was a good idea to connect Vin to the second channel of the scope. When I look at the incoming voltage from the PSU on CH2, I do not witness measurable high or low freq. ripples. Thus I suspect I just need to play around with caps on Vout rather than Vin.

The caps I had ordered in anticipation of issues like this are:

Ceramic 1210
10 nF
47 nF
100 nF
220 nF

Tantalum 1411
150 nF
220 nF
330 nF

Tantalum 2312
0.47 uF
10 uF
22 uF
47 uF
100 uF

Since I already have one tantalum 10 uF cap on Vin and one tantalum 10 uF on Vout, I will leave those in place but add a second 10 uF tantalum on Vout in the centre region to combat the low freq ripple. If it does nothing, then add 100 uF tantalum on Vout.

For the high freq. ripple, I'll first try adding 150 nF tantalum to Vout.

These are my starting points anyway. Any further insight welcome.

I would be kind of curious to know what happens when you attach it to a laboratory power supply instead of that linear regulator

A large lifepo4 might make a good temporary stable power source also

feipoa wrote on 2022-11-28, 05:00:

Since I already have one tantalum 10 uF cap on Vin and one tantalum 10 uF on Vout, I will leave those in place but add a second 10 uF tantalum on Vout in the centre region to combat the low freq ripple. If it does nothing, then add 100 uF tantalum on Vout.

For the high freq. ripple, I'll first try adding 150 nF tantalum to Vout.

no, as close to CPU voltage input as possible

The CPU voltage inputs are all over the PCB and go to a power plane. If you look at this image, download/file.php?id=150117&mode=view , what are you proposing for capacitor value and locations?

remember this Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP! ? 😀
Maybe those spots in the middle of socket will be fine, probably would be better if they were closer to the walls. Definitely dont hang a Christmas tree of capacitors on the Out leg of regulator, thats not where they are needed the most.
You can try replacing output tantalum with ordinary cheap 10-100uF electrolytic and see if that 800Hz goes away.

I haven't seen any 386 or 486 interposer use a dedicated cap on each Vcc like that. They have all used between 1 and 4 tantalums on the VRM's Vout in a rather centralised region.

So how many mhz can you get with it as is?

Remind me how much mhz you can get with different adapters/configs?

Personally, I’d load it up with a bunch of 100uf tantrums and see what that does…

But I’m crazy.

Sphere478 wrote on 2022-11-28, 06:31:

So how many mhz can you get with it as is?

yeah, seeing as all 386/5V 486 boards power cpu straight from the at supply maybe Im just inventing issues that arent there

rasz_pl wrote on 2022-11-28, 02:19:

feipoa wrote on 2022-11-27, 22:26:

ChrisK: I have definitely observed a good variance in how hot different linear regulators get under identical load conditions.

its pure ohms law, all linear regulators act as resistors radiating required voltage drop in form of heat. Difference between different models is in how precise (accuracy vs changing load, vs changing input voltage) and stable (thermal stability, load transient response, ripple rejection, internal noise) they are, how much margin (Vin-Vout delta) they require, max current rating, quiescent/standby current (how much power is wasted with zero load/how much power is wasted in off state. Important for battery powered devices). Then you get more fancy stuff like over current protection, under voltage cutout, thermal limit shutdown, inrush current limiter.

TLDR: The only factor influencing LDO efficiency is quiescent current, and this is measured in microamps nowadays 😀 Even granddaddy 7805 is barely ~5mA meaning it will waste "burn" whole 60 milliwatt.

That's what I intended to say. Only difference in total power consumption between different LDOs can be quiescent current.
Speaking of this, the MIC29302WT can have up to 35mA @ 1.5A load which would lead to 175mW max. quiescent power @ 5Vin (10mA respectively 50mW as typical value) plus 2.1W @5Vin/3.6Vout/1.5Aout.
Maximum LDO temperature depends on package type and thermal resistance from package to board. The lower that is the more heat can flow from the LDO to the board and the lower the LDOs own temperature.

rasz_pl wrote on 2022-11-28, 02:19:

feipoa wrote on 2022-11-27, 22:26:

I plan on trying a variety of PSU's. The 800 Hz may even be my crummy scope. 800 Hz is awfully close to the 1 KHz reference option it employs.

The unit is functional, but I'm trying to get more MHz out of it, and this is why I'm focusing on the seemingly unimportant 200 mV waveform. If someone has a cap. value to recommend for an 800 Hz ripple, it would speed things up some.

Another probe in the input will tell you if its from psu. The source of 800Hz might be LDO oscillation, might be your supply, highly unlikely its you scope unless you clipped ground lead somewhere away from measured circuit.

Also measure without the CPU/interposer to see if it's still there. Not sure if the CPU could generate such a low frequency. I'd also tip on some LDO oscillation or the like. But generally most modern LDOs aren't very prone to oscillation.
I also doubt that there's any reasonable cap to equal an 800 Hz disturbance in terms of size. 800 Hz is almost DC.

The datasheet for the MIC29302WT says it is stable with a 10uF Cout at full load and that Tantalums are fine. It says further "In fact, extremely
low ESR capacitors may contribute to instability.". So personally I'd stick with the 10uF Tantalum or maybe parallel two of them to see if the halved ESR has any effect.

It also mentions a minimum load current for stability which is given as 7 mA for this MIC29302 (sec. 4.3). Not sure how much the SXL2 consumes but maybe some dummy load behind the LDO is worth a test.

pentiumspeed wrote on 2022-11-28, 00:53:

ESR, less is best

Generally this is preferred to reduce output ripple as much as possible, but there are switching regulator topologies which rely on the output capacitor's ESR for stability (so called "current-mode" regulators).
They need this little signal as input for their error amplifiers in the feedback path. If that's too small they can't regulate the output and may get unstable.

Edit: typo

feipoa wrote on 2022-11-28, 05:00:

I'll reply with photos later. In short, the issue with 200 mV ripple at low freq. (700-800 MHz) does not occur with the commerci […]
Show full quote

I'll reply with photos later. In short, the issue with 200 mV ripple at low freq. (700-800 MHz) does not occur with the commercial product (Improve It 5V PGA168 to 5V PGA132 adaptor + another Gainbery voltage interposer) on the same testbed. There is also a measurable high freq. ripple at the FSB freq of 25 MHz, also around 200 mV. The amplitude of the both the low and high freq. ripples appear to reduce once clock double is enabled. The reduction is from around 300 mV to 200 mV. The low and high freq. ripple witnessed from the commercial product is much smaller, like 20 mV.

It was a good idea to connect Vin to the second channel of the scope. When I look at the incoming voltage from the PSU on CH2, I do not witness measurable high or low freq. ripples. Thus I suspect I just need to play around with caps on Vout rather than Vin.

The caps I had ordered in anticipation of issues like this are:

Ceramic 1210
10 nF
47 nF
100 nF
220 nF

Tantalum 1411
150 nF
220 nF
330 nF

Tantalum 2312
0.47 uF
10 uF
22 uF
47 uF
100 uF

Since I already have one tantalum 10 uF cap on Vin and one tantalum 10 uF on Vout, I will leave those in place but add a second 10 uF tantalum on Vout in the centre region to combat the low freq ripple. If it does nothing, then add 100 uF tantalum on Vout.

For the high freq. ripple, I'll first try adding 150 nF tantalum to Vout.

These are my starting points anyway. Any further insight welcome.

This sounds like a good plan of action

Linear regulators do need ceramic MLCC or tantalum capacitors right at the in and output pins to prevent oscillation, not electrolytic. This says so in the in any linear regulator datasheets. And in another datasheet for processor, these do state must have low ESR type, either tantalum or ceramic MLCC distributed capacitors in parallel too.

You have included these on your project, but 8 capacitors not installed for the CPU, please finish this.
You may need to include 2.2uF MLCC capacitors in parallel with each tantalum for the regulator, also another MLCC capacitor on the voltage adjust pin.

Cheers,

feipoa wrote on 2022-11-14, 10:11:

elbbar, I'm not sure what you are showing a photo of...

On the subject of capacitors and the design of a CPU board, what differences can you spot between the two designs pictured? Which design has stability problems?

Did you watch the videos I posted URLs for? Can you be more specific about which part you don't understand?

feipoa wrote on 2022-11-28, 06:20:

I haven't seen any 386 or 486 interposer use a dedicated cap on each Vcc like that.

Do you know what capacitors look like? There are also differences in what can be made by hand and what can be made by automated processes now (and in 1995). There are also differences in something designed/manufactured to a price point and something for performance/hobbyists. For example, "better" could mean cheaper/easier to make, or it could mean the ability to clock higher.

If you want to talk specifically about 486 chips, here is a picture of a 486-type adapter board. The surface mount CPU chip has capacitors lined up as close as possible/practical to many of the pins (directly beneath). I don't know the pinout for these chips but I have a pretty good guess as to which pins those capacitors are connected to.

Now if you compare a 486 to some of the newer CPUs, what can you observe about the type, quantity, and locations of capacitors? Is this making sense?

You posted a picture of your hand made socket adapter that has problems with higher frequency operation. Given the time you must have put into it, adding a few (or even a single) bypass capacitor would be trivial. This experiment would be completely reversible. Even accurately using a scope to measure signals like you've described is more work than simply adding a couple capacitors to see what happens.

I was contemplating earlier, should We be considering termination resistors?

It would be pretty tough to add them to the interposer, but perhaps a second interposer. The rotator board?

Btw, feopia, I am back. And did a little tinkering with the rotator.

It also will be a slow process. I’ll play with it in spare time

This upgrade module circuit board is sub optimal due to too much concessions due to circuit board when designed especially linear regulator and not enough capacitors closer to the VCC and VDD pins. If the guy knew the requirements, would keep this in mind when doing rough layout of all the components then route the tracks and vias. Back then, the software to do so is in infancy. Now we have much easier to do but still you have to research the datasheets and knowledge learned from good books on electronics engineering.

Good example same with sound blaster board design was terrible let alone few chips they designed was terrible too as like was trying to be cheap by using engineering people as little as possible.
Same with current issues with chinese made mini modules for projects.

For this reason, I wanted to someday redo the video card correctly for example and ISA multi-i/o cards for IDE and floppy and other ICs on 4 or 6 layer PCB instead of two layers.

Cheers,

@elbbar

I don't know if it is a language barrier at play or just the wording but the tone in your last message feels a bit off compared to the friendly thread.
Figured i should "see something say something".
Anyhow, back on topic.

pshipkov wrote on 2022-11-29, 01:46:

@elbbar […]
Show full quote

@elbbar

I don't know if it is a language barrier at play or just the wording but the tone in your last message feels a bit off compared to the friendly thread.
Figured i should "see something say something".
Anyhow, back on topic.

I think 'combative' is a good description. I was that type of asshole in my twenties, mm those were the days 😀