VOGONS


Gigabyte GA-586HX (Rev. 1.53)

Topic actions

Reply 40 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

I'm having a stab at a carrier PCB with rotary switch and resistors, which will hopefully fit in the TO220 holes. Like that will need a slight motherboard mod to change the middle adjust pin to a ground pin. But that'll look neater.

Reply 41 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

The stand-up type would be smaller and more universal! Many mainbords with small vertical heatsinks provide less space and wouldn´t allow horizontal fixing - let alone on a carrier board with switch, resistors and (if necessary) filter caps.

Modding the middle pin on the Gigabyte mainbords could be tricky. It´s also a massive via connecting the large area on the upside under the heatsink that extends unter the CPU socket (VRM-out) to the even larger area on the flipside that also connects to some Vcore CPU-pins.

On the GA-586DX a standing PCB is not an option. It would prevent the RAM sticks from being inserted / removed. That´s why I never replaced the original heatsinks with bigger ones.

Last edited by majestyk on 2022-01-21, 21:58. Edited 2 times in total.

Reply 42 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Oops, I was being dumb, it's pin 1 that's the adjust pin. 1-Adj, 2(+tab)-Out, 3-Vin. The adjust isn't needed (that handled on the regulator) but a reference ground is needed. Does that sound more sensible? Shouldn't be much current on the ground pin.

Reply 44 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

So I've done a first pass. Pin header at the bottom should line up with TO220 style 3 pin adjustable regulators. On the back is a rotary switch, I used the datasheet for this: https://uk.farnell.com/multicomp/mcrm3haf-16r … -pos/dp/2083001 but plenty of other rotary switches should fit. 0805 SM resistors (I think, just pulled out a library part that unhelpfully just says 'SM', but the dimensions look right). I normally use 0402, but they're a bit fiddly. Pin header can be a right-angled version so that it stands upright. Copper should probably be 2oz. Dimensions ended up as around 40mm x 16mm. The 12Amp module would be smaller, but doesn't seem to be available right now.

I've missed off a pin 1 marker for the switch, some of the thermal relief around the solder pads is a bit ropey, and I should probably leave a bigger gap for the copper pour. Not entirely happy with the Trim routeing, but it seemed to be a balance how nice that was, and how wide a copper pour I could get for Vout, and I've assumed it's more important to have lots of copper.

I've also got a nagging feeling I've got the pin header the wrong way around. So it could do with some silk screen labeling to say which pin is which.

Attachments

Reply 45 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

Looking sharp!

Some mainboards have linear regulators with transistors like "D45H2A" (PNP) and a "B-C-E" pinout (seen from the front) where the converter module can also be used.

In most cases the voltage regulator is placed next to the CPU-cooler so it might be difficult to turn the rotary switch without removing the heatsink. When replacing the CPU this is no hassle, but when experimenting with different voltages or overclocking this could be tedious.

I wonder if we could have 4 contacts at the bottom of the PCB like "Gnd - Out - In - Gnd" so the regulater could be rotated 180° and the 3 pin header soldered in the second position in such cases.

Reply 46 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

It would increase the volume a bit, but I could see about fitting a right-angle rotary switch so it could be accessed from the top? Might be able to sort out the input pins as you've suggested, but it'll mean a jumper wire because otherwise the ground trace will cut across the Vin/Vout copper areas or it'll have to go right around the edge of the board which feels like it's getting a bit long.

And I've definitely got the pin header the wrong way around at the moment. The left pin in the model (when looking at the side with the actual regulator) is Vin. But that's the way round it pretty much has to go because of where the inputs on the regulator are. So really, the side with the regulator is the 'back'. I'll add some labeling on the silkscreen.

Reply 47 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

Maybe something like this (and centered):

CoolReg_gnd.jpg
Filename
CoolReg_gnd.jpg
File size
11.54 KiB
Views
819 views
File license
CC-BY-4.0

The longer ground connection could decrease the efficiancy of the ceramic filter caps at the input and output of the Murata module. Same goes for any necessary wire jumpers on the mainboard to get the ground connection to the ground pin. This connection is about 2cm long on my GA-586DX at the moment. I think I will do some tests, if shorter / longer connections reduce / increase ripple.

The option to mount the regulator in two positions would also be useful in situations when there´s very little space on the mainboard at one side of the regulator.

Reply 48 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Boo, can't get surface mount right-angle mount rotary switches. Probably a good thing, I can imagine pads being torn off. So right-angle means getting taller and making enough space for through hole. It's currently about 40mm tall. Would it be better if it were ~50mm tall, but all the components on the same side, so it could lay flat?

On the ground pin, what you can't see is the couple of vias I've got stitching the top and bottom copper in that corner to try and increase the copper volume between that pin and Vin on the module. At the moment it should cope with about 15A before the trace starts getting a bit toasty. Which is fine even if it's set for 5V in/3.3V out. If I back that off and aim for 10A (could then use the same pin header layout if the lower current modules ever become available), and rejig the Vout copper pour area a bit, I can probably get the Vin to be single sided, and still cope with having a pin hole punched through it for a ground on that side. I'll try and make the trace a bit thicker. The copper pour on the Vout on that side is probably overkill.

Reply 49 of 71, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
snufkin wrote on 2022-01-22, 20:49:

Boo, can't get surface mount right-angle mount rotary switches. Probably a good thing, I can imagine pads being torn off. So right-angle means getting taller and making enough space for through hole. It's currently about 40mm tall. Would it be better if it were ~50mm tall, but all the components on the same side, so it could lay flat?

On the ground pin, what you can't see is the couple of vias I've got stitching the top and bottom copper in that corner to try and increase the copper volume between that pin and Vin on the module. At the moment it should cope with about 15A before the trace starts getting a bit toasty. Which is fine even if it's set for 5V in/3.3V out. If I back that off and aim for 10A (could then use the same pin header layout if the lower current modules ever become available), and rejig the Vout copper pour area a bit, I can probably get the Vin to be single sided, and still cope with having a pin hole punched through it for a ground on that side. I'll try and make the trace a bit thicker. The copper pour on the Vout on that side is probably overkill.

What about micro jumpers and a small smd resistor array that you add up to get to different voltages

A rotary does sound better though.

A small voltage display would be cool. i think you can get them quite small. But prob still too big. Prob too much space even if you just used smd leds to make your own display. 🙁

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 50 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie
Sphere478 wrote on 2022-01-22, 22:05:

What about micro jumpers and a small smd resistor array that you add up to get to different voltages

I was mostly having a stab at this to get away from jumpers. By the time you have a 4x2 jumper block, it's taken up most of the space of a rotary switch, and the rotary is way more convenient. It ends up being the same resistor selection.

A small voltage display would be cool. i think you can get them quite small. But prob still too big. Prob too much space even if you just used smd leds to make your own display. 🙁

Trying to get to not much bigger than TO220. Not really succeeding, but still want to keep size down if possible. Volt meter with a front panel display would be easier, there's usually plenty of space there.

I've got some copper pour areas that I'm a bit happier with now. But realised I made the layout for the resistors more complicated than it needed to be because I connected the common of the switch to Ground. If I connect that to the Trim then the resistors go from the Switch to Ground rather than Switch to Trim. There's now plenty of ground trace around, so they can just connect to that, rather than having to route a chunky Trim trace around the place. So going to see if that makes it any nicer.

Reply 51 of 71, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
snufkin wrote on 2022-01-22, 22:52:
I was mostly having a stab at this to get away from jumpers. By the time you have a 4x2 jumper block, it's taken up most of the […]
Show full quote
Sphere478 wrote on 2022-01-22, 22:05:

What about micro jumpers and a small smd resistor array that you add up to get to different voltages

I was mostly having a stab at this to get away from jumpers. By the time you have a 4x2 jumper block, it's taken up most of the space of a rotary switch, and the rotary is way more convenient. It ends up being the same resistor selection.

A small voltage display would be cool. i think you can get them quite small. But prob still too big. Prob too much space even if you just used smd leds to make your own display. 🙁

Trying to get to not much bigger than TO220. Not really succeeding, but still want to keep size down if possible. Volt meter with a front panel display would be easier, there's usually plenty of space there.

I've got some copper pour areas that I'm a bit happier with now. But realised I made the layout for the resistors more complicated than it needed to be because I connected the common of the switch to Ground. If I connect that to the Trim then the resistors go from the Switch to Ground rather than Switch to Trim. There's now plenty of ground trace around, so they can just connect to that, rather than having to route a chunky Trim trace around the place. So going to see if that makes it any nicer.

What if you pc104 it and start stacking small pcbs into a 3d shape

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 52 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie
Sphere478 wrote on 2022-01-22, 23:10:

What if you pc104 it and start stacking small pcbs into a 3d shape

Did briefly think about trying to put a small switch+resistor PCB in the space above the FETs, next to the inductor on the Murata module. So the carrier board would just handle getting current to and from the module and give a easy way to connect on to a motherboard. Could solder some rigid wire to the points on the module, (including the trim, ground and one other), which might make a tripod strong enough to take the stress of rotating a switch.

But decided not to. Partly because that'd make putting it together more complicated. Partly because that'd put the switch at the bottom, so difficult to get to if this is standing upright. Mostly because that sounded more fiddly to design.

Version 2. Now with silkscreen labeling, better copper pour areas, ground on bother sides of the header, no thermal relief on the header pins, and more sensible connections on the switch+resistors. Decided that the header didn't need thermal relief since you can easily see if you have a good joint, which is a bit trickier with the pads on the module, so they do still have spokes. I think (if I've used the design tool properly) that it should be able to handle 20A output/12A input, with a 30C temperature rise over ambient on the PCB traces, if made with 2oz copper. That goes up to 90C rise with the usual 1oz copper. Which sounds less good. Probably best to keep to about 12Amps with 1oz copper.

Switch looks big because I've allowed for one with a knob so it could be changed by hand. Looks like about 4mm can be saved if a flush switch is used instead.

I still have no pin 1 mark for the switch, because I once again forgot about that.

CoolReg2.jpg
Filename
CoolReg2.jpg
File size
30.72 KiB
Views
774 views
File license
CC-BY-4.0

[edit: now with added pin 1 marker. Honest. It's in the top right, hidden by the switch. Plus there's some more silkscreen stuff]

CoolReg2_NowWithPin1Dot.jpg
Filename
CoolReg2_NowWithPin1Dot.jpg
File size
28.02 KiB
Views
766 views
File license
CC-BY-4.0

Reply 53 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

I think the most power-hungry CPU would be a K6-III 450, this would mean a maximum current of 12.3A.

It´s a good thing the Murata modules drop to their minimum output voltage of 0.69V when the Trim-input is open. Otherwise loose contacts of the switch (age, dirt) could cause a K6-2+ to roast itself to death due to 3.3V core voltage without advance warning.

Reply 54 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Ok, so 1oz copper is probably ok unless the output is shorted. At which point the Murata module shuts down anyway. I'd still be tempted to go with 2oz copper depending on the cost difference.

Yeah, I wondered about the switch reliability. But I've seen cases of jumpers getting corroded and making intermittent contact, so I figured as long as no one spins the switch round hundreds of times then the reliability should be ok. Only way to do better is make it a fixed value output. As it happens, because R1 is always in circuit then the lowest the output can drop is 1.3V

Something I don't know, and haven't found in the datasheet, is the current on the Trim. I'm assuming it's going to be single digit mA given that the lowest resistance (all resistors on) is around 2.5k. Minimum current I've seen on any of the rotary switches has been around 25mA when under load, so they should all be ok. Probably best not to switch it whilst it's on though, I don't know if the switch has any guarantees about break-before-make, so the output might spike to 3.55V.

So I think this is pretty much done. The only pinout arrangement we don't have is GND/adjust in the middle, but at that point it just mean using flexible wires to connect it up. Do you think it's worth having a go at drawing out one that's about 50mm, rather than the current 40mm, but has all the components on one side so it could lie flat?

Here's the current gerbers:

Attachments

Reply 55 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

Great, this VRM will be usable in nearly every mainboard with existing TO220 regulator (integrated or Transistor) plus it can operate in both horizontal and vertical position. Even horizontal with the Murata module downside will not cause any heat issues.

To round it up we wouldd need an additional model for the 30-pin PI VRM connector. This connector is present on many S5/S7 boards or it can be added on as standard 2.54mm 2x15 pin connector.
Original VRMs (those were made for 2.8 / 2.9V only) are very hard to find today, they have the same overheating issue as linear onboard regulators and they are unable to provide the low core-voltages for K6-III+, 2+ and many regular K6-2 models.

PI_VRM_lin1.JPG
Filename
PI_VRM_lin1.JPG
File size
138.54 KiB
Views
742 views
File license
Fair use/fair dealing exception

A PCB half as high would host a Murata module, resistors, Switch and even input- and output-filter caps.

Reply 56 of 71, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
majestyk wrote on 2022-01-23, 20:33:
Great, this VRM will be usable in nearly every mainboard with existing TO220 regulator (integrated or Transistor) plus it can op […]
Show full quote

Great, this VRM will be usable in nearly every mainboard with existing TO220 regulator (integrated or Transistor) plus it can operate in both horizontal and vertical position. Even horizontal with the Murata module downside will not cause any heat issues.

To round it up we wouldd need an additional model for the 30-pin PI VRM connector. This connector is present on many S5/S7 boards or it can be added on as standard 2.54mm 2x15 pin connector.
Original VRMs (those were made for 2.8 / 2.9V only) are very hard to find today, they have the same overheating issue as linear onboard regulators and they are unable to provide the low core-voltages for K6-III+, 2+ and many regular K6-2 models.

PI_VRM_lin1.JPG

A PCB half as high would host a Murata module, resistors, Switch and even input- and output-filter caps.

There was a cool vid where a guy made a vrm for those lemme see if I can find it

https://youtu.be/J0NLGfocviU

Watch this, see if it gives ya any ideas 😀

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 57 of 71, by majestyk

User metadata
Rank Oldbie
Rank
Oldbie

Best idea there is to put the PCB between the pin-rows of a standard connector instead of searching for an angled connector.
Sadly this VRM won´t go below 2.2V. But they seem to be out of stock anyway.

He also uses the flamethrower a lot 😀

Reply 58 of 71, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
majestyk wrote on 2022-01-23, 20:53:

Best idea there is to put the PCB between the pin-rows of a standard connector instead of searching for an angled connector.
Sadly this VRM won´t go below 2.2V. But they seem to be out of stock anyway.

He also uses the flamethrower a lot 😀

Any good design should go down to 1.2 or 1.4v imho, because some people will play with mobile chips.

Yeah, pcb between the pins is a good way to do it.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 59 of 71, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Something like this: https://www.datasheetarchive.com/pdf/download … e=M&term=RCB002 ?

[edit: WARNING, that datasheet has an error in the pinouts, getting side A and B swapped round, see the next page in this thread]

That pinout seems to fit with the part in the video where he added the jumpers to the header, which would have connected V3 to both Vi/o and Vcore.

So it looks like it takes 5V (pins 14&15 AB), and puts out Vcore (pins 7,8,9,10 AB, except 9A which is GND. Because why not). Should be able to connect the Power Good directly from the Murata module (open drain, so it can pull it low and turn off the machine if it has a problem). Looks like it also need to loop the V3 pins to Vi/o. It's got GND for reference, so we don't need to worry about that this time.

Does that sound about right?

I'm thinking put the Murata module long-ways down, it's about the same length as a 15x2 pin header. Then make the PCB slightly taller and this time use a through hole right-angle rotary, rather than a SMD on the back, since it will definitely be mounted vertically.

There are 4 pins left that I don't know what they do or are: UPVRM#, SENSE, DISABLE, and V12. Any ideas?

It feels like Disable should be able to connect to the on/off input (low to turn on, high to turn off?), and the Sense would be a low current feedback of the actual Vcore. Which would be a nice thing to do to make sure the CPU gets the correct voltage, except that the Sense input for the Murata module isn't documented. I found a thread on some random vaping site (do people use 100W vaping kit?) where someone had asked Murata about it and it does sound like it just needs connecting sense+ to the load and sense- to ground near the load. If that's right, then it's probably nearly as good to connect sense- to any of the GND inputs on the header.

Ah, or perhaps one the pins labelled GND in that datasheet is actually a sense GND? Any chance someone can check if any of the GND pins actually has a trace going to it, and isn't just connected to the motherboard ground plane? If it's any of them, I'd guess 9A, 13A or 13B.

Last edited by snufkin on 2022-01-26, 22:58. Edited 1 time in total.