homemade VRM working!

[DanU]

I finally finished the home-made Voltage Regulator Module I was working on this weekend. The design should be good for 15 amps at between 1.6 and 2.3V. I even put DIP switches to select the voltage! The limiting factor is probably the relatively small heatsink on the power transistor. Since it's a linear design, it's gonna have to dissipate about 20 watts. I hope it holds up.

The guinea pig is a 4 year old Amptron PM8800. It's a 430VX board that only supports 3.5V and 3.3V CPUs. But it does have a mysterious VRM header that accepts a VRM which will allow split voltage CPUs to be run. The only problem is that these VRMs are hard to find and they're somewhat expensive... so I built my own!

I removed the jumpers that were in the VRM header, plugged in the VRM and set the voltage to 2.2V. Then I plugged in a K6-2 450 I had leftover from my K6-2+ upgrade. I set FSB to 75MHz and the multiplier to 6x(2x). There was some hesitation as I was about to push the power button. Mostly I was afraid the thing would start smoking. But I pushed it anyway. Surprise! It booted up and started loading windows. It's also stable! No crashes!

I think I'm going to have to put a larger heatsink on the power transistor. It gets really, REALLY hot. Yes, I do have the case fan blowing directly at it, but it's still almost boiling hot. Hopefully it can take it. The 2N3055 is specced at something like 150C, hot enough to cook dinner on.

[Eli]

Hey that's great, let us know how it holds up. I also have a VX board here that needs a VRM to run an MMX chip. Care to share your schematics? http://www.sysopt.com/forum/wink.gif

[Szech]

Hey! That's really cool! http://www.sysopt.com/forum/biggrin.gif

Could you be so kind as to post the schematic?

[Graham]

Great transistor the 2N3055, 150W and 10 a penny, well not quite.
MOS trannies dissipate less heat and can have fantastic specs.
Please post a diagram.

G

PS, I guess the transformer is pretty large, what did you use?

[This message has been edited by Graham (edited 04-30-2001).]

[DanU]

Here's the scematic:
<IMG SRC="http://members.nbci.com/solder/gfx/vrm_schematic.png" border=0>

A few notes about the circuit:
I tried to use commonly available parts. You should be able to get them from radio shack. The exception is the 30-pin header for the VRM. I picked it up at Jameco in Belmont, CA. I think it costs about 80 cents. But they do have a minimum order of about $10 for shipped orders. One alternative is to take an IDE or floppy connector and cut it down to size.

If you need the pinouts for the VRM header, you should be able to find it on Intel's web site. If not, let me know and I'll mail the PDF to you.

The 1 ohm sense resistor should be rated for at least 1 watt. You can make a 6 2/3 ohm resistor from a 10 and a 20 in parallel. In fact, you can make the entire voltage adjust circuit using only 20 ohm, 1/4 watt resistors in either parallel or series.

Since you'll be drawing up to 13A with the K6-2 550, don't rely on the four +5V pins of the VRM header alone to supply the current. I was reading 4.8V (!!!) on the VRM's +5V rail, so I connected a spare drive cable's +5V and ground wires to the VRM to supply the extra current. The six Vcore pins on the VRM header should be sufficient for the outgoing current.

The capacitors are a good idea, but you might be able to get away without them. I tried this circuit first without the caps, but the computer was a little unstable, so I threw them in as an afterthought. It works great now. 10uF is probably sufficient for the output, but 100uF doesn't hurt.

This circuit is limited to about 2.3V due to the ~2V dropout of the LM317 and ~.7 drop from the 1 ohm sense resistor (5.0V - 2.0 - 0.7 = 2.3V). This is fine for the K6-2/3/+ series of CPUs. To get more voltage, like 2.8 for the PMMX, you will need to replace the LM317 with a low-dropout regulator capable of supplying about 1A. Then you will be able to hit the maximum 3.1V allowed by the voltage adjustment switches. The disadvantage is that LDO regulators are harder to find that the ubiquitous LM317.

The other alternative to get more voltage is to use +12V to feed the LM317 and TIP30A, and use +5V to feed the 2N3055. The disadvantage to this is that the LM317 and TIP30A will get really hot and will require big heatsinks. Let me know if you want to see this circuit anyway.

Any decent power PNP capable of 1A can replace the TIP30A. I think Radio Shack has both the TIP30A and the MJE2955.

Any decent power NPN capable of 15A can replace the 2N3055. Radio Shack almost always has 2N3055's in stock. This is the only part that needs a heatsink because it gets really, really hot http://www.sysopt.com/forum/smile.gif.

[DanU]

Regarding the current demands of this VRM...
Yes, it will draw a lot of current from the +5V supply, up to 13A with a K6-2 550. This may cause havoc with old, cheap, or just plain weak power supplies. Solution: get a better power supply if you don't have one already! Believe it or not, I picked up a Turbo-Cool 300 for $5 at WeirdStuff in Sunnyvale. Sure, it's used, but PCPower and Cooling units are bullet-proof and are commonly used in mission critical servers.

Another alternative is to put in another power supply just to feed the regulator. Just don't directly connect one PS's +5V line to the other's or you might see some interesting electrical interactions. bzzzzzt!

This circuit was actually based on the one I saw on Kalle's page. The main difference is that I used a PNP and NPN in a compound darlington configuration instead of a single PNP. This eliminates the affect of a large base current swamping out the sense resistor. Plus I'm a lazy and a big fat 2N3055 is cheaper and easier to find than a big fat PNP.

And of course I have to throw in your standard disclamier. Use this information at your own risk! If something goes terribly wrong, it's your own darn fault! Take an electronics class and do your own circuit analysis to verify that this isn't really a motherboard destroying device http://www.sysopt.com/forum/smile.gif.

Comments, suggestions, criticisms, etc regarding the circuit are welcome.

[This message has been edited by DanU (edited 04-30-2001).]

[Jedi]

Many thanks to DanU for posting this, i am in the process of getting the components together and constructing the beastie to get a PC CHIPS M507 board (AKA Amptron PM-7600, AKA fugutech M507) to accept an AMD k6-2 500 (although i'll have to under clock it - thats the next tweak on the board!)

I'll post the results here when i get it running


The force may be with you, young Skywalker, but you are not a Jedi yet! www.teamjedi.freeserve.co.uk (http://www.teamjedi.freeserve.co.uk)

[tonym]

DanU,

Good job with the regulator!!

Ahhhh, the 3055 brings back many memories. It is a war-horse/work-horse transistor that was designed by TI sometime back in the late 60's! It is extremely rugged, almost impossible to destroy...a wide SOA (safe operating area)!

Instead of a linear regulator for the VRM, why don't you try a switching regulator, like a synchronous "buck" converter? They converters can be up to 92% efficient, so for a 1.6V@15A core voltage (22.5W) with a 5V input, you will dissipate approximately (22.5/0.92)-22.5=~2W. This is far easier to dissipate than the ~20W that the linear regulator dissipates.

Check out Analog Devices part ADP 3159: http://www.analog.com/pdf/ADP3159_0.pdf , note that Figure 3 does the VID "switch thing" that you do with the linear,
Linear Technology part LTC1753: http://www.linear.com/pdf/1753f.pdf , or
Semtech part SC1152: http://www.semtech.com/pdf/ic/sc1152.pdf

There's a little bit of "voodoo" in the circuit layout to minimize ground loops and parasitic circuit traces, but you'll learn by experimentation if you wish. However, these parst are as "canned" as you can get for switching voltage regulators!

Good luck...


Tony

[DanU]

Tony,
Thanks for the links.

I was thinking of building one around the National Semiconductor LM78S40, simply because it's the only controller that Jameco sells and I have no idea where to get any other switching reg controller.

For now the linear is working great. I had to install a front case fan to keep the temp at a decent level. CPUIdle also helps a lot. Without CPUIdle, the temp of the regulator shoots right up.

BTW, here is the datasheet for the LM78S40 for those interested: http://www.national.com/pf/LM/LM78S40.html

[Jedi]

I now have a complete set of parts.....

oh well, try or die http://www.sysopt.com/forum/smile.gif let you know how well it works tomorrow, gonna heatsink everyting ... just in case, with a BIG fan.

Anyone know how well an AMD K6-2 500 underclocks to 200 (66*3)?

Still, it's better than this **** Intel P133

Fingers.... scrub that... everything crossed

[Superdude688]

Hi, I know this is an old thread and I hate to dig it up, but for some reason it won't let me send PM's and I am very much interested in this thread.

I would like to build my own VRM as well, for my old IBM computer and drop a spare k6-2 in it.
The problem is your link to the schematic no longer works, which is understandable since it is over 6 years old. I was wondering if you still had the schematic and if you could send it to me or re-link it here?

I would just start tinkering and figure it out myself, but I figured I would give registering here and asking you a shot first.

Thanks,

- Dave