Hey everyone.

So I'm absolutely stuck. I've found what I need, but it's $450 and about $430 over my budget.

What I need: A regulator to drop either 5v or 12v down to 2.9v, still allowing at least 120w and 40a.

What it's powering: An 8 GPU video card with 2 ground, a 3.3v and 2.9v terminals. (It has a power supply board with a plug in and is bolted to the card [bolts transfer the power] but I can easily connect the wires to the card's terminals). The GPUs take the 2.9v, but it's confusing. The board says 2.7v on the terminal, the power board says 2.9v and the real PSU says 2.9v. So apparently... it's 2.9v >_>. Here's the best info on it I can provide:
http://www.thedodgegarage.com/3dfx/q3d_aalchemy.htm

My plan: I figured I'd get 2 of these fairly cheap computer PSUs and run them in parallel. If I use the 12v->2.9v method, then I have 2 rails to deal with per PSU. However, the '2 rails' are most likely (due to the cheapness of the PSU) just a split of one main rail and either current regulated or not. And I'd need to know which all 12v lines I solder together, as well as the grounds, would I use the -12v line? If I do the 5v->2.9v method, then there is no dual rails, but the max current put out is about 16a less. Same Q on grounds, -5v this time?

Skills: Cars and computer hardware. I know very very little about the stuff I'm dealing with, which I feel falls into Electrical Engineer and Computer Science. The 2 friend's I've consulted really weren't much help lol One is a Comp Sci grad, the other has worked in a PCB facility and knows a bit about this stuff, but again neither has sadly helped much. I know quite a bit about computers, how they work, building systems and overclocking. I'm Mr Fix-It to many people when a computer problem crops up, but this stuff stumps me.

I've looked into resistors, capacitors, diodes, bench top power supplies (expensive), already made regulators (pricey or not putting out the needed voltage) and pots. The other chap who's in my same predicament has been using a pot and gets the card to boot, output video and be read by drivers, but when he goes to test 3D the card locks up. My Comp Sci friend explained that when a pot has current draw the voltage drops and that's probably the problem.

I read a bunch of the pages in "Volume III - Semiconductors » DIODES AND RECTIFIERS »" and the Diode's part helped me understand thing quite a lot. That part where my friend's hadn't helped me much comes into play there. The one mentioned that the voltage that the Zeners are showing for specs equate to the drop they provide, where as I've now learned it's what they actually output. That was what I assumed when I first look at the specs :| So instead of needing a 9.1v diode to drop the 12v down to 2.9v, I need just a 2.9v diode. Problem there is they are all rated at 20-140mA, and assuming I understand things correctly, if I ran them in parallel I'd need between 450-1800. Again assuming I understand things correctly, with their power dissipation (what is lost during conversion?) I'd only have about 80w left over to power the card, assuming I use the 500mW.

And if you're interested, here's what I've found that doesn't meet my specs, and is far too expensive.

These ones are damn close to what I need, and are only $8. I suppose I could boost the voltage from 2.5, if that's possible. (Isolated DC/DC Converters)
http://www.cherokeepwr.com/product/dcdc.aspx


Then there's this, which is exactly, 100%, what I need. Could even do with 2 less chips and have a total current rating of 80A instead of the 4 chip 160A. At first I WAS looking at the chip and when I searched for it to buy, was $18 and got happy thinking it was the whole card. Then I realized my error and was sad again
http://www.irf.com/technical-info/refdesigns/irdcip2003a-c.pdf

So I come, begging, pleading, for a solution!!

Thank you,
-Clint

 

I can't help you directly, but this is a job for a switcher, if your application can take the noise generated. The thing is, you will need a lot of series pass transistors (to distribute the load). Personally I think it is very doable, but everything conductive will be very very thick at 80A.

If you find a switching power supply with reduced specs, think in terms of multiplying the output drivers by how many times you'll need to get the new current.

What was the matter with the second link, couldn't find one to buy?

 

Second gizmo costs $462 US from Newark. Looks like a piece of lab equipment for developers to use.

 

I can't help you directly, but this is a job for a switcher, if your application can take the noise generated. The thing is, you will need a lot of series pass transistors (to distribute the load). Personally I think it is very doable, but everything conductive will be very very thick at 80A.

If you find a switching power supply with reduced specs, think in terms of multiplying the output drivers by how many times you'll need to get the new current.

What was the matter with the second link, couldn't find one to buy?

Far far far too much for me. I'm only looking to spend $80 (at the most)


Would a mosfet regulator work?

 

Opps, missed the price. Nice gadget though. Like I said earlier, it looks doable.

 

If you want to try the modification way, you should get two ATX power supplies.
One that could handle the 5, 12 and 3.3V currents, and then one bigger which says that it is capable of 3.3V with at least the 40A.
Then you can try to change the reference voltage on the second supply and get it to work at 2.9V instead of 3.3V, leaving the rest unconnected.

Then just connect the grounds together and try.

The biggest problem is to test the PS under load, because you probably don´t have an 0.1ohm 120W resistor, and I wouldn´t recommend trying it on the card right away.

If you want I can dig up a schematic of some ATX supply, they mostly use the same topology.

 

If you want to try the modification way, you should get two ATX power supplies.
One that could handle the 5, 12 and 3.3V currents, and then one bigger which says that it is capable of 3.3V with at least the 40A.
Then you can try to change the reference voltage on the second supply and get it to work at 2.9V instead of 3.3V, leaving the rest unconnected.

Then just connect the grounds together and try.

The biggest problem is to test the PS under load, because you probably don´t have an 0.1ohm 120W resistor, and I wouldn't recommend trying it on the card right away.

If you want I can dig up a schematic of some ATX supply, they mostly use the same topology.

Thanks for the info.

To clarify to everyone, I'm only needing around 40A total. The PSU can supply 2 cards and has a 75A rail, which I think is probably a bit over for redundancy (If you will). In reality I probably only need 30A.

All the cheap PSUs I"ve found, and even most of the big names, don't have that high of a 3.3v for amperage. The 3.3v is an unknown factor in this, but since it's only supplying the rest of the card, I'm thinking whatever even just 1 PSU is will satisfy those needs, and 2 will be far more than enough (especially looking at the real PSU).

This is the PSU I plan to use (x2):
http://www.newegg.com/Product/Product.aspx?Item=N82E16817339019

When I get home I can try and get a schematic on it.

As for the resistors, I don't have one, correct, but I have found one (pricey heh)
http://www.newark.com/jsp/search/br...&Ntt=categorynumber88077&Ntx=&_requestid=4703

Now, bear in mind, I'm a total newb when it comes to the workings of PSU and all the components. So for me to fiddle with any wirings and making it put out a different voltage is something I'd need a laymans-drawing for (essentially step by step picture instructions lol). I've been trying to learn as much as I can to help people tell me what I need to do, it's coming along slowly.

I wish I could get in touch with the head engineer at Q3D, he helps a bunch of people and repairs the 3Dfx Voodoo 5 6000s that weren't released but engineer samples are all over. But no luck in that, and I don't feel like guessing his email (wouldn't be hard since I Know his name) and bugging him like that. If I could get ahold of him I could tell you guys the specifics on voltage ripple and the like.

I appreciate the help here, and I find this to be fun. I'd love to be able to just dive into something like this and make my own adjustments to how I personally would like something work And if there's anything I can do to repay someone for drawing me out some idiot proof plans, I'm open to that!

Thank you again.

 

You really need two PSUs, because you can´t hack them to give you 3.3 and 2.9 together.
For the reisitor, you can buy 5 20W ceramic resistors 0.47ohm each and connect them in parallel, they should handle 100W continuously and should stand 120W for a short time.

The PSU should have similar connection like in the attachement.

You don´t really care yet about all the parts on the left, you now need to locate most of the parts on the right side of the transformer. Then you should disconnect/desolder all RED, ORANGE, YELLOW and BLACK which are 5, 3.3, 12 and GND wires, leaving only one of each so you know what was where.

Then you have to find the two resistors called FEEDBACK in the picture, and see to which power line are they attached to and you should measure the voltage that is behind them, like in the junction of R25 and R26.

 

I think a regulator would be a much better way to go. No 2 power supplies are identical, so their voltages won't be either. This is a major issue.

Finding an exact match is not too likely either, so the schematic is informative, but probably won't help too much.

 

There is no need for matching, one will supply unmodified 3.3V, and the other will be modded to produce 2.9V.

 

You really need two PSUs, because you can´t hack them to give you 3.3 and 2.9 together.
For the reisitor, you can buy 5 20W ceramic resistors 0.47ohm each and connect them in parallel, they should handle 100W continuously and should stand 120W for a short time.

The PSU should have similar connection like in the attachement.

You don´t really care yet about all the parts on the left, you now need to locate most of the parts on the right side of the transformer. Then you should disconnect/desolder all RED, ORANGE, YELLOW and BLACK which are 5, 3.3, 12 and GND wires, leaving only one of each so you know what was where.

Then you have to find the two resistors called FEEDBACK in the picture, and see to which power line are they attached to and you should measure the voltage that is behind them, like in the junction of R25 and R26.

Wow! Now I feel a bit more confident about this actually happening...

For the voltages, I planned to leave the 3.3v alone seeing as I need it, and use the 12v bucked to the 2.9v. The reason behind that is because I won't be using the PSUs for anything but this project. Although I WOULD prefer not to desolder anything till I for sure have the setup working. Just in case either the card is bad, and the seller doesn't have anymore, or it flat out didn't go as planed. In either instance, I'd be able to use them as regular PSUs still, or at the very most sell them. HOWEVER! If you feel this plan isn't a good idea, possible how I want it, or too hard for a novice such as myself, then I'd be more than happy to go about which ever way you (or others) see as better.

Would I be able to run 6 of the resistors to allow for more wattage, or would that actually be the reverse in this instance? Seeing as 100W is only an assumption based on another card that is QUITE different than this one. The only similarity is the graphics chips being used and type of memory. Otherwise it uses two other chips that talk to their own set of 4 GPUs, and then communicate with a HiNT chip as the SLI bridge. So there's easily the possibility of it even requireing 150W. Either way if that all is the case, and I start @ 5(100W~) , I should just be able to solder on another resistor correct?

I think a regulator would be a much better way to go. No 2 power supplies are identical, so their voltages won't be either. This is a major issue.

Finding an exact match is not too likely either, so the schematic is informative, but probably won't help too much.

You are right, no two anything (except a clone pretty much) are alike. Processors are a great example where one can overclock like a mad man and the other is locked up tighter than Fort Knox!

If someone could provide me a method of doing this, I had planned to run some sort of fail safe so it doesn't drop below 2.7v and go above 3.0v, or maybe 3.5v would be a better ceiling. The one gentleman who I'm working with on our problem said with 3v they were getting fairly warm. That can easily be due to them getting SOME of the power they need and actually working. Or too little amperage and working too hard. Either way, I don't want to over-volt this card heh

Would a rectifier be a bad investment for this too? EDIT: To help cut down on any voltage ripple (if it even helps with that *shrug*)

 

You don´t have the card yet? I don´t know, but this can become quite lenghty because of this kind of discussion..

If the amps you need are higher than what the PSU can supply, you can have another problem because you may also have to bypass the overcurrent protection in it, again depending on the actual design of the PSU.

Would I be able to run 6 of the resistors to allow for more wattage, or would that actually be the reverse in this instance? Seeing as 100W is only an assumption based on another card that is QUITE different than this one. The only similarity is the graphics chips being used and type of memory. Otherwise it uses two other chips that talk to their own set of 4 GPUs, and then communicate with a HiNT chip as the SLI bridge. So there's easily the possibility of it even requireing 150W. Either way if that all is the case, and I start @ 5(100W~) , I should just be able to solder on another resistor correct?

The resistors will be only for testing the modded PSU if it can hold the needed current.

Let´s say you want to load it with up to 150W. Then you should have 8 20W rated resistors 0.47ohm resistors.
Each of the resistors added will load the PSU with cca 18W, provided it holds the desired voltage.
The total current for 150W is cca 50A, which is about a half more than the PSU should supply.



You have four options, based how good are your electronic skills:
a) find a cheap supplier of the 2.9V supply you need
b) modify some PSU, preferably used
c) make your own supply from a scratch

and d) make someone else do it

BTW do you have any instruments, at least DVM, and maybe even oscilloscope?

 

You don´t have the card yet? I don´t know, but this can become quite lenghty because of this kind of discussion..

If the amps you need are higher than what the PSU can supply, you can have another problem because you may also have to bypass the overcurrent protection in it, again depending on the actual design of the PSU.



The resistors will be only for testing the modded PSU if it can hold the needed current.

Let´s say you want to load it with up to 150W. Then you should have 8 20W rated resistors 0.47ohm resistors.
Each of the resistors added will load the PSU with cca 18W, provided it holds the desired voltage.
The total current for 150W is cca 50A, which is about a half more than the PSU should supply.



You have four options, based how good are your electronic skills:
a) find a cheap supplier of the 2.9V supply you need
b) modify some PSU, preferably used
c) make your own supply from a scratch

and d) make someone else do it

BTW do you have any instruments, at least DVM, and maybe even oscilloscope?

I have the card, it's sitting in the antistatic bag. What I don't have yet is the PSUs. I want to make sure this sort of mod is possible before I start shelling out cash for parts. If it is, and I can get a parts list, then I'll start purchasing the supplies.

Also, remember, I'm planning to run 2 PSUs with a total of at least 60A on any of the 3 rails. +3.3V@32A,+5V@34A,+12V1@16A,+12V2@17A. So supplying what I need (which I still estimate to only be around 35A) shouldn't be a problem.

I do have access to an oscilloscope actually. My friend just got one. Not sure what DVM stands for. Volt meter? If so I have one, but it's not high dollar, only can do 15A. And I think that fuse is blown so I'd need to replace it first heh. EDIT: Just came to me, Digital Volt Meter. So yes, I have one, but again it's only like a $45USD version. My friend might have a better one if it's needed.


EDIT: OK So I've been doing more looking around and on the site where that schematic was found, had a link to this site:
http://www.webx.dk/oz2cpu/radios/psu-pc1.htm

Is there any way that would help any of you explain to me what I need to do? >_>


Also, as for noise and ripple, which was mentioned earlier, I'd go by what this table says and assume that 50 would be a good target, and that possibly using the 5v may be better since it's already there?

Output-----------Maximum Ripple and Noise (mV p-p)
+12 V1DC--------------120
+12 V2DC--------------120
+5 VDC-----------------50
+3.3 VDC---------------50
-12 VDC---------------120
+5 VSB-----------------50

And to answer my own question to find out which lines I need for the 12V1 and 12V2 rails, I found out which ones.

 

Had a more detailed reply but Firefox/acrobat crashed. But basically I said that initially liked Bill's buck converter -it's what's used for getting 1.7V or so for your CPU on the actual motherboard from 3.3V. But without a good background in power electronics to design a top notch buck converter with synchronous rectification, the modding of the power supply would be the obvious choice. It would also be a LOT cheaper. You're just changing a couple of resistors really vs designing a pretty high current converter. This could be reversed if you document what you do and later want to use the PSU for 3.3V again!

Oh and consider taking your 2.9V reference from the card (and the large capacitance you'll put there) rather than measuring it at the power supply. Just be careful as there could be stability issues. The advice earlier that you test it out on resistors is EXCELLENT!

 

I think you should buy the PSU for the 2.9 rail, take it apart and see what you can do with it. You don´t have to cut the wires out, it just makes room for manipulation with the PSU.
If the operation doesn´t go well, you can always bring it back to the "regular PSU" state, even though it won´t have warranty anymore.

The only problem that could happen is that the PSU won´t be able to work stable with the lowered voltage, beacause it is not designed to work that way. But 3.3V to 2.9V is 12% less, so it should not be the case.

 

Well I'm all for doing the mods, but then my question is, what resistors do I swap to change the voltage? Or would that be something I'd have to take pictures of the card for your to actually tell me?

 

If you mean the card of the PSU, then yes.
You have to take the the board out of the box, take some photos of the top and the bottom of the PCB.
If you had the diagram of the exact PSU, then I could tell you right away, but as it is you need to do some tracing, tweaking and testing before you get to the finished modded supply.

From the picture I posted before it would be R25 and R26, and probably some more. But that is an old 200W PSU, so yours can be more like in the link you posted, so that would be different.

 

If you mean the card of the PSU, then yes.
You have to take the the board out of the box, take some photos of the top and the bottom of the PCB.
If you had the diagram of the exact PSU, then I could tell you right away, but as it is you need to do some tracing, tweaking and testing before you get to the finished modded supply.

From the picture I posted before it would be R25 and R26, and probably some more. But that is an old 200W PSU, so yours can be more like in the link you posted, so that would be different.

Alright. I'll see if the manufacturer will maybe give me a diagram of it, or worst case I'll buy just one and post pics.

Worst case, I can replace my AGE old 430w Antec that's in there. This PC is amazing, runs stable as hell, heavily overclocked (700mhz over stock on air cooling and still 30c range) and I have around 5 HDDs, a DL-DVD R/W, couple fans and a heavily overclocked x1950 Pro. I don't want to replace it due to possibly jinxing myself, but that's my worse case scenerio if things wouldn't work out haha

Or maybe just hair dryer the warranty sticker and safely peel it off >_>

If I get the specs and post em in here, should I PM you?

EDIT: If I use 1 PSU after modding, would I be able to use this to see what it's drawing and if I need the 2nd one? (If it's a retarded idea, just call me a retard lol) http://www.newegg.com/Product/Produ...c=EMC-IGNEFL090208-_-Gadgets-_-L0D-_-82715001

 

Then the other positive thing is that if the new PSU doesn't go well, you can try the old one too

It is not completely retard, but unless you already own it it just isn't worth the cash. The problem is that the current draw of a PSU isn't exactly sinewave and who knows if the meter can cope with it and display accurate measurements.
And anyway borrowig or buying an clamp-on current meter would be better.

 

Well, I fired an email off to the HQ and US branch of that power supply company asking for a schematic of that PSU, but no reply. I'm doubting I'll get one :\ So I'll just have to fork out the cash for one and snap some pics.