a cheap power supply?

[Philly_Boy]

Actually HiPro George has a well documented thread regarding a 5k watt PSU he built last year.

I hope your build goes well and you achieve the results you are looking for.

[Alexontherocks]

yes and I have read that thread. Very interesting stuff but let me point out:

- I have not seen the psu finished, but let us assume it has been completed and rocks as it should.

- can anyone build it? The layout Uncle Jimbo ha devised is surprisingly simple yet effective. Multiple volts, independent rails, limited number of components and none of them are very expensive.

- How much would a toroidal transformer cost you if you decided to build that psu?

- What would happen if it broke? To whom would you turn to? A market product would be too expensive to buy so repairs would have to be made. How many hardware failures would occur and what would be the total cost for the end user?

- The type of modular design we have here is bulky (but not by much. low wattage psus are not that big) but has the benefit of being custom in terms of number of units. The mutiple relays are for the amps envolved. Also with some attention to ventilation the PSus can be disposed in different manners. I have found the most efficient in terms of space on the floor vs height but combinations are possibile.

- How efficient would my system be compared to the one Hipro is building? Let us say I load each one of my psus at 50% capability or slightly more. Isn't that the point on the performance-to-watt curve where psus can achieve max efficiency?

I have not even started building the system but honestly it starts to sound a "distributed vs central" debate similar to the ones which often emerge when reading about computational power requirements in modern super computers.

Anyway I am just weighing pros and cons before ordering components. What needs to be addressed is probably the switching mechanism for the tec voltages but that can wait.

[littleowl]

no matter! we want to see lots of pics and I mean LOTS of pics.

[Uncle Jimbo]

You are gonna spend all this money on a high powered TEC set up....10x PSU's in a custom case...to achieve what temps?

I messed with a TEC set up last spring. A 437 watt TEC powered by a proper 24v meanwell PSU with proper guage wires connected from the TEC to the PSU. I put 2x PA120.3 rads, 2x DDC3.2 pumps (with Petra's top) and a D-tek Fuzion on this set up and my h2o temps still averaged 28-30c!!

All this to cool a QX6700 Kentsfield to -10*c idle and +3*c loaded. That allowed me to raise my OC from 3.8ghz (on straight water) to 4ghz. Big whup....

Needless to say I abandoned that set-up and have sold it.

I have a single stage unit on this rig now that I built myself. My CPU has -35c idle temps and -25c loaded temps. The single stage runs 24/7....something I would never trust for the TEC set-up lest it burn my house down....

Philly -
That's a very clean set-up, and those are great temps. Beautifully done.

The so-called '437' TECs currently being sold by ArcticSpider are not 437s at all - they are 360W Qc max units. The '437' on FrozenCPU is also not a 437 QC max unit, and is not the same as what is on FrozenCPU. I have been commenting on that on another thread - the original supplier for the 437 TECs quit making them and I think both sites have changed what they are selling but their documentation on the new TEC is not very transparent. See '437 watt TEC exposed as fake' for more detail -
http://www.overclock.net/peltiers-te...ml#post4033918

Running a 199 element TEC at 24V is a recipe for a big power bill and a hot water loop. It sounds like you did a great job on your TEC rig but got screwed by the people selling you the TEC.

With your setup, you were probably throwing over 800W into the cooling loop, and only getting 180W of cooling at 30C dT - not too good, and I can see why you were disappointed. The TEC just doesn't have the muscle to do much with that heat load. But if you could keep the cooling loop at 30C with that heat load, you had a great water loop, and could have gotten much better temps with the right TEC. You need something that moves your heat load with less than 75% of Imax. You get only 10% more cooling at full power, but use twice the power to get the job done.

A QX6700 Kentsfield is a hard load to cool, and none of the 'commercial' setups will do a good job. The right TEC for that load is a 24130, but those are hard to get. I'm currently working with a Chinese supplier to try and get some. At 24V, the 24130 is at 75% of Imax, total input power is about 500W, and you move over 250W with good differentials. So for less power in, the 24130 will move a lot more heat.

But phase cooling for that kind of load is the right way to go - it's hard to get big sub-zero temps with a heat load over 200W even with the best TECs.

[Alexontherocks]

Uncle Jimbo, I was wondering whether it is really necessary to have multiple relays (for a rated 120A on each line) when realistically I won't be using switching lines (for TECS) with more than 30/35A at most. In terms of maintanance and failure rate I'd rather have less relays at work and possibily separate the main psu assembly (powering my system) from the TEC power supply. I know this is slightly less flexibile but the beauty of this setup is that if I put a 4 psu system to the test (powering a TEC) I can simply add another 2 supplies and so on until my power requirements are sattisfied.

Don't get me wrong, your layout is very clever and I would be extremely happy to see it completed but I question the usefulness for me of having switching lines capable of 120A right now. For the moment I would like to have two independent units. One for the pc and for the TEC. I will put both to the test and would rather "isolate" any failure and be able to idenify which unit experienced failure. In the first few weeks this would also give me the possibility of changing elements rather quickly (less faulty components to identify) Given that I will be using screw type connectors and faston connectors, upgrading from a simpler design to the more integrated one you posted will be fairly easy.

What do you think? Also I see no reason not to upgrade the TEC power unit and add another 2 atx supplies if I find it necessary (multiple tecs in my system maybe?)

Also I would be able to separate the two units (independent power cables) and possibily stress test the TEC one in a more controlled environment (my desk) without having to detach everything from my pc (only computer in the house...damn )

[Uncle Jimbo]

Multiple volts, independent rails, limited number of components and none of them are very expensive.

Alex makes some good points. My original interest in this was because my nephew has a rig that really does need at least a 1200W supply, and he was set back by the price tag. But there are some other advantages.

The picture attached is of a triple-redundant supply from a Compaq Proliant server. This will deliver rated power all day every day for years. Any supply failure is isolated to that supply, and supplies can be hot swapped without taking the system down. There are heat, total power, and status outputs that can be used to monitor the state, and send notifications to the host machine. The chassis fits nicely into a standard tower case without much work - basically cutting out the back plate and screwing it in. The triplicated rack has dual thermally controlled 120mm fans with a total airflow capacity of over 300CFM, so no need for additional cooling fans.

This particular unit has triple 275W supplies. I bought 2 of these on eBay for $35 each. Power distribution is a little different than an ATX - the triplicated supply delivers 70A at 5V, 50A at 3.3V, and 36A at 12V.

Just below the triplicated supply is a dual redundant set from a different server - those have two 500W supplies with a similar mounting scheme. Those are rated more like an ATX - the redundant pair has 50A at 5V, 50A at 3.3V, and 50A at 12V. I got two of those units for about $70 each.

I also have a redundant 'rack mount' server supply setup. Those are 12V only, delivering 72A at 12V. If I need more 12V power, I could use those in place of, or in addition to, any of these other units.

So for a pretty low cost, I can build a monster supply that delivers all the power I could ever want, tells me if it's getting sick, allows replacement of failed supplies without turning off the unit, and gives me a wide selection of voltages. Also, spare parts are easily available.

[Uncle Jimbo]

Uncle Jimbo, I was wondering whether it is really necessary to have multiple relays (for a rated 120A on each line) when realistically I won't be using switching lines (for TECS) with more than 30/35A at most. In terms of maintanance and failure rate I'd rather have less relays at work and possibily separate the main psu assembly (powering my system) from the TEC power supply. I know this is slightly less flexibile but the beauty of this setup is that if I put a 4 psu system to the test (powering a TEC) I can simply add another 2 supplies and so on until my power requirements are sattisfied.

Don't get me wrong, your layout is very clever and I would be extremely happy to see it completed but I question the usefulness for me of having switching lines capable of 120A right now. For the moment I would like to have two independent units. One for the pc and for the TEC. I will put both to the test and would rather "isolate" any failure and be able to idenify which unit experienced failure. In the first few weeks this would also give me the possibility of changing elements rather quickly (less faulty components to identify) Given that I will be using screw type connectors and faston connectors, upgrading from a simpler design to the more integrated one you posted will be fairly easy.

What do you think? Also I see no reason not to upgrade the TEC power unit and add another 2 atx supplies if I find it necessary (multiple tecs in my system maybe?)

Also I would be able to separate the two units (independent power cables) and possibily stress test the TEC one in a more controlled environment (my desk) without having to detach everything from my pc (only computer in the house...damn )

Alex - the only reason to use this setup is to take advantage of the unused capacity on the base PSU's. If you want to split it into a Max ATX supply and a separate TEC supply, then you get down to a pretty small number of relays.

Let's say you do the 6 as a big ATX supply, and then take the remaining 4 and do a 2+2 setup for TECs. If you are only using 1 or 2 TECs, you will have plenty of juice. You could probably do that with a single relay for each selection position, since it is unlikely that you would need more than 30A even for 2 TECs if you are in the good efficiency range.

That would require only 5 relays, or 6 if you provide a 12V float and 24V output.

[Alexontherocks]

Alex - the only reason to use this setup is to take advantage of the unused capacity on the base PSU's. If you want to split it into a Max ATX supply and a separate TEC supply, then you get down to a pretty small number of relays.

Let's say you do the 6 as a big ATX supply, and then take the remaining 4 and do a 2+2 setup for TECs. If you are only using 1 or 2 TECs, you will have plenty of juice. You could probably do that with a single relay for each selection position, since it is unlikely that you would need more than 30A even for 2 TECs if you are in the good efficiency range.

That would require only 5 relays, or 6 if you provide a 12V float and 24V output.

I understand. Since I have no problem really in building one or the other (apart from maybe controlling the relays easily) I guess I'll go with the full design, you are right about tapping into the unused 3.3. and 5 lines.....I ask forgiveness for I have sinned and strayed away from the path

Only real problem is controlling the relays..... suggestions?

[railmeat]

best $50 p\s in the world is the 450 watt FSP on newegg.its a tank and peformas like 1.. type in FSP in newegg search bar

[Uncle Jimbo]

I understand. Since I have no problem really in building one or the other (apart from maybe controlling the relays easily) I guess I'll go with the full design, you are right about tapping into the unused 3.3. and 5 lines.....I ask forgiveness for I have sinned and strayed away from the path

Only real problem is controlling the relays..... suggestions?

To switch all the voltages, the easiest way is to use an 8 position 3 pole rotary switch. You need a break before make type switch. The lines in the table represent the leads to the relays, which have the same markings. The 0V float connections are just open. Common is 12V (probably through a small resistor, like a 10 ohm 5 watt, just in case you short something). Each switch position will select the relays on that line. You probably want to be sure the relays are being selected correctly before putting real power on them.

[Alexontherocks]

best $50 p\s in the world is the 450 watt FSP on newegg.its a tank and peformas like 1.. type in FSP in newegg search bar

possibly so but for 50$ I got 10 500w supplies.

[Alexontherocks]

just out of curiosity....should 4 or 6 or 10 powersupplies be connected in parallel, the modification to the psus should be made to all but one correct? Or should it be done to half of the number of supplies in the rig?

Also: is there any chance (have to know for safety precautions) that the atx supplies will charge each other and explode? Sounds dumb but many have pointed out this possibile outcome. Is there any risk of this happening?

Ivestigating the physics envolved in the project...

[Uncle Jimbo]

just out of curiosity....should 4 or 6 or 10 powersupplies be connected in parallel, the modification to the psus should be made to all but one correct? Or should it be done to half of the number of supplies in the rig?

Also: is there any chance (have to know for safety precautions) that the atx supplies will charge each other and explode? Sounds dumb but many have pointed out this possibile outcome. Is there any risk of this happening?

Ivestigating the physics envolved in the project...

Alex -
The PSUs in the 'base' - 6 in your setup - do not require any modification at all. Only the 'float' supplies need to have earth ground isolated. That is true if you use one in the base and 1 in the float, or if you use 20 in each.

There is no chance the supplies will explode even if you wire it wrong - you might blow a fuse. People who do idiot stuff like connecting the red wire on one MOLEX to the black on another to get 17V will get a home-made toaster instead... but you seem to know what you are doing, and if you use a little care, you should have no issues.

I have had some people question the idea of connecting supplies in parallel, because one might have 12.3V on the 12V rail, and another 11.8V. I did some testing to see what would happen, by intentionally putting mis-matched supplies in parallel. The answer is that the one with the higher voltage will take a larger share of the amperage load, but aside from that, there is no other effect. They don't 'fight' each other.

That's why it is a good idea to use identical supplies in a parallel setup. All of these supplies also have a fine tuning pot internally to trim the voltage, and it might not be a bad idea to set them up to have the same voltage into a medium load - say half of full power - to keep load sharing equal. But I have found in practice that it doesn't make much difference. The regulation on these supplies is not that tight, and the supply carrying the larger load will have its voltage droop a bit, so they equalize. It's like the load sharing resistors in large parallel power circuits (a small resistor between collector and ground) - the increase in current causes self-regulation.

[Alexontherocks]

Thanks for the info. I was pretty sure about what to do to build the system you designed. I just wanted to understand for myself so I looked into the physics behind "floating" a supply (I have no specific education in electronics) and came up with several sites.

Also this came up:

http://www.procooling.com/index.php?...s&disp=52&pg=1

I tried to understand how multiple supplies work when combined. Floating supplies is also used to increase precision of the output voltage or so it would seem from some sites I found on the web.

Anyway it would appear that in a simple parallel build only one psu should be grounded, the others should be "floating supplies". From the article it seems that for any numer of psus in parallel or series only one should be grounded.

My question is: what would happen (in a hypothetical build) if, for any reason, that single psu (connected to ground) failed and the ground connection got severed? Wouldn't that be dangerous and ultimetely affect the voltage values of the output lines?

[Uncle Jimbo]

Thanks for the info. I was pretty sure about what to do to build the system you designed. I just wanted to understand for myself so I looked into the physics behind "floating" a supply (I have no specific education in electronics) and came up with several sites.

Also this came up:

http://www.procooling.com/index.php?...s&disp=52&pg=1

I tried to understand how multiple supplies work when combined. Floating supplies is also used to increase precision of the output voltage or so it would seem from some sites I found on the web.

Anyway it would appear that in a simple parallel build only one psu should be grounded, the others should be "floating supplies". From the article it seems that for any number of psus in parallel or series only one should be grounded.

My question is: what would happen (in a hypothetical build) if, for any reason, that single psu (connected to ground) failed and the ground connection got severed? Wouldn't that be dangerous and ultimately affect the voltage values of the output lines?

Alex -
Your last statement is correct - having all of the base supplies connected to earth ground is an important safety consideration. That's a nice article but not written by an engineer. Just for reference, I am a graduate electrical engineer with over 20 years experience, including design of high current switching power supplies from the ground up.

Let's talk about the 'base' supplies and parallel hookup. If earth ground is connected to the black wires on one of those, it is connected to all of them. Some people think that having multiple earth grounds will cause 'ground loops'. This is false. There should be no current in the earth ground - if there is, something is not wired right. The earth ground is there for safety, and every one of the base supplies should be connected. Not only is it safe to leave the earth ground connected on the base supplies, it is safer than disconnecting it.

Also. modern supplies do not need a big dummy load on the 5V line. That has been true since the development of the PC97 ACPI spec, which is the same spec that made idle power available with the supply powered down. The spec says the supply has to maintain regulation with a draw as low as 300mA. Think for a moment about a sleep state where the only power is to a USB mouse - obviously a big resistor is not needed with any supply made to spec in the last 10 years.

Floating a supply does not increase precision - it degrades it. The total regulation offset is the sum of offsets of the supplies in the stack. This is not a big deal for computer PSUs as the voltages are all regulated again on the component boards. But if you wanted to build a tightly regulated supply, you would need to do a lot of work to the internal feedback to get precision.

I built a supply for an instrument control system that delivered 100A at 24V with better than .005% regulation from 1A to 100A - that means that the voltage at the load dropped only .1V from essentially no load to 100A load. The supply used an operational amplifier with open loop gain of over a million, and also temperature sensors to predictively adjust the voltages to account for component variations.

The biggest source of voltage drop between a PSU and load is the wiring to the load, so a precision supply has to sense the voltage drop in the wire and adjust for it. To do that, the supply has to have sense lines which can be attached directly to the load. The Meanwell supplies have these sense leads. No current goes through the sense lines, so they can pass back the actual load current to the regulating elements.

[NotSoCoolJ]

Hmm, thats very interesting. I was wondering about that. I have seen people just parallel ATX supplies before without separating one of the supply's grounds from the chassis ground first. I was told that this would pose problems. Don't get me wrong, I believe you. I just think I may be misunderstanding what your saying. I have never done this before and I may want to at some point in the future. I was always leery of doing something like this in the past because of conflicting information on the subject.

Granted the procooling info is for AT supplies and not ATX. but here is a direct quote.

"IF THE BLACK LEADS ARE COMMON TO CHASSIS GROUND IN THESE SUPPLIES AND YOU TRY TO CONNECT THEM IN SERIES OR PARALLEL THEY WILL SHORT OUT PERHAPS DAMAGING THE SUPPLIES!!!"

I really fail to see how this is would short out anything in a parallel configuration. If they are in series, how would that work without creating a direct short to ground?

[Uncle Jimbo]

Hmm, thats very interesting. I was wondering about that. I have seen people just parallel ATX supplies before without separating one of the supply's grounds from the chassis ground first. I was told that this would pose problems. Don't get me wrong, I believe you. I just think I may be misunderstanding what your saying. I have never done this before and I may want to at some point in the future. I was always leery of doing something like this in the past because of conflicting information on the subject.

Granted the procooling info is for AT supplies and not ATX. but here is a direct quote.


I really fail to see how this is would short out anything in a parallel configuration. If they are in series, how would that work without creating a direct short to ground?

AT supplies are no different - I have several supplies identical to the ones he modded. This is basic electricity and has nothing to do with the design of the supply.

The base supply, whether a single supply or a parallel group, should always have earth ground connected. In series the earth ground MUST be disconnected from the 'floated' supply (or group of parallel supplies) or you will get some very hot wires. It's only common sense - the 'ground' for the floated supply is actually at some other voltage, so connecting it to ground shorts out the base supply.

[jspace]

All you need to build a DC supply is a transformer, some diodes (or a diode bridge), and a handful of capacitors

[Alexontherocks]

I am really grateful to Uncle Jimbo for shedding light on the subject. Unfortunately in the the pc business (or should I say modding business) you are likely to find very creative and ingenious individuals but not that much theory to back up every successful modification.

This is all very interesting and I'd rather know what I am doing rather than simply building. It is a good opportunity to gain a better understand of basic principles which may prove useful for any modification to power supplies in the future.

[leuler]

My credentials pale in comparison to Jimbo's, but there is no doubt that
everything that he has said is correct. The only possible reason for not
grounding all the the base PSUs is IF AND ONLY IF they are
not supplied by ( plugged into ) the same circuit. It is possible to find
that the earth grounds for different circuits to have a potential difference
( though it should be small ) , causing a flow of current between the
grounds ( ground loop ).

If the house wiring was done competently, this ground loop current would
be low or non-existent. Would this cause a problem ? With A/V equipment
that are powered by different circuits and are connected together by
A/V cables, it is a source of noise. For parallel PSUs, I don't think so.

Even if a ground loop causes a problem, here is the proper and safe thing
to do. Ground all of the base PSUs. Always do that. Then,
IF the PSUs don't share the same circuit, physically ground the PSUs
together. There are multiple ways of doing this, but the basic idea is that
the earth ground is then common for all the base PSUs and then no current
flow between the PSUs on the earth ground is possible.

Most likely, none of this would be necessary. I was just trying to emphasize
that you should always ground the base PSUs.

[leuler]

This thread or a new one like it ( how to guide ) should be stickied. The
discussions lately in this forum involve some ideas using TECs that use
more than one PSU. Given that it can be more cost effective to mod PSUs
meet the power requirements of these designs and not difficult to do with
the right info available AND can be done safely, it would be
good if the info didn't get buried.

[Alexontherocks]

This thread or a new one like it ( how to guide ) should be stickied. The
discussions lately in this forum involve some ideas using TECs that use
more than one PSU. Given that it can be more cost effective to mod PSUs
meet the power requirements of these designs and not difficult to do with
the right info available AND can be done safely, it would be
good if the info didn't get buried.

My thoughts exactly...

I believe that multiple psus is the way to go for heavy load applications and hardware. If you look at the 280 chip from nvidia under load you can see that the power requirement bar has been raised. 1000w psus can be considered a safe bet for the next year or two of high end graphic cards. Add sli or tri sli (or quad sli) and 1200w psus seem an even safer choice.

I don't know much about electronics but it only makes sense that extracting all that power from a single array of components is taxing on the components themselves. Efficiency suffers and cost/maintenance rises.

I hope that more efficient solutions (interestingly enough developed for no-fuss applications) can only be adopted in the home pc industry.

From an "historical" point of view it also makes sense that after RAID/Multiple cpu Cores (or even multiple cpus)/multiple stream processors and multiple vga cards even PSUS undergo a similar evolution even in the domestic field of computer hardware.

I may be wrong but it seems a logical outcome...

[Uncle Jimbo]

All you need to build a DC supply is a transformer, some diodes (or a diode bridge), and a handful of capacitors

True - but simple transformer based DC supplies are big, heavy and expensive, and provide no regulation. I have a 24 VAC 20A transformer - it weighs 30 lbs.

For most people, using off the shelf PSUs is a much easier thing to do.

[Uncle Jimbo]

This thread or a new one like it ( how to guide ) should be stickied. The
discussions lately in this forum involve some ideas using TECs that use
more than one PSU. Given that it can be more cost effective to mod PSUs
meet the power requirements of these designs and not difficult to do with
the right info available AND can be done safely, it would be
good if the info didn't get buried.

I also agree. The questions asked are very insightful and provide a good sense of what concepts are hard to understand.

If I can find the time, I intend to sort from several threads into a FAQ type format, but with pictures. I am planning on at least three threads -
1 - this one, on various ways of using standard PSUs to get higher voltage or more current.
2 - Controlling TECs from desired load temperature
3 - Multipe TECs, undervolting, and how to select the right TEC for the job

There is a lot of very good material on the various threads, it just needs to be organized and cleaned up.

[Alexontherocks]

I also agree. The questions asked are very insightful and provide a good sense of what concepts are hard to understand.

If I can find the time, I intend to sort from several threads into a FAQ type format, but with pictures. I am planning on at least three threads -
1 - this one, on various ways of using standard PSUs to get higher voltage or more current.
2 - Controlling TECs from desired load temperature
3 - Multipe TECs, undervolting, and how to select the right TEC for the job

There is a lot of very good material on the various threads, it just needs to be organized and cleaned up.

Please do so if you have time. I will try to supply pics of the build so that there is a visual reference for the part envolving rigging the supplies according to your layout.