Worldwide EA-300 Re-upgrade

[Pentium4]

I recapped this guy a while ago. It ran for 6,000 hours after a recap and 12V rectifier upgrade https://www.badcaps.net/forum/showpo...&postcount=851

The thing runs hot, as you can imagine, so I'd like to do some more mods to it for fun
I had originally replaced the original 16A ultra fast with two 12A ultra fasts. I don't want to waste shottky rectifiers on it but I do have two matching 20A ultra fasts I will throw in there. And since the 5V and 3.3V use the same 30A rectifier, I'm going to replace that one with two matching 30A packs, or even 40A since I have so many of them. This is basically to increase the efficiency.

Is it even worth trying to replace the FET for the 3.3V with something better?

I wanted to replace the single toroid with something better but can't find any scrap PSU's with a similar design. I'm hoping that the increased rectifiers will help reduce the heat in that area.

I can't believe I didn't do this before, but I changed the values of the load resistors on the secondary. The 12V and 5V ones were running so hot that the wires were melting! It had 100Ω on the 12V, 15Ω on the 5V, and 50Ω on the 3.3V. I changed them to 450Ω on the 12V, 50Ω on the 5V, and 100Ω on the 3.3V. Just running it with shorting the PS_ON before and after, the idle went from 14W to 11W!

[Pentium4]

Re: Worldwide EA-300 Re-upgrade

I do have a FET with lower RDS ON
Original:
On hand: https://cdn.badcaps-static.com/pdfs/...693bb1ffbc.pdf

[Pentium4]

Re: Worldwide EA-300 Re-upgrade

Okay so I got two 20A ultra fasts on the 12V, and two SBL3045CT's on the 5V and "3.3V" I left the FET for now. It's amazing how much cooler it runs already. I'll upload some pictures tomorrow.

[kaboom]

Re: Worldwide EA-300 Re-upgrade

I do have a FET with lower RDS ON
Original:
On hand: https://cdn.badcaps-static.com/pdfs/...693bb1ffbc.pdf

RDS doesn't matter for a linear regulator. It only matters when the device saturates, like in a switching regulator.

For a linear regulator's pass transistor, you're interested in power dissipation.

Before derating, the replacement device is 20W better. Be sure to use a mica insulator to keep device-to-heatsink delta as low as possible.

[Pentium4]

Re: Worldwide EA-300 Re-upgrade

Excellent. Thank you for pointing that out to me or I would have had no idea. So I did a specific test to see how much wattage was being used, and also made note of the voltages. So I'm going to replace that FET with the better one, and then do the exact same test to see what kind of difference it made to wattage used and voltages

[kaboom]

Re: Worldwide EA-300 Re-upgrade

Excellent. Thank you for pointing that out to me or I would have had no idea. So I did a specific test to see how much wattage was being used, and also made note of the voltages. So I'm going to replace that FET with the better one, and then do the exact same test to see what kind of difference it made to wattage used and voltages

You're not going to see any difference.

Take a common 7805. At 10v in and a 1A load, dissipation is 5W. Now "current boost" it (http://www.repairfaq.org/sam/samschem.htm#schslp2), giving much higher capacity.

The "boosted" circuit will still dissipate 5W at 10V in and a 1A load.

Linear regulators aren't used on motherboards much anymore because of the large amount of heat they produced. First to go with DC-DC converters were CPU regulators, back in the mid-late 90's. Then memory and chipset supplies used them. Some M/Bs as recently as P4s used linear regulators for chipset and RAM.

The only thing you'll achieve with the upgraded MOSFET will be higher allowable max temp for that heatsink. Anytime you go with a higher Pd device for a linear reg's pass transistor this will happen.

It is not done to improve efficiency.

More sensible supplies place a magamp choke in of the 3.3V rectifier's legs. One leg goes directly to one of the "5V" hots from the TX secondary, the other goes to the magamp choke and then to the other "5V" leg.

The magamp choke has an extremely high inductive reactance with zero DC current through it. So at this point, the 3.3 rectifier is only fed by one of the transformer secondary legs. Our 3.3VDC is around 2.4-2.8V at this point.

There is another TL431 based circuit in the vicinity of the 3.3V filter caps and output inductor. It controls a small transistor, that backfeeds DC, via some 4148s and sometimes UF4004s, up the magamp choke, towards and into, the transformer secondary, returning to GND/common via the center tap. It's a small DC current to keep the transformer happy, but more than enough to greatly reduce the XL of the magamp choke.

With it's XL reduced, the magamp choke allows more current to feed the associated leg of the 3.3V rectifier, causing the output voltage to increase. Which doesn't happen til after the 3.3V output inductor; as in any SMPS, the output inductor (and filter caps) is a duty cycle to voltage converter.

As the load changes on 3.3-out, the feedback circuit sends more or less DC into the magamp choke, raising or lowering the output voltage, achieving regulation of the 3.3V output.

You can clearly see this in the diagram. L6 is the magamp choke. That example is interesting- the 3.3V output inductor is wound on the same toroid as the other outputs. It usually has its own toroid.

The other outputs are common wound to improve their cross regulation. Since 3.3 is independently regulated, this is not required or desirable.


Then there are the big supplies, which are nothing more than 12V-out units capable of 60-100A. Usually with synchronous rectification- a must at those output currents. And a PFC booster feeding the highside DC bus- a simple caps and diodes doubler would pull massive current spikes just before the peaks of the AC waveform as the caps recharged.

The single 12V output goes thru several shunts, for current monitoring to satisfy UL's volt-amp requirements of no single output exceeding a given capacity. Say there are four shunts, monitored by four opamps- a "four rail 12V" supply. If the voltage across a given shunt goes too high (too much current), the output of its respective opamp is used to kill the main PWM.

The other outputs in these beasts are derived from DC-DC converters, exactly the same way a motherboard buck-converts +12 into VCore. Usually just +5 and +3.3 though. -12 is of such low current that it can be taken care of with a few diodes off the secondary of the main transformer. With fusible resistors of course.

[Pentium4]

Re: Worldwide EA-300 Re-upgrade

You're not going to see any difference.

You're absolutely correct. Not a single change in anything.

Then memory and chipset supplies used them. Some M/Bs as recently as P4s used linear regulators for chipset and RAM.

Yep, easy to spot. You can usually see how discolored that section of the board is

More sensible supplies place a magamp choke in of the 3.3V rectifier's legs. One leg goes directly to one of the "5V" hots from the TX secondary, the other goes to the magamp choke and then to the other "5V" leg.

The magamp choke has an extremely high inductive reactance with zero DC current through it. So at this point, the 3.3 rectifier is only fed by one of the transformer secondary legs. Our 3.3VDC is around 2.4-2.8V at this point.

There is another TL431 based circuit in the vicinity of the 3.3V filter caps and output inductor. It controls a small transistor, that backfeeds DC, via some 4148s and sometimes UF4004s, up the magamp choke, towards and into, the transformer secondary, returning to GND/common via the center tap. It's a small DC current to keep the transformer happy, but more than enough to greatly reduce the XL of the magamp choke.

With it's XL reduced, the magamp choke allows more current to feed the associated leg of the 3.3V rectifier, causing the output voltage to increase. Which doesn't happen til after the 3.3V output inductor; as in any SMPS, the output inductor (and filter caps) is a duty cycle to voltage converter.

As the load changes on 3.3-out, the feedback circuit sends more or less DC into the magamp choke, raising or lowering the output voltage, achieving regulation of the 3.3V output.

You can clearly see this in the diagram. L6 is the magamp choke. That example is interesting- the 3.3V output inductor is wound on the same toroid as the other outputs. It usually has its own toroid.

The other outputs are common wound to improve their cross regulation. Since 3.3 is independently regulated, this is not required or desirable.

I found this fascinating. Thanks for taking the time to write it out in more simple terms for me, I appreciate it Is the way that it's independently regulated, the reason that even cheap PSU's will have such a clean and steady output on the 3.3V rail? And like momaka said, this linear regulation creates clean power, so with those very low ESR Rubycon ZLJ caps, this thing probably has <5mV of ripple?
I'm still a little bit confused (I'm a slow learner) so please bear with me on how the 5V and 3.3V use the same set of rectifiers, and are also regulated from the same toroid. Once the 5V out is rectified, is the voltage then stored in the secondary of the transformer, or the toroid coil to then be stepped down and rectified to 3.3V by the FET? Or, is the regulated 5V then stepped down to 3.3V to then filtered by the capacitors? I find linear regulation interesting, even though it's obvious why using a mag amp is the smarter method to use.

I was tired of that stupid voltage selector switch getting in the way, so I used your trick to make it permanently 120V, and then put an appropriate sticker on the back I also added two MOV's, and switched out the 330uF YC input caps to 470uF Lelon.

I also switched out the 2200uF 10V ZLJ caps on the 5V to nichicon PW 2700uF 6.3V. I've been a little too paranoid lately with using such low ESR caps in PSU's like these, unless for instance it's on that linear generated 3.3V rail. I also tightened the switchers on the primary heatsink. They were not on there very tight. Believe it or not, the primary heatsink runs hotter than the secondary one.

Check out the original washer for the original FET. It got so hot that it discolored! This looks like a nice PSU now that I feel I can trust in something that won't use more than 100W. I moved the fan thermistor to a cooler part of the PSU, since it would just rev up when it was in the toroid. Obviously that small toroid is being stressed quite a bit. With an 80W DC load after all the improvements, it gets up to 55C. I know that's not too hot for it, but it seems pretty warm for such a load.

[Pentium4]

Re: Worldwide EA-300 Re-upgrade

Also, do you think this PSU would do better in a 12V based or 5V based system? Because as far as I'm aware, 5V based systems also use more 3.3V than 12V based systems. And this is where the extra heat in the PSU would come into play. But I'm also curious in general as to which type of system would be harder on that toroid coil.

[everell]

Re: Worldwide EA-300 Re-upgrade

Here is a drawing of a typical 3.3 volt mag amp circuit. It might be a little easier to follow than the one KaBoom posted.

[kaboom]

Re: Worldwide EA-300 Re-upgrade

Not a single change in anything.

Yes, but that heatsink can now run hotter before that MOSFET dies.

Yep, easy to spot. You can usually see how discolored that section of the board is

Or not... Intel D865GBF boards used Nichicon VR caps for general purpose decoupling, including for the NB and SB. Those often started acting funny, or perhaps, slower and quirky. I think I junked mine back in early 2008- I wasn't crazy about S478 at that point. Maybe on a better board I would've recapped it, but the 865G was a "Celeroned" northbridge. Leave it to Intel... Shoulda got an 865PE...

And I probably took for granted the Nichicons being good- those 85 degree caps often went bad.

I found this fascinating. Thanks for taking the time to write it out in more simple terms for me, I appreciate it Is the way that it's independently regulated, the reason that even cheap PSU's will have such a clean and steady output on the 3.3V rail?

It is fascinating. You'll find that 3.3 is always independently regulated. The first ATX supplies used an opamp or '431 using a MOSFET as a linear pass device. Lowest ripple of all, but lots of heat. Also, all current supplied by 3.3-out comes from the +5 rectifier, which must be sized for the load on +5, as well as +3.3.

Then came the magamp-derived +3.3. Now we had two output rectifiers, one each for 3.3 and +5. The one for +5 has [b]both[/i] anodes connected to the transformer secondary. But the one for +3.3 has only one of its anodes directly connected to the tx; the other goes thru that magamp choke.

The supplies with magamp 3.3 always have two seperate rectifiers for +3.3 and +5. In fact, if you remove the magamp choke and replaced it with a jumper wire, the 3.3V ouput would become a second 5v output! And the supervisor or comparator, if present, would kill the PWM for "output overvoltage."

"But if both the 3.3 and 5 volt circuits are so similar, why are there two different output voltages?" Good question! With the magamp choke's inductive reactance, and its modulation by DC current, itself modulated to keep output voltage constant, we are effectively reducing the AC input voltage to the +3.3 rectifier.

The magamp circuit must have its own rectifier, and, by extension, its own output inductor (toroid).

And then the Big Silverstones and Corsairs, with the DC-DC converters, deriving 3.3 from +12.

And like momaka said, this linear regulation creates clean power, so with those very low ESR Rubycon ZLJ caps, this thing probably has <5mV of ripple?

You mean in your supply with the MOSFET in series, between +5 and +3.3? Then yes, you'll have very little ripple- that entire circuit works like a giant 7805. Or 7803.3...

I'm still a little bit confused (I'm a slow learner) so please bear with me

A learner is a learner, no matter what. And far better than an "antecwhore," "antecpimp," or even an "antecrep."

Once you begin to figure out even a little of this stuff, the rest of the pieces fall into place.

on how the 5V and 3.3V use the same set of rectifiers, and are also regulated from the same toroid.

The output inductor is not there for regulation. It's there to keep the primary switching devices and output rectifiers from exploding, to keep ripple into the first filter cap (per output) reasonable, and most importantly, to allow a constant-amplitude, variable duty cycle squarewave to become a smooth DC voltage.

Take the 12V output in a typical half bridge supply. The common cathode of that rectifier goes from ~20V to zero on each alternation (twice per cycle). How come the output is "only" 12V? Because the pulse width is variable. Pulse [b]width[b/] modulation- the voltage is always the same amplitude; the width is varied according to line voltage and load demand.

Assume a 60% duty cycle to the input of that inductor. But how is there a constant DC output with a pulsing input? Because of the output filter!

Flux builds in the output inductor during the the on time, and current begins to flow. As this current increases, the output cap charges up. Suddenly, we're in PWM off time, with no diode current supplied by the primary. While the flux collapses, the current that is still flowing in the inductor continues to flow; out the end of the inductor into the cap, out of the cap into ground/common, from common into the tx center tap, out of the tx into the rectifier, and (finally) back to the other end of the inductor.

Incidentally, forward converter supplies always have a diode with its anode to common. Transformer current and inductor freewheel current have their own diodes. A dual-common-cathode rectifier will have the cathode going to the high side of the output inductor; one anode to the tx hot, the other to gnd/common. The return path thru the tx is blocked or infeasable during off-time and cannot be used to commutate the inductor current as in a half/full bridge, hence the dedicated return diode.

This is why it's referred to as an output inductor; it stores and releases energy. Whereas the magamp uses a choke or reactor; we want to choke off, retard or hold back, the incoming AC. In the case of a magamp, we also want to modulate this degree of "choke-ing."

The magamp does this with only minimal core and copper loss- far less heat than a linear regulator.

Now the multi-winding output inductor, as used in computer supplies, does improve crossload regulation, but that's due to the sort of "transformer action" that occurs- if current flows on winding X, Y and Z get helped along.

That means all the rails supplied by such a common output inductor (+/-5, +/-12) usually have a minimum current they must be loaded to.

Which is another thread... d:

And the old AT supplies had yet their own reason for a minimum load, which I will get to.

Once the 5V out is rectified, is the voltage then stored in the secondary of the transformer, or the toroid coil to then be stepped down and rectified to 3.3V by the FET? Or, is the regulated 5V then stepped down to 3.3V to then filtered by the capacitors? I find linear regulation interesting, even though it's obvious why using a mag amp is the smarter method to use.

Yes- the "finished product" 5V is used to feed the 3.3v regulator ckt. It isn't "stepped down" as much as it's "burned away." Linear regs...

In either case, the +5 out is derived in the usual way. With a 5->3.3 linear regulator, the very same 5V output that feeds the red output wires also feeds that MOSFET and related circuitry, possibly with a 12V feed for increased gate-drive voltage. With the magamp 3.3, the +5 out just feeds straight to the red wires.

Again with either one, the feedback opamps/references would be fed a sample of +5 thru a resistor network.

There is no energy storage in the transformer, except from leakage reactance, which is a parasitic aspect of a real world component, hence the snubbers.

I was tired of that stupid voltage selector switch getting in the way, so I used your trick to make it permanently 120V, and then put an appropriate sticker on the back I also added two MOV's, and switched out the 330uF YC input caps to 470uF Lelon.

Nice...

I also switched out the 2200uF 10V ZLJ caps on the 5V to nichicon PW 2700uF 6.3V. I've been a little too paranoid lately with using such low ESR caps in PSU's like these, unless for instance it's on that linear generated 3.3V rail.

Right, you don't want to make the regulating action "hunt" around.
+12:


-12:


The actual switching ripple are the individual vertical green lines. Then there are about thee sawtooth envelopes per div, spanning 2.5-3 vertical divs from baseline- something's compensated poorly, making it have worse ripple than it appears to. Those are actually oscillations in the output voltage; the feedback loop is repeatedly overshooting, then undershooting its setpoint. And finally, you can just barely see some low-frequency line-freq ripple making it thru- there are actuall three waveforms on that trace.

The "wonderful" test mule:
http://hardwareinsights.com/wp/the-2...ply-roundup/2/

I also tightened the switchers on the primary heatsink. They were not on there very tight. Believe it or not, the primary heatsink runs hotter than the secondary one.

This is why they often blow up. I've seen bad examples of assembly in pix of reviewed units. No lockwashers.

Check out the original washer for the original FET. It got so hot that it discolored! This looks like a nice PSU now that I feel I can trust in something that won't use more than 100W.

I moved the fan thermistor to a cooler part of the PSU, since it would just rev up when it was in the toroid. Obviously that small toroid is being stressed quite a bit. With an 80W DC load after all the improvements, it gets up to 55C. I know that's not too hot for it, but it seems pretty warm for such a load.

While it's surprising, that's not too bad. That will heat up once continuous inductor current is established, probably once you've got 5A/2A on 5V/12V.

[kaboom]

Re: Worldwide EA-300 Re-upgrade

Also, do you think this PSU would do better in a 12V based or 5V based system? Because as far as I'm aware, 5V based systems also use more 3.3V than 12V based systems. And this is where the extra heat in the PSU would come into play. But I'm also curious in general as to which type of system would be harder on that toroid coil.

I recently did what I once thought was impossible- brought a P4P800-E back to life.

There's actually quite a bit that runs off 3.3, directly or indirectly. Both NB and SB do, after linear onboard regs, as well as the RAM supply.

I also have a partially recapped Radeon 9500 pro in there. It has a dual buck converter, taking both +3.3 and +5 down to its Vcore, and three more +5 fed DC-DCs for its onboard RAM.

The linear-regulated 3.3V is a holdover from olden times when not much used +3.3 and the resulting heat wasn't as much of a problem. You may want to put that fan thermistor near the 3.3 pass MOSFET.

Here is a drawing of a typical 3.3 volt mag amp circuit.

Looks good. You can tell it's a forward converter w/o even looking at the primary. See D30 with one anode to GND?

It might be a little easier to follow than the one KaBoom posted.

KaBoom? No, just "kaboom," all lowercase, nothing blew up today!
And it's not an "earth shattering" kaboom, either...

[kaboom]

Re: Worldwide EA-300 Re-upgrade

That supply was made in a magamp version as well.

I've traced it out and drawn in missing components.

See how the pass MOSFET for +3.3 is located at D22 on the board?

Two spots where magamp chokes are missing? What looks like a spot for a large "pencil" inductor on the secondary, near +3.3?

I knew I saw this one before- nearly the same supply that came in a tigerdirect case I had years ago. Mine powered an Albatron KM266 and 1.3GHz Duron.

P4, yours has a bigger transformer and output wires. Mine had 20 or 22AWG for everything! Yours has a trimpot, mine had precision resistors instead.

And mine certainly didn't come with as nice a fan as yours...

They do run warm- you'd think mine was feeding a hungry Prescott!

[momaka]

Re: Worldwide EA-300 Re-upgrade

I can't believe I didn't do this before, but I changed the values of the load resistors on the secondary. The 12V and 5V ones were running so hot that the wires were melting! It had 100Ω on the 12V, 15Ω on the 5V, and 50Ω on the 3.3V. I changed them to 450Ω on the 12V, 50Ω on the 5V, and 100Ω on the 3.3V. Just running it with shorting the PS_ON before and after, the idle went from 14W to 11W!

Good job! Those loading resistors can get silly-hot if the manufacturer didn't pick proper values. I don't know who came up with "100 Ohms on the 12V rail is fine" idea, as it is not. With that resistance, you have
((12V)^2)/100 Ohms = 144/100 = 1.44 Watts
That is quite substantial. A 2W resistor will get scorching-hot, and even a 3W resistor will be toasty.
And 15 Ohms on the 5V rail - that is the lowest I have ever seen (previously, this was in Macron PSUs - they usually use 25 Ohms). With 15 Ohms, that's 5*5/15 = 25/15 = 1.67 Watts. <- way too much heat to dump in a small resistor, especially near the caps.

Linear regulators aren't used on motherboards much anymore because of the large amount of heat they produced.

Actually, you will be surprised by how much they are still used.

Southbridge almost always uses a linear regulator - at least when the PC is in stand-by or "soft-off" (and that's fine, because in these modes, the SB doesn't use much power, so the linear regs for that rarely run hot).

As for the NB, this varies between motherboards. Some not-so-power-hungry NB chipsets still may use linear regs. But it is usually several "chained" linear regs, so that the heat is dissipated across more devices (something that a few early Pentium 4 / Socket 939 ASUS designs did NOT do and ended up with toasty-looking boards).
Another popular design with Pentium 4 era (and especially with i8x5 NB chipsets) was to directly take the CPU V_core and use that to feed the NB through a linear reg (if, say a 1.7V Willamate P4 was used. For 1.5V Northwood core, it was easy - just feed CPU V_core directly to chipset Vcc).
As for the newer motherboards - or at least what I have seen - is for PSU 3.3V rail (and sometimes 5V rail) to be buck regulated down to 1.8V for the RAM Vdd (if DDR2) or 1.5V (DDR3). Since most NB chipsets started using 1.5V or less way, way back in the Pentium 4 days (most nowadays use around 1V), the latest trend has been to linearly "step-down" the RAM Vdd voltage (either 1.8 or 1.5V) down to whatever the NB chipset uses.
So you have PSU 3.3/5V rails -> 1.5V or 1.8V RAM buck regulator -> NB chipset linear reg.

With RAM, I think DDR (2.5V) was the last to use linear regulators. In almost all cases it was derived from 3.3V rail to minimize the energy needed to be "burned", as kaboon noted.

Anyways, enough of me distracting from the main topic in this thread.

Also, do you think this PSU would do better in a 12V based or 5V based system? Because as far as I'm aware, 5V based systems also use more 3.3V than 12V based systems. And this is where the extra heat in the PSU would come into play. But I'm also curious in general as to which type of system would be harder on that toroid coil.

Depends on motherboard chipset. As far as I know, all motherboards still use 3.3V and/or 5V rails for their chipsets. So the motherboard with the more power-hungry chipset (bigger chipset heatsink usually) will draw more power from 3.3V/5V rail. However, what you really want to avoid with your PSU is heavier 3.3V rail usage. So in that case, specifically look for a motherboard that draws power from the 5V rail for its chipset rather than the 3.3V rail. I find this to be more common on older AMD motherboards such as socket 462, 754, and 939. However, you have to be careful with socket 462, because many use 5V for the CPU. Ideally, you want a motherboard that uses 12V for CPU and 5V for chipset. RAM on these older motherboards (DDR) will be from 3.3V rail. But even 4 RAM sticks don't draw that much power.

I also have a partially recapped Radeon 9500 pro in there. It has a dual buck converter, taking both +3.3 and +5 down to its Vcore, and three more +5 fed DC-DCs for its onboard RAM.

Ah yes, the Radeon 9500/9700 and 9800. These were the last ATI video cards to use 3.3V and 5V rails for RAM and GPU_core.
I find these video cards very useful for adding in motherboards with AGP that use 12V for the CPU. The high 3.3V/5V draw from the video card and the high 12V rail draw from the CPU makes for some very stable and well-balanced voltages - even from crappy power supplies (since all rails are loaded down equally well).

And it's not an "earth shattering" kaboom, either...

Sometimes your user name reminds me of that obnoxious "HI, BILLY MAYS HERE" commercial that ALWAYS MADE YOUR TV SPEAKERS NEARLY BLOW UP. .