low or super low ESR + impedance caps

[trodas]

Re: low or super low ESR + impedance caps

willawake - you win. I can't produce them, so it might be just as SoddemFX say - the DFI design produce too much spikes and stuff, witch - together with my fanless madness, killed the Chemi-cons easily. Now I use Panasonic FM ones and we see about them. Also read bellow - I thinking about add ceramics to filter the crap out... Witch in fact also kill then the possibility so we don't know for absolutely sure that the DFI LP B/Ultra Infinity mobos are cap-killers...


davmax - guys, I don't (also can't because I lack so many english terminology words I desperately need) want to fight. No way. I have better things to do. But please, please please next time try to read what I write, thanks!
I mean this:

1: ESR. You measured 0.1 ohm.

If you actually bother to read what I trying to say (excuse typos and stuff, pls), I was trying to tell you THIS:

...not try to tell me that the toroids on my mobo heating-up because of resistance of the thick cooper wire, witch (as I measured recently) has not measurable resistance (that it is - the quality multimeter of mine measuring 0.1 ohm for it, but the very same 0.1 ohm it says for the measuring wires only, so...) then we should get along fine

I don't know - it is really so hard to understand that when the wires used to measure with my multimeter has 0.1 ohm, then the coil has to have less that measurable resistance? Do I really need explain this to you? You never measured something that is close to your measuring equipment resolution so you know, that you have to measure also the wires you using to connect the coil to the multimeter? And the fact, that if they show the excactly same resistance, then the coil was far less, probably less that 0.01 ohms resistance?
Again - the thick cooper wire, witch has not measurable resistance - okay now?


And now - I reporting sucess!

Imagine this - yesterday I removed the heatsinks from the AGP/PCI voltage regulators - it is Arctic Silver epoxyde, however when you heat it up with big 250W soldering iron to 160°C, it get off easily. But WTF? Machine become really stable after this "modification", so I get a little theory what happen and why.
IMHO when the caps fail, the regulator(regulators) started to overheat like mad. They will for sure get burned-out, luckily the heatsing on them and it's weight pulled them out a little, when the tin get melted and therefore the mosfet lost's it's contact. That was the few sec after the explosion and before the hangup witch started all this...
Now, when I heating them up, I made them solder a little better their backs to the mobo, so that is why the mobo started to like PCI devices... So, I soldered them real good, reseted bios (long time, it was quirky again) and then even with the added capacitor 470uF at the position 10 it booter right to WinXP with FX5600 and Audigy inserted, where before it produced only 4LOD
I was wanted to test 6800LE as well (replacement card), however due to the big SB cooler and the 6800LE equiped with a Silencer it won't fit. So the final test will be done into the box, seems

Happy!

I probably better wait for the SMD ceramic caps to make the filtering better and also add the caps from back to support higher FSB where Last Viking suggested

(yes, replace the two 1000uF 10V ST RLG 352 caps with Panasonic FM ones and add two 2200uF 10V Panasonic FM ones at the bottom of the mobo - hell, another hole to the metal backplate on witch the mobo get screwed up is need, LOL - like my machine don't have enough holes already!)

[Rainbow]

Re: low or super low ESR + impedance caps

I don't know - it is really so hard to understand that when the wires used to measure with my multimeter has 0.1 ohm, then the coil has to have less that measurable resistance? Do I really need explain this to you? You never measured something that is close to your measuring equipment resolution so you know, that you have to measure also the wires you using to connect the coil to the multimeter? And the fact, that if they show the excactly same resistance, then the coil was far less, probably less that 0.01 ohms resistance?
Again - the thick cooper wire, witch has not measurable resistance - okay now?

OK, so assume it's 0.01 ohm. And now, compute the power loss at, say, 30A:
P = R*I^2 = 0.01*30^2 = 0.01*900 = 9W <- this is why it gets hot

[gg1978]

Re: low or super low ESR + impedance caps

OK, so assume it's 0.01 ohm. And now, compute the power loss at, say, 30A:
P = R*I^2 = 0.01*30^2 = 0.01*900 = 9W <- this is why it gets hot

Actually even .01 Ohms sounds a bit high. I mean, we're only talking about maybe 6 inches of mag wire here, probably 20Gauge or thicker. You would need a milliohmeter to get a proper resistance measurement on one of them.

Not to mention that some toroids have 2 or 3 wires in parallel wrapped around the torus, drops the series R by 1/2 or 1/3, and increases the current capability as well by 2X or 3X.

Also the Vcore current is split among the phases of the VRM.. You would be correct if this were a single phase VRM, but typical "performance oriented" boards have at least 3 phase. My Gigabyte board has 6 phase VRM, but each phase is "smaller" than the typical phases found in a 2 or 3 phase design. Hence why most boards with >20A Icore capability have more than one phase.


Then again if we're talking about a P4 Prescott, with a TDP of ~125W, with Vcore being 1.3V, thats almost 100A..

[davmax]

Re: low or super low ESR + impedance caps

Yes Trodas I did say the 0.1 ohm is very high. I gave you the calculation so that you can work it out for yourself. For 0.01or 0.005ohms for 15, 30 or 50amps. I can only give examples and explanations to assist you. I hoped that the article covering "DC chokes" would help you.

As the information reveals the mostly likely cause of heat is ESR added to surrounding heat generating sources. Core losses relate only to ripple which is a much smaller amplitude than the DC current.

[PeteS in CA]

Re: low or super low ESR + impedance caps

Not to mention that some toroids have 2 or 3 wires in parallel wrapped around the torus, drops the series R by 1/2 or 1/3, and increases the current capability as well by 2X or 3X.

That is likely true with really high current VRMs. One reason is the difficulty of trying to wind AWG #10 or 12 on a 1" or 1.5" OD toroid. Another is the problem of "skin effect" with high frequencies. OTOH, the O/P inductor in a P/S will often be a "coupled inductor", with windings for several O/Ps. If the number of turns for each winding is designed properly, it improves the regulation of the +/-12V and -5V O/Ps.

[davmax]

Re: low or super low ESR + impedance caps

PeteS.
In essence there are many variations between systems and even CPU loads. eg two and three phase VRM and differing inductor design and possibly core materials (even faulty caps). So each case needs to be addressed. Now in Trodas case he states the inductors are very hot, the source of heat is probably ESR loss added to heat from other components.

I would use a IR thermometer (non contact) to establish hot spots. Not everybody has an IR.
This really helps isolate any problem area.

We can guess all day about the value of ESR (it can be calculated on a case by case basis) It appeared that Trodas originally may not have appreciated that ESR watts is proportional to the square of the current (heavy currents at that).

PeteS what is you opinion as to the source of heat?

We do seem to have drifted off the Low ESR caps topic.

[trodas]

Re: low or super low ESR + impedance caps

Yea, we are far, far away from the main topic. And yet there come some very good points, like that there is 3 coils and not one - witch mean the current to the CPU is divived by 3 for each of the coils.
The second problem is to determine the resistance of very thick cooper wire - and we talking about less that 3 inches of it!
The wire has diamater 1.35mm. Anyone willing to do the math is more that welcome
To move to the real world, let's assume I try and manage stable run fanless at 1.850Vcore and 2700Mhz. This give according to the CPU heat typicaly 86W of heat and maximaly 109W. ( http://www.cpuheat.wz.cz )


And now a little update!

In recent minutes I get my hand on another dead (destroyed Vdd mosfet by puling the heasing - broken in half) DFI LP B mobo - Taiwan made one. But the point why I write this is - over the board, it is full of Teapo caps!!! Now this is quite a bad thing, since they are known as BAD CAPS BRAND!

The Hall of Shame - Badcaps Photo Montage
Teapo - https://www.badcaps.net/forum/showthread.php?t=424

List of Bad Cap Manufacturers
https://www.badcaps.net/forum/showthread.php?t=388
Teapo - http://www.teapo.com.tw/

So, no wonder the max OC is getting lower and in time - that is ever worser that the ST caps I have on both my china-made ones...!
It is likely the voltage regulation/something is better on them or somethings, as the OC is claimed by SAE that are better on Taiwan made ones, IIRC Vdrop did not react on China ones, but does on Taiwan ones.

But maybe it is time to check the specs of the caps
Caps could do huge difference - I mean - maybe the Teapos are better in the ESR/ESL aspect that the ST ones my China mobo have, but just have shorter life?
After all, I have now in my possesion 2x China and 1x Taiwan mobos and I'm positive I could made them all work (if I get the need parts), not only the one I actually resuscitated, so some testing coudl be done...
But I definitively replace the Teapos all, because the previous owner having hard time at 200Mhz FSB... witch is not a overclocking at all...


PS. davemax - yes, I have IR thermometer. What readings you want? :
These: http://ax2.old-cans.com/s.php?p=wc&id=116&c=8&d=1&v=v2
Yes, the clock is very low and there is the Zalman airflow... But reduced to 1100 RPM, so...

[davmax]

Re: low or super low ESR + impedance caps

Hi Trodas. That is good you have IR. Does it have narrow focus so that heat spots can be seen? Some are very good. Mine is a low cost one with a 10:1 focus ratio eg at 10cm the beam is 10 mm diameter so a close-in distance og 2cm gives approx 2mm.

Now I do not really neeed readings. Hopefully you can read the temp on the motherboard where the inductor is connected. Then measure the temp of the inductor core. The difference between these readings with give you the temp rise of the inductor. If you have a high resolution IR you may be able to detect if the copper on the inductor is hotter than the core.

Those readings would be interesting.
Is there any difference in the inductors? Are they all the same temp?

Because of skin effect a copper resistance calc with not be 100% but near enough to get an order of magnitude i.e approx. I will have a crack at it later, time is short right now.

Yes, I apologise that I did not make myself clear about the ESR calcs for the inductor. I did read that the 0.1 ohm was mostly test lead. I should have calculated at 0.01ohms (2.25W) a very simple side ways shift. I looked at the CPU heat link you supplied. How did to obtain the 109W max? Was it for 1.85V and 2700mhz? Both increasing the wattage. Taking 110W and 1.85V this would mean close to 20amps per inductor or 4 Watts in 0.01 ohms. 4 watts on a device that size can be hot, but a lot of the heat will be piped directly through the copper wire into the motherboard. But we are still guessing. I think the IR is a good tool.

[davmax]

Re: low or super low ESR + impedance caps

Well the guessing is over for ESR of the Trodas VRM inductors. If the copper wire is indeed only 3 inches long and 1.3mm diameter (16AWG) the resistance is 0.001ohm with a dissipation of 0.4W at 20 amps. This would confirm the Trodas statement that the inductor high heat is not copper loss, even if skin effect and temperature doubled the resistance it is unlikely that this is the cause of high heat. Similarly it would appear that the small excursions of ripple current would not generate high heat either, this would be very hard to establish not knowing the ripple characteristics.

3 inches represents 4 turns at 0.24 inches diameter. Is that right Trodas?

I hope it is now understood that no heat will be generated in the inductance (not inductor) by the ripple current (it being lossless, storing energy as current increases and releasing the same energy as current decreases).

This would suggest as previously proposed that the heat could be conducted from the motherboard plane by the copper wire.

Any other thoughts anybody.

[trodas]

Re: low or super low ESR + impedance caps

Ah, well, here we go. First at all - I already posted the area in question, davmax. If you followed the link there:
http://ax2.old-cans.com/s.php?p=wc&id=116&c=8&d=1&v=v2
...you will see it + temp readings. For temperature measuring I use this toy:



Witch is basicaly opticaly-striped down version of better one, but for the price... well... I picked this one. Specs there: http://www.deltatrak.com/infrared_thermo_8.shtml
8:1 aren't very good, but the 33:1 version is very expensive for home-use, so...
The measuring is quite precise (the electornic are same as in the hi-end version) however the targeting is harder

I dubt I could measure precisely at the base of the coil, but I could try it. Besides, the wires will be probably the same (cooper is known for excelently low heat transmition resistance) or lower (no, resistance is not the heat cause there) temp as the core of the coil, witch does heat-up there, thanks to the power transformed trough it when the turn-on phase of mosfet end and the LC stuff create the oscilation around the desired voltage.

Picture showing the RMS ripple on DFI board, before the owner modified the circuit a little to produce better looking results are there:



Very bad, right? Well, that could be what damaged the caps anyway...


I obtain the 110W by the CPU heat calc program. You pick your CPU type (Barton), the CPU voltage and the CPU clock and there you go. 85W is Iddle, 110W is 100% load (priming AND doing something else might come close to this...)

3 inches represents 4 turns at 0.24 inches diameter. Is that right Trodas?

Again, the pic! You can see, that there are 6 turns on each of the coils. The coils aren't round, but more likely square with cut coners. The width is 6,5mm and the height of the material around it the wire go is 3mm. Now do your math - but I quess that we need to add about 15-10% for the fact that the wire (as you can see) don't copy the coil precisely, but because of it's thickness crete more round turns.

Knowing the ripple characteristic is possible from just looking at the oscilloscope screen pic SoddenFX give me

PS. I still don't understand how you can say that transfering the power (inductance?) are loseless, when the coil obviously heating-up, but I did not want to cause another fight I rather would like to complain that 30 pieces 10uF ceramic Panasonic SMD caps cost 32$ Now that suxx (8 caps like this missing on single DFI LP B mobo)

[willawake]

Re: low or super low ESR + impedance caps

this is what you need trodas http://www.flirthermography.com/

[trodas]

Re: low or super low ESR + impedance caps

Would you be my sponsor, willawake?

I will be loyal to you and place your logo ony my T-Shirt :

[willawake]

Re: low or super low ESR + impedance caps

hehe my friend sells them but he wont give me one to play with

[davmax]

Re: low or super low ESR + impedance caps

Thanks Trodas for the ripple waveform. But I an unable to tell how bad it is. You need to quote the oscilloscope Y sensitivity setting in mV/cm.

Your IR looks good enough at a 20mm distance the spot size is about 2.5mm.

Thanks for the coil dimensions. This will help but it will not change much, I had a feeling 3 inches was short, the real length works out to 125.4mm or 4.94 inches this means a ESR of 0.0017. If we again double that to allow for temperature and skin effect (0.0034) this would mean at max 110W and 20A per inductor a wattage of 1.36. So nothing too bad.

The inductor losses are: ESR at 1.36W plus core losses that should also be low (the DC magnetic field has no effect and the ripple magnet field is small, but the high frequencies involved increase the losses at this low level, this has to be defined by your scope setting). Finally the remaining inductance contributes no heat (inductance resists current change thus acting as a current smoothing device. It resists a current rise and stores energy and it replaces that energy as current decreases thus help to minimise the current drop).

I hope that you will understand this inductance bit. It is correct. If the inductor gets hot you simply find out where the heat is coming from, difficult I agree, but the IR should help. The calculations with normal operation indicate that the inductor heat is not great, so there may be abnormal operation somewhere.

For comparison on my Epox 8RDA+ the three inductor temps are:
Mb Inductor
58 54
58 54
52 49

So here the inductors are cooler indicating that they are not adding heat but actually radiating or losing heat for the motherboard.

[davmax]

Re: low or super low ESR + impedance caps

Further note. My system is fan cooled. If you are still running fanless it is most likely that the motherboard heat is the problem. Just look at my temps. I think PeteS warned against fanless, it could kill inductor cores, lytics and FETS. Surely you have some fans.

[linuxguru]

Re: low or super low ESR + impedance caps

Trodas, you need to ground your oscilloscope probe very close to the capacitors where you're measuring the waveform. The overall shape of the waveform is typical and expected. The spikes could be:

1) ground bounce due to overlapped conduction of the VRM MOSFET totem-pole when switching. It's undesirable, and can probably be minimized by using FETs with smaller gate capacitance, but it's not a serious problem.

2) caused by lead and foil inductance between the VRM inductor and the output capacitor bank. You can get rid of it by putting suitably large X7R SMD ceramic(s) close to the VRM inductor, between
Vcore and GND. Something between 100 nF and 10 uF should do the trick - try experimenting.

[PeteS in CA]

Re: low or super low ESR + impedance caps

The overall shape of the waveform is typical and expected.

You can see 3 elements in that w/f:

* The ramp - positive-going while the cap is being charged and negative-going while it is supplying current to the load (discharging);

* Steps or discontinuities at the changes from positive-going to negative-going and rom negative-going to positive-going - this is due to the ESR of the cap;

* Spikes - these are due to inductances in the circuit, e.g. the ESL of the cap and stray inductances mentioned in lg's post, and due to noise radiated into your probe if you haven't grounded it carefully.

Avoiding noise pick-up depends very much on the layout of the circuit being monitored and your specific 'scope probe. You can get a reasonable idea of how much radiated noise you are observing by connecting the probe ground in the circuit and then probing that point. If you see no spikes, that's good; if you see significant amplitude spikes, you need to try to improve the grounding. If the grounding is as good as is practical for the MB and your equipment, just remember the limitations of your measurement set-up.

[trodas]

Re: low or super low ESR + impedance caps

My final idea (after discovering how bad the other that Chemi-con caps on the DFI LP B mobos are) of recapping the mobo looking like this:



As you can see, pretty much all bigger caps get replaced.
Caps 1 - 5 was 3300uF 6.3V Chemi-con KZG ones, but now there are better the Panasonic FM ones with 4700uF 6.3V.
Caps 6 - 8 was 1500uF 16V KZG ones as well, replaced by 2200uF 16V Panasonic FM ones.
Capacitor 9 was 1000uF 10V OST crap, now there are 2200uF 16V Panasonic FM one, as Last Viking recommended high capacity there.
Empty space 10 recieved 470uF cap.
Empty space 11 seems blockin when fited with 470uF one the start of ram-controlles, so ATM remaining empty, I might try it once again to be sure...
Empty spaces 12 - 24 recieved 120uF 16V Panasonic FM caps.

And todays I started dumping the OST crap out of my mobo, four pieces of 470uF 16V (25 - 28) get replaced for the Panasonic ones. When you look at the picture, it seems that they are into sensitive areas as well, so it might help. Two are right up to the cap 1 and cap 6 - I quessing powering of the onboard NIC and USB, witch is important. The third one is just on the right top AGP slot corner, close to the coil. This could be important cap as well, OST has no place there - filtering some of the AGP voltages proably? And fourth one is at the bottom of the mobo, right next to the yellow USB ports and next to the hole. Filtering USB voltage might not be the most important thing, however the crap has to get off from my mobo and I had some USB problems right before the "big bang", so it has to get off...

Now is pissing me only the four 220uF 10V craps (close and between the ram slots - 40 to 43), next six 1000uF 6.3V around the ram slots and stuff (29 - 34) and the ram's powering by 2x 1000uF 10V (36 and 37) + 470uF 6.3V - 35. This one is going to get replaced by 560uF one. And yes, exactly these two, where Last Viking recommend to add two 2200uF caps from the bottom side to these ones, as you can see on his pic.
And so I will add these! :

Now there (as well, as on the picture) is missing only eight empty space pieces for 10uF 16V polymer caps, where I will add 10uF 16V ceramic ones (ceramic is better in specs even that polymer, I suppose, right?) and another eight pieces of 5.6uF ones. Half of them are missing in and half bellow the CPU socket, where I will add hi-quality 4.7uF 10V Taiyo Yuden caps.
(sadly they did not have 5.6uF or 4.7uF for less that 10V - 6.3 or even 4V are much better choice there, as the Vcore are nowhere near 10V...)

And oh well, I almost forget that caps 1 - 10 and 36 + 37 will be bypassed by 100nF 16V filtering SMD caps from the bottom side of the mobo for better filtration and getting rid of as much noise, as possible :

Now only get hold on the caps, finally!

And at least the mobo working entierly stable with all the added caps now. 200x10.5 only, CPU limited (oldie Barton), but at least with tight timings on ram: 11-2-2-2-7-12-0-0-3-3-2-0-3-E-E-F-E

More to come. Soon, hopefully.

[trodas]

Re: low or super low ESR + impedance caps

Preliminary recapping results

I spendt last two days soldering new caps on the DFI LP B mobo and even there is still 6 the old bad caps (I ordered wrong size, forget to check, my fault), the results are so far not amazing, but stunning!

I have a oldie Barton, unlocked, 2500+. I killed my Epox 8RDA+ mobo with desperate attempts to run it at 2200Mhz (3200+ 200x11), because it need 2.1Vcore (!!!) to run stable there. Todays I made slight mistake in bios and to my shock, the mobo posted fine with PR3200+ rating settings at only 1.850Vcore witch I used to reach 2100Mhz...!

With little tweaking I discovered that 1.900Vcore is Prime95 stable at 2200Mhz, while before I need 2.1V on the Epox 8RDA+.

Now that IS what I would call improvement!

I just wonder how much more Last Viking could gain for FSB or how much more MHz SAE can get if theirs mobos was modded the way I do it - adding not only hi-quality caps, but bridging them with 100uF ceramic SMD filtering caps where possible and then also adding hi-quality ceramics 4.7uF ones into (4x) and bellow (4x) the CPU socket to futher filter the voltage.
I added the filtering everywhere I can and the results are, well, outstanding. Im so happy with them

Brief history of what I did:



CAP 18 canno't be installed of mobo did not POST!
CAP17 I made the 220uF 10V one isntead of the 120uF 16V one.
CAPs 36 and 37 are bridged from back of the mobo with 2200uF 10V caps and also at the same time bridged with 100nF filtering SMD caps.
CAPs 7 and 8 canno't be bridged with 100nF filtering caps or the mobo did not POST!
CAPs 1 to 6, 9 and 10 are bridged with 100uF SMD caps from bottom of the mobo for better filtering as well

Then there are 9 empty spaces where SMD 10uF electrolyt caps are soldered elsewhere on the mobo. I used 10uF SMD ceramic ones, however there are 3 positions witch cause NO-POST scenario. One of these are capable of serious bios corruption, as I experienced when I soldered the cap there Beware!

Again in graphic - do NOT put cap in there:



OR there:



...and that it is! Now the area around the Vdimm + Vdd mosfets are pretty crowded with caps and once I get hold of the 1200uF 6.3V caps, I try (with reasonable cooling and usable CPU) how far this beast will clock

Obsession with large bulk caps? Hell yea!

[Per Hansson]

Re: low or super low ESR + impedance caps

Big congrats trodas on your excellent work!

I'm a bit curious though, could you do a test with the system as it is now and see how hot the capacitors really become? Because UCC KZG caps should not fail...