low or super low ESR + impedance caps

[trodas]

Re: low or super low ESR + impedance caps

Okay, thx.

Now promised images:


Yep, these two are faulty...


I include this foto for fun - look, how sharp it is Never mind the blurry caps, the PCI slots behind them are "razor-sharp", aren't they?


Focus on the first one - got beated up pretty bad, fella...


Focus on second one - you can almost see the black stuff inside into the crack - please note that both was bumped much more, when I pulled them from the mobo... The bump on their top must shrinked a little.


From side - like I said - the bump was higher when it happend - now are only mediocre...

Another observation - I looked on one of older photo I made, this one:
http://ax2.old-cans.com/s.php?p=wc&id=114&c=8&d=1&v=v2
...and I definitively notice that the cap is already bumped a little...! Yes, this one in the middle of the picture, near the additional 12V connection to the mobo + visually close to the hose going into the chipset block. Full size there:
http://ax2.old-cans.com/wc/0115_full.jpg
(on the mosfets was cooler, sadly not glued to one of them, that is why they are with silver top - Arctic Silver epoxy used. And the black crap in the coils? A hi-temp silicone meant to quiet them. Unsucesfully. I should rather install decent caps, but how the hell I was supposed to know?!)

[trodas]

Re: low or super low ESR + impedance caps

davmax - thanks As they are superb, the best caps ever, so... I don't have Paypal, but friend has, so maybe he can help me, we see. The delay because of the New Year is not great, so let's look out for alternatives still.
And yep, I know the price is pretty low.
And as for the capacitance, it could be even a little bit bigger, it did not matter much, I think. Definitively 1500uF is not enought, so 2200 or 2500 or 2800uF won't hurt, do you agree?
OTOH when it come to that, I want buy a few spare ones as well. There is at least two positions for 470uF missing ones and replacements for future is also good to have, not to mention about 4700uF / 6.3V ones for the CPU power section.
The Samxons GA could definitvely do better job that Chemi-con 3300uF / 6.3V ones what are there now, don't you agree?
I dop him a email, and let's continue searching for alternatives. The Nichicon HZ ones looking excelent as well...

[trodas]

Re: low or super low ESR + impedance caps

Yes, the Farnell shop I was reffering about into one of my previous posts, that they offer mediocre Nichicon caps (worser that already exploded on me) and that they also refused to send me anything to my country or even to Germany (refer to the Germany trick part ).
I looked again and they stock only UPS1C222MHD NICHICON cap, witch has laughable useless 1550mA ripple current

The problem with http://uk.farnell.com/ is, that even for Germany it says:
"You're not allowed to export to that country. Please select a different shipping address."

...but yep, they have the Rubycons ZL ones with fairly great "punch" at 2770mA and only slightly higger impedance, tough once again:
"You're not allowed to export to that country. Please select a different shipping address."

So, unless someone from England is willing to re-send me these, I have no idea how to get them
Pretty sad, don't you think so?

I try Froggle again... And someone EXCEPT Big Pope selling them? Aren't there a list of distributors on their webpage or something? There has to be more ways to get these.

Samxon GC is fine and available when Big Pope returns.

Well, GC aren't as great - there are far better ones. What about GA ones?

[willawake]

Re: low or super low ESR + impedance caps

you can get Sanyo MV-WX at Elfa 2200uf 16v 12.5x25
http://www.seapraha.cz/

[trodas]

Re: low or super low ESR + impedance caps

willawake - thx, I jsut noticed it as well They aren't from the best ones, tough they are acceptable at least for testing stage : Just to know, if the mobo still can kick-in. I writen them already, if they could get any of the "choosen ones", so we see. In case of emergency and unreasonably long time to get better caps, they might do the job. Please note that they have higher ripple current, yet also higher impedance!


I would like also elaborate now a bit by bit, how the cap problems are come to the final loud thud from my computer, witch after a 15sec or so ended up with complete lockup.

Even from the very beginning, I was not able to break 200Mhz "barrier" (not even 202Mhz) w/o sound distortions. First these distortions happen only on key press. Then all the time.
Backing down to 200Mhz make them disappear completely.
I have no solution and none of all the overclockers as well.
Yet still, play AC3 hi-bitrate movie, pause and un pause it resulted sometimes from strange loud noise from all 5.1 channels. It happened mostly with AC3Filter, since it use 24bit output, normalizations, amplifications and so on - and therefore stress the machine (PCI bus and so on) more. Later it move from the occasional to 100% sure noise from all speakers when I pause/un pause AC3 movie. I blamed AC3filter. Yet no-one else seems to experience such bug, so...
Recently I realized that my max. OC is severally limited. My max. post was 2813Mhz and CPU-Z record was 2700Mhz - yet when I try it now, I get complete lockup! Also last year I run with 200x13, and now it is not stable and after few hours a strange lockup come. (no blue screen, no reset, lockup) I blamed 6800GT card, witch for sure draw more power and therefore might limit my max. OC, but as we know now, it was also false accusation.
Then sometimes, depends on BF2 initialisation, phase of moon of whatever else, BF2 game play was interrupted with noise. Usually when more sounds has to appear on the game. I blamed drivers, then changed to the better U-pack ones, yet no help. I blamed game, but reboot fixed it. Always. I was clueless and think about adding a additional grounding wire from the mosfet on Audigy2zs to the mobo (as I once did with suggestion from the overclockers to get past 200Mhz w/o bugs) and/or capacitor to better filter/hold the voltage when spikes of energy consumption occur.Another idea was to slap a chipset heatsink on the Audigy2zs chip. Tough all these ideas are probably worth implementing, I'm now almost convinced that they would not help, at least except the cap not much...
But back to the problems.
Then USB stopped working and any attempt to install it end in lockup. Also, some of the services are a bit ficked up, since I cannot even change status of the service (eg. how it run, Automatic, Manual of Forbidden ) - the process always lock up. I thought that it was probably something screwed up in winblows and reinstall will fix it. Hell, if things are THAT simple to fix...
And then finally a week and half before final showdown the machine used to lock during booting. Sometimes. Later most of the time. No increasing voltage helps, nothing. Right when the GFX card driver deg initialized it simply freeze. Reboot fixed it, always and I blamed winblows. It turns out that winblows are innocent at this time.

So, this is just my short story how bad caps happened to change my machine behaviour.

[yanz]

Re: low or super low ESR + impedance caps

I see, but you put the standard to KZG so it would be difficult to find another caps that have the spec better than or the same with it. I know you want everything to be exactly as what you need, i was once like that..

What i see about KZG is that it has the spec that "super" compared to another caps, while has the size 2/3 of them. But the endurance is half of them.

You sure what you did with your case system is safe? I too am a freak for silent system, but its not mean i would remove all the fans. That's not healty. I prefer the case to be like this (ducted PSU, large fans on exhaust and inhaust)


This is my opinion from my experience but could be wrong: anything that has >=1500uF >=16v >=10mm(diameter) would be enough for your need. Also the ESR that <= 60 mili ohm would be just fine.


Dont worry too much about too much capacitance. I replace the 2200 uF with the 3300uF and no problems (i have a lot stock of panasonic FC and chemicon LXZ/LXY). But it's just my experience..









For 1500uF 16V application, I have rubycon ZL in my stock, but it's 12,5mm in diameter. Another one is Panasonic FJ, 10mm, this cap has no data i can find yet, but i guess it's at ~1600uF, lower than what you want, but again you put the standard too high with KZG (imo).

It take a long time for me to get a skill of putting a 12.5mm caps in place of 10mm caps. Sure it better to get 10mm caps in first place, but its hard to find. Damn distributor, they just take anything available from Singapore and don't care with the brand/series (as long as its low esr). Buying from internet would be ridiculous because its high price. Ironically, I could find a plenty of low esr teapo here...

As for the soundcard, make sure you get the case grounded properly and your PSU has active PFC to filter the noise from AC circuit. At least that's what i can think.

[davmax]

Re: low or super low ESR + impedance caps

Trodas I am concerned that you are taking some risks with your system. You need to check the system carefully stage by stage.

Caps are killed by excess ripple, over voltage or excess temperature.
The cause must be found , the system apparently worked well. It seems that failure may have occurred due to your modification ie temperature rise, or that the power supply is faulty (excess ripple or voltage).
I have read all the posts and there are some missing facts.
1. What voltage is applied to these caps? 12V or 5V? The 16v rating would imply 12V but manufacturers can and do fit 16V caps on 5V supplies.
2. What make of caps are in the power supply? Do not assume that all is OK. The voltage may be correct at light loads but at working loads the output ripple may be excessive with poor caps. Check the brand of caps and if necessary change all over 1000uF with a reliable brand (there are not too many of those).

Now you rejected the Samxon GC cap because GA is better, but GC 1500uF 16V ratings (2900mA and 10mohm) are better than KZG , MBZ, MCZ and much greater than FC (1800mA).The cause needs to be fixed, getting the ultimate caps is not the way to go.

Many of us still suspect the PS, it feeds the power to these caps and apart from excess heat this would be the stress mechanism. So check it out thoroughly.

Now if the voltage to the caps is indeed 5V you can use 10V caps. Maybe easier to get.

Fix the cause.

Drilling into backplanes is disastrous for motherboards. Unless you are a mother board design expert it is not wise to modify, you will create problems.

Go carefully.

[trodas]

Re: low or super low ESR + impedance caps

yanz - you got a point, mate. Yes, the KZG Chemi-cons are pretty good (reasonable ripple current 2550mA and super low ESL - only 0.013) however either at least one of them are bad and dragged the other down with it, or I simply need better caps there. Remember that I had issues over 200MHz FSB (witch is quite normal on DFI LP B mobos to run) and since I had enought of them, then I logically seek a solution and not just a workaround.
Therefore - like I mention earlier - I seek to replace them with a slightly higger capacity ones (I feel that 2200uF over 1500uF could be sufficient, maybe 2500 or 2800uF could be as well, but it depends on dimentions) AND I also plan on adding caps to the places originaly on the mobo intended to have them, but removed in order to cut cost down. Maybe you heard it too, but caps are normally pulled out from the reference mobo, untill it stop working. Then the last one is put back in and mobo is now "cost-optimized"
There are one bigger (470-680uF?) missing right next to the two failed, 4 small are missing near the main PSU connection, 2 into the memory power part and 6 (!) smaller into the PCI slots. I do believe that puting a quality caps back in could:
a) dramatically increase the lifespan on current caps
b) let me run over 200Mhz w/o sound issues that obviously come from inadequate power/signals filtering on the PCI slots.
c( myebe even reduce the heat, because adding other caps make distribute the stress among them, so...

What is wrong with seeking for quality components? Look at my list:
Samxon GA
Nichicon HZ
Rubycon MCZ
Chemi-con KZJ
Samxon GC
Nichicon HN
Panasonic FM
It does NOT containing only Samxon GA, Im ready to go as low, as Panasonic FM But not lower.

The endurance (life span) is interesting thing. My mobo worked over 1 and half under 100% load (folding all the time) and the CPU caps are okay, even they should not be. Let's do the math. The machine run every day for at least 15 hours, sometimes 24. Therefore this is (15 x 365) 5475h per year. And they still working well!
It IMHO indicated (plus the fact that there aren't single other DFI overclocker who did not beat 250Mhz FSB) that the cap in question WAS faulty from the very beginning. It just get worser and worser over time. And when I add the caps with higher capacity, better specs (lower ESL also means lover heat produced) and redistribute the load to more of them, I should be fine again even with 2000h caps.
BTW, all caps from my list, except the 7000h Panasonic FM ones are 2000h rated.

I disagree that I could allow the ESL as high, as 0.060. I believe the 0.013 are there for purpose and the replacement should not get higher at all, especially when I planing to stress the mobo harder that before. And also there is many caps with far lover ESL - Nichicon HZ with 0.0006.5 is fine shining example. Damn, I want it bad!

Big thumbs down for distributor that just take anything labeled low ESR and did not care about type, specs and stuff. The caps obviously difer a lot, so... such short minded way is shamefull.
Buying from the internet seems ATM for me as the only one way to get good quality caps. Yet still ATM I managed to find only mediocre ones, like Nichicon HE, online. I did not even counting other ones like useless Nichicon VZ series...
(both mentioned are at stock in http://www.digikey.com /me tinking about the HE just for testing/other non-critical purposes usage, as they still have more that twice the impedance and still less the ripple current that KZG...)
So the question is not the price, but WHERE to get the quality ones.

About the soundcard - I don't dubt the case-grounding connection done into the Neo ANtec PSU and it have the active PFC as well - you can check it in my fongic link in my sig I believe the bad caps are the issue and not anything else. But to test is throughly I need quality caps. Something OVER the Chemi-tech KZG specs
And yep, Im perfectionist and I love my fanless watercooling and there is no way in hell I could install one sigle fan in my machine - hell I love how it it quiet too much! I rather limit my OC


davmax - well, thanks for your concern, but no risk - no fun!
Yes, you are right - I should check my system stage by stage, yet we still did not even get nowhere near the stage 1 - power up the mobo again with quality caps. You are right about the reasons, yet forget to include possibility that the cap was somehow faulty from it's beginning, witch explain a lot. I use, from the start, also other PSU - a old 431W Enermax witch hardly cope on the 12V power line and therefore might damage the cap even more? Dunno. For past over year I use Antec NEO 480W PSU and it rules.
1) - about the actuall cap voltage - I don't know.How should I? They are very close to the additional 12V PSU connector, so I assume 12V, but let me check on the mobo, if the 12V line is somewhat connected to the cap or not.... Confirmed. As I suspected, trough the coil the 12V is directly connected to the 12V rail/connector from the PSU. BTW, the missing cap is connected before the coil and obviously should help filtring anything that could get inducted into the rails as well, as stabilize the voltage under load. I'm bound to add at least 680uF there.
2) - caps (okay, their tops only) looked all fine into the PSU, I did not checked them more throughly. OTOH checking visually did not mean much, except to find obvious leaking at bottom and bumped top... The cap should still look perfectly well outside and has only half of it's capacity inside, so... I'm affraid I should buy any oscilloscope with show or not show me possible output ripple under maximum load. (ed. Doom3, Quake4, 3DMark05, BF2 or something like this....) I don't have one ATM, tough. Not even know a friend ho has, so, this is a little bit of problem...

And I did not say that I rejected Samxon GC, as you can check - they are still into my list, just not at the top:
Samxon GA
Nichicon HZ
Rubycon MCZ
Chemi-con KZJ
Samxon GC
Nichicon HN
Panasonic FM
These I want, other I reject. And yes, GC ratings is sufficient to say at least - certainly a lot better that KZG ones and please also note that I want increase the capacity a little, so if you look at the 2200uF ones in the GC you see impedance 0.009 - ripple current 3190 witch is excelent Fothermore, there is the bigger diameter ones, what can do even better... But if there is a chance to get GA, I go for them for obvious reasons. They can lose almost half of their specs and are still above KZG and certainly above failing KZG ones!
And yes, my mobo has to be fixed, because even typing on THIS:
http://ax2.old-cans.com/s.php?p=badt...0&c=8&d=1&v=v2
suxx like hell But it did not mean that I go for the low-end. What is the problem with aiming high? After all, there is other 6 types of cap's I find more that suitable to go. (the key was, obviously, rated ripple current over 3000 and lower impedance that KZG line have and then sorted from best to worser yet still perfectly suitable alternatives)

By drilling to backplane (plate?) I did not mean the mobo - God forbid!
(I once cut out piece of a "obviously unused corner" on my GF2Ti card and from the time the card know only 32MB of it's 64MB capacity...)
I was aiming to improve the cooling. (eg. give the caps longer life)
The story goes like this - I was desperate to get over the cursed 200Mhz limit, so I was (and I'm still are) ready to do almost everything. So I created this cooler and glued it to the mosfets with Arctic Silver temperature epoxyde:
http://ax2.old-cans.com/s.php?p=wc&id=123&c=8&d=1&v=v2
before:
http://ax2.old-cans.com/s.php?p=wc&v=v2&d=1&id=122&c=8
Unfortunately it means that one of the screws are removed... and that is where the irony begin. So, since there is the Vmosfets voltage regulator chip (the square one bellow), I had to crate 0,7mm thick "extension" to the heatsink in order to get it cooling everything. Now the problem is, that I failed to realize that into fanless case the heatsink did not help much, not to mention the mosfets emiting most of the heat into the mobo PCB (they are designed that way). What irony? Now it come. Since I removed the screw between these hot components, I killed the heat-link between the PCB (where most of the heat go) and the big metal backplane, on witch the mobo is screwed up!
The big thing provide undubtelly FAR better cooling that the heatsink - especially when there is other heating-up components right away from it - SB bellow, hdd near, and GFX card extremly close...
So, in short, this "ingenious" heatsink idea was faulty and now I took it out that give special care to the screw - I use brass screw with brass washer too and I also added brass washers from both sides of the screwing point, and also added a nut on the back side to make the heat transmit better to the backplane, witch, by - just by it's size - provide FAR better cooling.
And my modifications did not stop there.

The text that you have entered is too long (10871 characters). Please shorten it to 10000 characters long.

Oh, my...

[trodas]

Re: low or super low ESR + impedance caps

Bellow the CPU is already a 92mm hole into the backplane, used on Epox 8RDA+ to add helping caps there:
http://ax2.old-cans.com/s.php?p=wc&id=80&c=8&d=1&v=v2
And I added another 92mm one (partly overlaping) so I could add a piece of such heatsink I added to the mosfets on the back side of the PCB, right bellow the CPU mosfets. At 1.725Vcore this is not necessary and I run at 1.800 last year and it worked as well, but still - it might be better. I planing to stick it by thermal both-sided sticky - like the one added to the ram cooling stuff. I hope the sticky holt with the temperature well...
There are few minor other modifications to the case, like 2x 102 and 1x92mm holes in bottom + the case it on anti-vibration about 1 1/2 high silentblocks:
http://ax2.old-cans.com/s.php?p=wc&id=77&c=8&d=1&v=v2
...so there is a nice intake
Also in top middle you can find 92mm hole:
http://ax2.old-cans.com/s.php?p=wc&id=67&c=8&d=1&v=v2
...and the 80mm holes in back/front are driled trough to make 0 resistance to airflow:
http://ax2.old-cans.com/s.php?p=wc&id=79&c=8&d=1&v=v2
(no pic ATM of the front one)

And the drilling? Well, no picture yet, but basically I created a two sets of 6x36 7mm holes into the case side, one up like there on my old case is something like it:
http://ax2.old-cans.com/s.php?p=wc&id=37&c=8&d=1&v=v2
...and one down. Purpose? Air circulation into the area, where previously was no air circulation. Planed benefits? Enhanced cooling of the backplane with will now cool the Vdimm and perhaps the Vcpu mosfets as well.
Hope you see whare I'm going it with

[trodas]

Re: low or super low ESR + impedance caps

Oh, well, I made a foto of the drilling work.
http://ax2.old-cans.com/s.php?p=wc&id=115&c=8&d=1&v=v2
It was not easy, as you can probably imagine. First draw where the holes will be, then punch them, so I can drill them all with 3mm drill and then finish them with 7mm one. And then clean their sharp corners and stuff with 14,5mm drill (both sides).
The math go as follows: Two times 6x36 = 432 holes. But the number of drilling operations is 4x higher - 1728. Ugh. I never do it again, my arm hurt

[willawake]

Re: low or super low ESR + impedance caps

fucking awesome

[davmax]

Re: low or super low ESR + impedance caps

Trodas. I am not sure of your understanding of ESL v ESR. ESR represents equivalent series resistance and therefore a source for heat generation. ESL includes inductance L that can reduce the ripple through the capacitor and increase it in the CPU/load. The inductance L is lossless and therefore does not add heat.

Now low ESR is most import for the VRM output (this is where you are not having problems). The effective load resistance of a CPU with 30amps at 1.66V is 55milliohms and the output cap ESR must be considerably lower than this. Example on 8RDA+ there are six 3300u caps with an ESR of 9 milliohms each, in parallel this equals 1.5 milliohms or 2.7% of the load value meaning that the ripple through load is 36 times less than the caps.

Take the VRM input caps (the ones that are giving problems) with a 12V supply and for simplicity considering 100% efficiency the load resistance reflected at this point is 52 times larger (2.860 ohms). Cap ESR suddenly becomes very much less significant (can be effectively 52 times higher). So ripple is a consideration. Remember that the ripple frequency may be around 100Khz and 1000uF has an impedance Xc of 1.6 milliohm so four in parallel represents 0.4 milliohms, so low compared with 2.86 ohms this is insignificant and making ESR the dominant value for ripple. Therefore in terms of ripple any increase in cap value above 1000uF is not very significant. The main purpose of bulk capacitance is to support transient load changes and these are short term and do not create significant heat.

In summary aiming to get caps with higher uF value and low ESR for application to the input filter gains nothing significant in terms of heat dissipation. So for all things being equal your problem points to external heat, faulty cap or high ripple from the PS.

Hoping this is helpful

[davmax]

Re: low or super low ESR + impedance caps

A further note about ESL (equivalent series inductance). As VRM frequency increases the effect L increases, it stops ripple being bypassed by Al electros. Motherboards include SMD ceramics that have vey low ESR and ESL to bypass the high frequency spectrum.

So basically ESL is not too important in Al caps and you do not need to add any caps (tantalum or ceramic) to improve matters. In fact manufacturers add some ESR and ESL to help share the ripple load passing through parallel caps.

[trodas]

Re: low or super low ESR + impedance caps

willawake - thanks Be ready for my other projects as well


davmax - I do understanding the ESR and ESL. You explained it pretty well, tough. The only thing I could disagree is that the inductance is loseless - this is not true and you know it. Late you elaborated it pretty nicely that you are aware that inductance could cause several loses in high-frequency.
Don't you think that this loses has to generate heat? I think they does. Of course, they are unwanted and should be filtered with ceramic caps in parallel, but some of them remain there - because hey - it is mostly about pulse switching on mobos, is not it?

Besides, the most important thing is, that this is only the third reason and I even started it with the world "maybe" Therefore I'm well aware that if there is ANY benefit, it is going to be meanligness (maybe less that 0,1°C gain?) - but still, there will probably be such gain, when Samxon GA is used instead of other less excelent caps.

Could I ask where the "CPU with 30amps at 1.66V is 55milliohms" value come? And what is value for AXP-M at 1.800Vcore and 2600Mhz?

Anyway, then end part of the DFI LP B mobo consisting of only there 3300uF KZG 6.3V caps, and they have 0.012 ohms, so /3 = 0.004 ohms (or 4 miliohms) witch is considerably HIGHER that on the Epox 8RDA+...! I dont like it. I want there 4700uF Samxong GA ones. They got not only far more "punch" - instead of poor 2800mA they can deliver smashing 4890mA ...! Now THIS could reduce the number of switches need for the mosfets and therefore cool them a little, don't you think?
Also, they got 6,5m ohms impedance per piece, so 2.1m ohms as result. Half!
(tough to get where Epox is I need to add another in parallel to them...)

I must be mad if I did not want them!

I like you calculation you done for the exploded caps, even I'm not sure where you come with the load resistance 2.860 ohms - you calculated this from the power drawn out of the PSU? Hmmm...
After all, both of the caps are there in parallel (the exploded 1500uF ones on my mobo) and since I probably todadys get hand on Nichicon HE, we see what happen.

Chemi-con KZG 1500uf 16V - impedance 0.013 - ripple current 2550 (2000 hours) 10x20
Nichicon HE 2200uF 16V - impedance 0.027 - ripple current 2230 (10000 hours) 12.5x25

Sure HE means HighEndurance, but the impednace is somewhat high and ripple somewhat low... Is there a way how to make the cap better - I mean long time ago I heard something about like "cap formating" - eg. first slowly (using resistance) charge the cap to it's capacity and then slowly (say 4,7k resistor) discharge it few times before use it? Could this help obtaining the most optimal from the cap?
I asking because the cap parameters could differ by 20%. This means that the ripple current of the KYG cap could be smaller or bigger by 510mA. Eg. 2040mA or 3060mA.
Same for the Nichicon HE ones, so, getting the best out of them is in order.

Sadly I can't measure them. My multimeter can go only to 200uF
...and even the dead caps show farily much more capacitance, after all...

I finally understand your concern with the PS ripple - since no more reports of Chemi-con KZG cap failure there was before me, you logically looking for other reasons why the failure happen. And the PSU is the logical conclusion. I gotta get osciloscope to check this... What do you think about this one?
http://www.professionalequipment.com...qx/default.htm
I'm especially interesed into the "glitch capture" and "power quality by capturing noise spikes, glitches and those evasive intermittents"

And still, reading your conclusion I come to conclusion that hi-quality SMD ceramic caps soldered between each of the caps aren't bad idea, after all. It definitively helps to:
Filter very fast ripple spikes out.
Lover the ESL.
Incease the ripple current (tough very little).
...and definitively adding a 47 or 100uF ceramic SMD ones in parallel can't hurt, can it?

Witch are the best ones?

[davmax]

Re: low or super low ESR + impedance caps

Trodas, I am using standard electrical theory. Inductance is definitely lossless just as capacitance is. In caps the loss is in ESR and in L it is in the resistance of the inductance material eg copper or in low frequency components in the core material. Yes there is impedance and voltage drop but no energy loss, the voltage and current in L are 90degree phase shifted, in fact opposite phase shift to that of a capacitor. At resonant frequency the phase shifted components of L and C are equal and opposite leaving only the circuit resistance. You should study this up.


The 1.66V and 30 A is used as an example of the type of values in a CPU. To illustrate the kind of circuit load resistance and cap ESR levels that exist. And to enable an example of the change of load value seen at the VRM input filter. The 2.8ohm is not taken from the power supply, again it is standard theory. For simplicity it is assumed that the VRM is 100% efficient (of course it is not). Impedance through the VRM changes according to the square of the ratio of input and output voltage using the formula Watts = volt squared/Z. So taking 12V/1.66V = 7.23 then square it = 52.26 now multiply 55 milliohm X 52.26 = 2.87ohms.

You want something like GA at 4980 mA. I honestly say that you will get nowhere near that ripple current in the caps. Yes the GA will run cooler.Where will the highest temp be in your system? Internally in the cap or externally from MOSFETS. Yes the lower the ESR the higher the ripple current in a cap will be for a give ripple voltage. Most mother boards as you know distribute the ripple over several caps, the 8RDA+ with six 3300u in a conservative design that is able to use higher ESR caps. Yes the DFI needs lower ESR caps.

Pulse switching in power supplies is normal, pulse switching produces a wide spectrum of frequencies, that is why ceramics are used to filter the higher harmonic frequencies. I believe you need to leave that part alone and just focus on the problem of the input caps.

Is buying a scope good value? When the cost of replacing caps in the PS can be so low.
That scope will do but for future VRM's operating in the Mhz region it will not be any good for looking at waveforms in that area.

I do not agree that you need to add ceramic caps.They are already installed near each CPU socket. Give the designers some credit, they do understand these things.The problem has been the purchasing people, buying low quality electros.

[trodas]

Re: low or super low ESR + impedance caps

Hey, let's not fight about nothing! A good shop suggestion would help much more to me!


If you really want fight, then just remember, that the inductance create a magnetic field when current is passing through the coil. The field was not created out of zero energy, so that is where the loss begin.
Also the loss is proportionaly higher as the frequency and increases - eg. high impedancy is has almost 100% resistance to the high frequency currents.
Another reason why inductance IS loss. In mentioned case (imagine BIG coil (over 100 000 threads) on HUGE iron core and 10Mhz frequency : ) is the loss 100%
(if there is no loss in the inductance, then please explain to me, how come the coils are burning hot! : )

And yep, I'm avare that the coil delay current while the cap delay voltage, as you mention. In ideal case shift it back/forward by 90 degrees.

You have point, tough, into the resonant frequency are the inductance of the cap somewhat irelevant - eg. affecting only and only quite minor, the resulting frequency of pulses need to supply the CPU with voltage and current.



It also explain why LastViking "get away" with adding a lot's of cap's to his mobo. Yes, he added also a hell lot of inductance (in parallel connection and with caps - inductance are actually building up, while resistance is going down...) but it aperantly did only slow-down the resonancy clock and therefore he was the only one who almost touched 300Mhz barried on FSB on nForce2
http://www.xtremesystems.org/forums/...ad.php?t=31703



Thanks for explaining the calculation. God know how much my CPU consume ampers But for 1.800Vcore and say 40A it should be 0.045 ohm (45m ohms) and therefore 12/1.8 = 44.44 x 0.045 = 2 ohms.
Not so much different and certainly far away from the mosfets specs.

So, if none of the ESR anfd ESL does matter much, how come my Epox and DFI mobos died so fast when I added another (not low ESR) caps in parallel to - in case of Epox - dying ones?

You want something like GA at 4980 mA. I honestly say that you will get nowhere near that ripple current in the caps.

Well, according to the specs - I should get it :
The point of having such nice caps in mobo is not that they will run a 0,1°C cooler, but that the mosfets probably run up to 10°C (?) cooler with them - at least I hope.

And about the ceramics - yes, that is EXACTLY what I hoping they help me, when put in parallel with each major cap on the mobo :

You think that future VRM circuit move from about 50-100kHz or whaever is used now into Mhz? Sounds logical to me, I tend to believe this move, but why they aren't did it already, when in quite short - frequence buy power? There has to be some limits and that is why the frequency stay relatively low, nope?

And yes, into the CPU socket there are lot's of ceramic caps as elsewhere on the mobo. Designers credit assured. However there are FOUR bigger ones missing on the back of the CPU already, see there: http://ax2.old-cans.com/s.php?p=badt...3&c=8&d=1&v=v2
(the last 4-th position is hidden behind a cap)
And it did not end there. Similar 4 bigger caps are omited from the inside of CPU socket! Also there is 6 empty postions for smaller ones. I'm affraid someone overoptimized there. Also, filtering the voltage better before (adding the small ceramic caps in parallel to each major cap on the mobo) could only make things slightly better, right?

[linuxguru]

Re: low or super low ESR + impedance caps

The heat in the VRM inductors does not come from the inductance - it comes from two sources:

1) Ohmic (resistive) heating in the copper wire used in the windings.
2) Core heating in the Ferrite Toroid due to hysterisis and/or saturation.

Both of these can increase substantially with increased current draw from the VRM. That's one of the reasons why you can't simply add a low ESR cap and overclock your CPU. The limiting component now becomes something else - inductor, PCB traces, MOSFET, upstream capacitors, etc.

The ceramics on the back of the CPU package are installed during manufacturing based on the current drawn by the die during power-supply binning. If it's a low-power part, the currents drawn are proportionately less, and fewer bypass caps are required. Just a manufacturing optimization, which has the unfortunate side-effect of making overclocking difficult.

[trodas]

Re: low or super low ESR + impedance caps

linuxguru - well, your first point - resistance - is sure right. But how much heat does it can generate? There is just 6 (!) threads by very THICK cooper wire. How much resistance can this has? 0,001 ohms?
Got my point?
The resistance is maybe responsible for 0,1% of the heat. Nothing more.

Your second point about core heating because of hysterisis and even possible saturation is also walid and major cause of the heat. Tough this it is just substitute word for impedancy loses - okay, far more sophisticated and precise, however we talking about the same stuff.
Impedance is determined by the core material (and shape, toroids being the best) and number of threads of wire around the core. Now it is the (maybe wrongly writen, by my vocabulary is limited, so...) inductance loss, witch is consumed by the polarization of the core material. (and at some point - saturation point - does not the inductance increase anymore, because the magnetic flow into the core reached the maximum for user core material and therefore increasing voltage/current futher lead only to more and more loss = heat)
In short - creating magnetic waves is not energetically for free and the loss of energy could be - at least I believe it could - simplified as induktance loss.

The problem with OC you are talking about quietly assume that with current components are the machine running at it's limits.
This is not true. W/O modifications I was able to hit 2813Mhz, witch is IMHO not so bad for unmodded board. Others scored even more. My friend Stephan from Germany, aka SAE hit 3200Mhz, witch is and remaining to be the WR for AthlonXP - with the very same board. Of course the quality of the components does vary and he reach such OC just because chilled CPU can run higher with less voltage need. Not to mention his mods to stabilize the Vcore, disable the OCP protection and so on...
My goal is not to attempt WR, but to make the mobo reliable longtime respectable performer. It was running well (except the audio glitches, tough) at 200x13 - 2600Mhz, witch is not entierly bad for fanless overclocking. I would like more run at 230x11 or even 225x12 = 2700Mhz, witch could be possible with quality components and well-dimensed caps. Samxons GS sound lovely for CPU powering at clocks like that :
I should also mention that when I did not play games, I fold and therefore the machine is under 100% load all the time

And exactly the manufactring optimalization is what I looking forward to disable and return the caps where they belong : Since I was running fanless, I has to keep hard to limit the Vcore / V dimm voltages to minimum in order to prevent overheating. So, if using Samxon GA 4700uF caps, bypassing them with 100nF ceramics and soldering back the ceramic caps into and bellow the CPU (at least the 8 big ones for sure) coudl result that I need only 1.775Vcore for 2600Mhz and not 1.800Vcore - then it was worth it

But where to get such quality caps?

[linuxguru]

Re: low or super low ESR + impedance caps

Trodas, you're mistaken about inductance. Inductance is loss-less, i.e. the Impedance jwL or sL in an ideal inductor is lossless. An air-cored coil is close to an ideal inductor, as also a ferrite-cored coil with a small air-gap, in which the core does not reach the saturation flux-density, Bsat.

The current in an ideal inductor will tend to flow for ever if the ends of the inductor are shorted - this is similar (by duality) of the voltage in a capacitor staying constant if its terminals are open-circuited.
Of course, in a real capacitor, the charge will leak out over time, and the voltage will drop. Likewise, in an inductor, the current will drop to zero eventually due to ohmic losses.

If a toroid is getting hot, then magnetic flux-density in it is close to or exceeding the saturation flux density. This is due to the current in the inductor exceeding the maximum rated current. Over time, the magnetic properties of the toroid will get affected due to the heat, and the MOSFETs will fail due to excess current.

[linuxguru]

Re: low or super low ESR + impedance caps

The actual energy lost in a ferrite core has two components - hysteresis loss (due to the B-H curve enclosing a finite area within it) which occurs at any frequency, and eddy-current losses (relatively small when the ferrite particles are fine, but frequency dependent). In this case, it is the hysteresis loss that dominates.

Note that in a toroidal coil with a vacuum-core, the core losses are zero - even though the coil has a non-zero inductance L = N^2*mu*A/l. (Here, N is the number of turns, A is cross-sectional area, l is the circumference of the center of the toroid, and mu is the absolute permeability of free space). The impedance in such a coil consists of the inductive component, sL and the resistive component, R. The inductive impedance sL is completely lossless.