Titanic questions

[LaurV]

Originally Posted by Madpoo

I wonder if you couldn't just wire in a POT in parallel, have a nifty little knob to adjust the resistance across the shunt instead of shorting it entirely?

No way! Unless you can find a pot about 0.01 ohm, standing 30 Amps through it, haha. A normal pot will let the magic smoke out immediately. If the card consumes say, 300W from the 12V bar, then P=UI, I=P/U makes a current of 300/12=25 Amperes through that resistor. You need a power mosfet there, to work as a controlled resistor, but a power mosfet still takes about 0.2V drop at 25A (and you need a good one, a normal one has higher drop). But we have on our resistors only U=IR=25*0.005=0.125V. The best mosfet you can use at ~25A will not give you so low drop.

You may be able to "manufacture" such potentiometer (or better say a winded power rheostat) by yourself, if you take a 3mm or 4mm dia iron wire (or nichrome), about 5-10 cm long, shape it like a horseshoe, invent a kind of cursor from the same material, that can make a good grip, etc., you got the idea. It will also have the size of a horseshoe, and it may work, if you manage not to lose more than few hundred micro-ohms over the soldering points, contact wires, and contact between the cursor and the wire shoe. See how you manage to put this inside of the case, possibly under the card's cooler, because I don't know if you can afford the resistance of some long wires (and another soldering points) to put it outside the case. And if you manage to put it inside the case/card, see how you can adjust it by hand... My idea is that you first start trying to place a real horseshoe inside of your VGA card, and if you succeed, than you can think how to make the potentiometer

Well, this was a bit exaggerated...

Joking apart, the thing can be done, but you need to replace the acquisition part too. Eventually don't touch the resistors, but only cheat the opamps/comparators that read the voltage drop. This can be easily done, but does it worth? If you stress that card too much it consumes double power for only 50% (or so) more output. Better buy another card, which also doubles the power, but doubles the output too...

[ewmayer]

LaurV, your geographic locale is getting in the way of your GPU-blasting - you need to work out some kind of deal with someone living in cold climate: you send them a super-duper hot-running GPU and maybe subsidize their E-bill, you get the crunching credits, they get a free high-tech space heater and maybe some quality gaming time-sharing. Got any pen pals living in Greenland or at an Antarctic research station?

[kladner]

Originally Posted by ewmayer

Got any pen pals living in Greenland or at an Antarctic research station?

Iceland or Norway. Sometimes really cold, with cheap power, as I understand. I think both have hydro, and Iceland has geothermal, to put it mildly.

[Mark Rose]

Originally Posted by kladner

Iceland or Norway. Sometimes really cold, with cheap power, as I understand. I think both have hydro, and Iceland has geothermal, to put it mildly.

90% of Icelandic homes are geothermal heated. Iceland doesn't really get cold, either. Like Norway, its temperatures are moderated by the ocean.

If you want cold with cheap electricity, try Quebec. :)

[kladner]

Originally Posted by Mark Rose

90% of Icelandic homes are geothermal heated. Iceland doesn't really get cold, either. Like Norway, its temperatures are moderated by the ocean.

If you want cold with cheap electricity, try Quebec. :)

Really! (on the cheap power, anyway.)

I thought Iceland used some of those many heat sources to generate electricity, too.

[LaurV]

Originally Posted by ewmayer

LaurV, your geographic locale is getting in the way of your GPU-blasting - you need to work out some kind of deal with someone living in cold climate: you send them a super-duper hot-running GPU and maybe subsidize their E-bill, you get the crunching credits, they get a free high-tech space heater and maybe some quality gaming time-sharing. Got any pen pals living in Greenland or at an Antarctic research station?

[edit: quote included, everything in between is crosspost, I started to reply to Ernst hours ago, but I also have a job to do!]

Ha! You know, during my life I had many job offers which I didn't take, and I don't regret, except for two.

One, which actually was not a job offer, but only a couple of interviews, was Microsoft, beginning of 90's when all my colleagues (classmates, sorry Serge , but also friends not in the same class, and some of my former professors and former students) went to work there, other few of them made their own companies (Red Hat is an example, one of the founders was my student). Microsoft did huge recruitment in eastern Europe in the beginning of nineties, and it took from there mostly Romanians, and mostly from my area/city/university. Bill Gates itself said that one in five of his collaborators are Romanian. I could add, one in 10 is from my area But for my part, I screwed up bad one of the interviews, when they asked me if I want to relocate, I replied something along the lines "for some years, but not forever". After which they didn't talk to me much, anymore. This I found later, from a friend working there... At the time I was young and stupid and I was thinking that I am a good Romanian, and after few years of working for "real money" I could return to my country and actually do something there, in that silly country. I didn't consider that no company in the world would be so stupid to invest money in me to let me go exactly when I was starting to learn and somehow return their investment... Well, that is that.

The second time was Frisk, the producer of F-Prot, arguably one of the biggest antivirus producers in the world at that time (mid of year 2000). I think I already boasted here around about my diploma thesis involving computer viruses and data protection, and about my "small" collection of few thousand families of viruses, all disassembled and commented and stored on floppy disks. Still have them. I did many submissions to anti-ve companies too, so they could improve their products. And I may have been the first in the world able to remove "one-half" virus, including the polymorphic part and decryption of the encrypted part of HDD (the healing/decryption program was in use for a while at the Uni, at the time when mcafee was only "learning to recognize signatures", that could be added to an external signatures.hex file; as a joke, we - the lab assistants working there - used to add signatures of games like pac-man and prince of persia to mcafee's signature file - in hex - to forbid students in the labs to play games on Uni's computers - the programs "recognized" were automatically blocked to run or exited from memory on the old x86 machines - mcafee was more useful as that than as antivirus). Well, in the mid of 2000, I was freshly moved to Thai, and beginning to like the warm climate, but in the same time finding new, unexplored territory (read: new Asian viruses), so after one of such submissions of virus analysis, I had a long talk with Mr. Skulason, the founder of that company, and we got to the point where we had to fix the salaries and the relocation costs, but then I most probably upset him somehow (no idea if I said something, or did something) because he didn't reply to my last mails. It may be that he didn't get them too, but I didn't insist too much, my family didn't really like to move to "north pole". The company is based in Reykjavik.

For the M$ job I was upset few months then I forgot about it. Till later, when I talked to my man and found the reason, I went upset again (angry with myself) for another few weeks, then I forgot again. Time heals things.

But for Frisk job, hmmm... I regretted ever since I started playing with noisy computers and GPUs and I am still regretting it today. I think it was the biggest failure in my life that I didn't go to Iceland. Beside of a "nice climate" to cool my hardware, they have pension plans, they give citizenship with ease, and you can buy your own house, none of these three things are possible here in Thai, for a farang (unless you use some border-of-the-law tricks like making a "fake" company, which produces nothing but then it can buy that house, and get you an investor visa, etc), and after 15 years here (and half), I am starting to regret it even more... Iceland would be a paradise to live now, already having my place and my frigorific room full of GPUs... yarrrrr....

But I am targeting southern Australia in few years, my girl has a fix idea to study there, and I can't let her down...

[snme2pm1]

Originally Posted by Madpoo

I wonder if you couldn't just wire in a POT in parallel, have a nifty little knob to adjust the resistance across the shunt instead of shorting it entirely?

I don't know where you can buy a POT for milli-ohm purposes.
Even wires to attach such a device might incur unwanted milli-ohms of unwanted resistance.
I have equipment that can measure very low Ohms measurements, and such can be used to detect the shorts on small PCB's, for example to isolate a rogue Tantalum capacitor on one occasion that I recall.
You suggested in parallel, which means that only a portion of current would pass through such a device.

Originally Posted by LaurV

No way! Unless you can find a pot about 0.01 ohm, standing 30 Amps through it, haha.

Well, Arron spoke about a parallel arrangement which didn't need to take the entire current, but perhaps it might rather quickly take a huge share of the current if the person undertaking such an experiment went crazy, which is of course entirely likely!

LaurV went on to talk about this being a MOSFET application, however I would find this a challenge to implement usefully.
A couple of years ago I was planning for some low voltage distribution about the house, but wanted an electronic auto-reset fuse.
I figured that I could use the MOSFET Drain Source Resistance (Rds) to detect the load, and couple that with some rail to rail op-amps to then for several seconds discharge the gate upon an over-current circumstance, thus denying the load circuit.
Whilst I still might implement this when I get around to it, it is a preference to keep 0V connected thorough a system, so for example breaking that arrangement by introducing load sensing resistance on the low side would be a nuisance.
From what LaurV has described (and I haven't had cause to glance) current sensing is conducted on the +12V (high) side in recent GPU equipment.

The simplest first consideration might to be to deploy a P-channel MOSFET, such as would be suggested by that arrangement, however unless the price and availability of P-Channel devices has significantly changed in recent years, since I last bothered to look, affordable P-Channel devices would have very high Rds characteristics, so likely untenable.
Another approach would be to consider an N-Channel device, and there are many to choose with attractive Rds figures.
The trouble with that is that such a device would need to a control circuit able to pump the gate from a potential from at least 6, more probably more than 10 Volts above the 12V rail.
Thus some awkward gymnastics using a volt pump or converter would be needed to entertain such.
Further, driving the gate of a MOSFET is very unlike turning the knob of a potentiometer.
It is most awkward to contemplate controlling the absolute Rds from the gate; the transfer relationship is not well regulated among manufactured devices.
Manufacturers do not supply devices with precise relationship between Vgs and Rds.
Mostly MOSFETs are best to be either ON or OFF and spend as short as possible gate charge or discharge time in between.
That is very unlike a potentiometer, and more like a two speed driver, at any moment utilising either Go... or Stop...

[LaurV]

Originally Posted by snme2pm1

Well, Arron spoke about a parallel arrangement which didn't need to take the entire current, but perhaps it might rather quickly take a huge share of the current if the person undertaking such an experiment went crazy, which is of course entirely likely!

You didn't really get it. It is not about the current, but about the voltage drop. On the actual 5 mR resistor, if the board consumes (say) 240 watt, at 12V, you get a current through it if 240/12=20Amps, therefore the voltage drop U=IR=20*0.005=0.1V, or 100milliVolts on that resistor. THIS 100mV is measured by the (differential?) opamps and the power is computed. Higher voltage drop means higher current, therefor higher power, so the "internal regulator" will cut the clock or voltage down. You have to make your "internal regulator" believe that the power is smaller, so he would increase the clock and/or the voltage. This you can do by making a LOWER voltage drop on the resistor (or instead of it)*. In any case, the branch with lower drop (lower resistor) will also digest higher current. I=U/R. You can not modify Ohm's law. Mosfets: higher current=higher voltage drop. You have to make a voltage drop of 0.05V over that resistor (or whatever you put there) to make your card believe it only consumes 120W. And you can not make a "separate trick" with the opamps, like without feedback, because in this case the power will increase until the card burns and it will still believes it consumes 120W.

The problem is not the current, but the (extremely small) voltage drop, which you have to make it smaller, but still mentain the feedback. This you can't really do it "adjustable" (this was the topic, making it adjustable) without putting your own opamps there, and in this case you cancel the measuring part completely, let the resistor there untouched, but put your own measuring opamp, with a lower gain, and cut out the tracks for the old one. Or change the gain, etc. Assuming you can find out how it is done, because from all this discussion we still have no idea if and how the opamps are used (or specialized circuit) to measure the power. I said "if" the diff opamps are used to measure the drop, but we still don't know that.

The discussion was only that you can not make it adjustable, because with all the wires, potentiometer, soldering joints, whatever, the resistence (therefore the voltage drop) will be higher. Five milliohm is freaking low resistance. To decrease it is easy, just mount another resistor in parallel, or substitute. You may even use few parallel resistors and power switches (with resistance of microohms) to enable/disable them, giving different power ranges to your card, like "normal", "hot", "hotter", and "keep our freaking hands out" (or "kiss me I am hot", like some bikers use to write on their exhaustion pipes, hehe), but I won't call that "adjustable".

In real life, I mean for variable-power applications, you don't change the shunt resistor to vary the power. You let it there, as it is, but you "read" some values from it, amplify them with some gain, and change the "action" you do with the values you read from this resistor. Well, not the best formulation, but you understand what I want to say. You wold need more hacking into the board, and at the end the best way is still to put a lower resistor there, and change the software (card's bios).

----------
*or by modifying the table or the calculus in the bios of the card, but we were not talking about that.

[Madpoo]

Originally Posted by snme2pm1

I don't know where you can buy a POT for milli-ohm purposes.
Even wires to attach such a device might incur unwanted milli-ohms of unwanted resistance.
I have equipment that can measure very low Ohms measurements, and such can be used to detect the shorts on small PCB's, for example to isolate a rogue Tantalum capacitor on one occasion that I recall.
You suggested in parallel, which means that only a portion of current would pass through such a device.

Correct... my thinking was that instead of shorting it entirely, you wire a POT with a similar resistance range in parallel, just enough range to give you something between 0 and 0.5x of the original resistance, instead of just 0 with a solder bridge.

Gives you some finer control over it... but then I don't know what that part of circuitry does on the GPU apart from me skimming LaurV's post so I don't know the power involved. I was assuming that since there's a resistor already there, it must have reasonable power requirements...not the kind of thing you'd need a huge "dummy load" resistor you could fry an egg on.

But hey, I may have misunderstood the setup and my idea is stupid. I can live with that.

[snme2pm1]

Originally Posted by LaurV

You didn't really get it.

Wow, are you serious or just trying to wind me up? No wonder the moderator has changed the thread title.
I would have imagined, from some of the language that I used, it should have been obvious that I am an electronics engineer that actually does known a thing or two about the subject.
I easily understand what you are trying to achieve.
There may also be some cross language communication difficulty.
The notion of using a MOSFET was introduced by yourself in response to Aaron proposing the use of a potentiometer.
I furthered that discussion by pointing out some technical difficulties to make such a proposed deployment a dubious practicable exercise as a replacement for said non-existent potentiometer.
I agree that it would be a more desirable approach to have a fixed current sense resistor, as exists, unless you just want to bridge that with another suitably chosen watt waster.
As to the idea of modifying the operation of the sense circuit; Apart from fault repair work, I have been known to analyse and slightly modify the operation of manufactured equipment in various ways, though it can be an eye strain gathering the information to comprehend existing equipment, and identification of components, and getting their specification documents can be a challenge.
Very close to me, my bedside table has a couple of devices that annoyed me sufficiently that I did modify them!
By all means, if source for the firmware can be modified, that might be a convenient approach.

[retina]

Just buy another resistor exactly the same as the existing one, and solder it directly on top of the existing one, in parallel. You'll halve the total resistance and double the power input to power measured ratio. You might also turn your card into a fire hazard but that is your problem to solve. Don't say I didn't warn you.