International Rectifier is one of the oldest semiconductor manufacturers and it's Californian. Infineon is the semiconductor shard of Siemens, Germany, and it was basically an international behemoth from the onset, running semiconductor manufacturing both in Germany and as a joint venture with UMC in Taiwan, and they've acquired things left and right over the years.
You mentioned that you didn't feel like you did I great job on this video... you're wrong. You did completely fine. I actually enjoyed sitting through the full 40 minutes (sped up at 2x of course, so I guess 20 minutes). The only thing I would mention would be to link some example data sheets you pulled your numbers from in the description. Thanks for teaching us a few things.
So my display card is dead, and i'm loking for solution, and woala and here i am on how vrm work series. Thx for free knowledge, very much appreciated.
There are some sound problems at the end. 39:36 for example. Also I recommend watching in 1:5 speed. I would also like to see more of this types of videos.
Didn't Volterra made MOSFETS out of silicon? They were smaller than your regular MOSFET, included high and low side FETs in one package and were rated for 40A. I remember having those on my EVGA X58 Classy.
i wanna know which chip is switching the frequenzies on wich the fets are switched on and of ... because im in gpu repair for some time now and wanna know why my vrm ic is blowing every single time its replaced and first fired up :D
For the equation for the switching loss, shouldn't the voltage have its own triangle? This would result in higher current limits, which was the point of not using continuous drain current.
The equation 100% works I just didn't want to work the proof for it(I also did just copy it out of an article written by an Texas Instruments Employee). The basics of the proof go: W = (0.5 * V * A + (Tr + Tf))(this whole part converts to joules) * KHz(this converts the joules back into watts). Also the current rating using the switching + conduction loss is higher than the continuous rating. So this method does exactly what I was aiming for. High side FETs like the IRF6811 go from 32A at 125C to 45A+ at 125C(1.5V 500KHz)
So would it be wrong if I squared the 0.5, (Tr+Tf) and frequency and used the delta of input voltage and minimum voltage drop instead of just the input voltage? Would the higher current this results in blow it up?
That should be the voltage for just the switch times. You explained the current component of the heat with the triangle, but the voltage component changes the same way.
This is too long to explain in just a comment. But the equation works as is. I could do a proof for it in another video but I'm not a physics channel and really don't want to spend the time on that.
I have a good question... I think. You said on the other video that the way you convert 12V to 1.2V is with this inductor thingamabobby charging and discharging producing a zigzag voltage pattern. The 12V coming in is converted from 120V... but it's AC. Does this conversion in the power supply also use an inductor? My main question is does the main 12V coming in also have a zigzag pattern? Is anything done to alleviate this or does the CPU VRM inductor charging completely negate the need?
Shouldn't the V in the equation for the measure of the W for the VRM be 2.2V instead of 12V? And another question on Tr/Tf are them from the controller(like an IR35201) or from the power stage(like the TDA21472)?
Well done. I don't see in the formula de internal resistance. With higher current output the mosfet gets warmer, what up the internal resistance and therfore more heat. With double the switching frequence expotentially the internal resistance should go up too You have in the conduction loss the I2 but i don't see that habit in the switching loss in your equation? Or is this common electrical circuit behavior covered bei the ROJC? PS: Where does this C° originate from - I'm used to °C.
why don't they just stick out a probe - through the casing that one can use measure the j-temp? Why do they make things so complicated when there are simple solution?
I'd like an answer to a question I just couldn't find anywhere. ASUS STRIX RX 470 4GB has such configuration that VRMs are quite close to one of the VRAMs. Tom's Hardware raise a question of dangerous temperatures reaching 96C on one of the Samsung VRAM chips that are, I think, rated for 95C max even while the GPU core is just hovering around 62-63C. In which scenario is this dangerous, and will slight overclocks, or power draw limitation adjustments hurt the lifespan of the card?
VRAM and RAM are generally really really hard to kill. The only issue I could see is that hot VRAM clocks worse and it might not be stable if it gets really really hot.
ou, that was one of interesting math lesson in a looong time))well done) but...still...how i can measure current of 6600k when it runs at 4.83Ghz under 1.425v up to 1.488v. passing prime...93C-cpu, ambient around 18C whitch is not so much comfortable)))) well...and i think it can a little bit more....pff)
Hi Buildzoid I found out about you from Gamers Nexus channel. Your videos are great. Anyway, I wanted to know if I could discuss some technical things with you? Eg. I have a project where I want to generate 10+ V potential as excitation signal for a device which then returns me the signal in milli volts as a function of some real world value. My board can only work on a cheap 6V lead acid battery. The excitation signal needs to be noise free as any noise in it will lead to noisy data signal (the milli volt signal). How to get 10+ V from 6V battery? Step up buck boost converters are noisy and costly. I was thinking maybe I could somehow generate a +5V and -5V signal using linear power converters and thus end up generating the 10V potential (peak to peak). Can you share your email or something with me so we could discuss? Thanks.
So that "Super duper low RDS MOSFET's !" - that all MB guys are so happy about on boxes - is only useful in Low side ? I guess marketing guys didn't know what to do with Tf and Tr for high side... Good to know RDS comes from "I squared R times duty cycle" part.
Super low RDSon is easy to market. RDS = Drain To Source on Resistance. Looks like I forgot to mention that. Basically it's electrical resistance of the MOSFET when it's turned on.
Good to know :) Here's another question : You said that we can't measure silicone temp inside MOSFET's and can only do Tcase with special tools. Q: How can a VRM controller (?) know that one of the phases (ie. MOSFET's) is overheating then ? I would like to know why AMD AM3+ MB's can throttle themselves despite low-ish CPU temp (CPU being FX-8xx0 series). I'm assuming that R9 290X or Fury VRM's can do this as well.
They don't know internal temp. They just have a thermal probe somewhere in the VRM section and when that detects a too high temp it starts to throttle. This temperature is set low enough that under normal conditions there should be no risk of the FETs dying. So basically they do a calculation that say at XC the VRM can't handle Y current and when that X temp is reached they throttle the CPU. However this is usually a temp only safety system and doesn't actual take current into account.
Thank you for explanation :) Wish they (MB manufacturers) shared that probe singnal somehow with users. Sometimes it's just tiring to explain, why there is a thermal throttle on performance when CPU/GPU temps alone are OK. PS. Great video !
Nice video,learned lots from it,but even though you tried so hard and got so far to about 40min. it's just too much ramble in the end(it doesn't even matter). I have a question,my RX 480 is a reference sapphire watercooled(just the chip,still have the default plate cooling VRAM and VRM with a fan on it),i'm currently bios editing it and want to lower the memory timings and keep it at 2000MHz,if i try 1500MHz timings copy pasted will it work fine generally?or even at 1625MHz timings i could do 2250MHz right?
I'll talk to myself a little,it's good,1500MHz no good,blank screen while flashing and crapped out,i reflashed with another card a modified clocks and voltage bios with settings that worked in windows,but no good,exclamation point in device manager,only default bios works,i thnk i can only change the power limit and maybe timings without exploding everything.
talk to myself some more. what the hell?i just can't use polaris bios editor on this card,no matter what i change,it fails to work with drivers,is PBE only for aftermarket cards?
Are you on the new ReLive drivers? Because those re-enabled BIOS signature checks. www.techpowerup.com/228536/amd-bios-signature-check-re-enabled-with-relive-locks-out-polaris-bios-modders
They will probably disable the singature checks again some time soon. Up until the 16.9.1 driver there were signature checks. Then they stopped. Now they're back again.
Question: Do GPU Makers have better specsheets then you have? Because they also have to know that stuff and if i only know the typical RØJC i would tilt pretty hard and dont buy MOSFETs i cant trust.
yeah there are non public versions of datasheets that have more information than what I can usually get my hands on. The thing is if I design a VRM to do X. and the typical spec is capable of more like 3X then I'll just use the cheap FETs anyway.
First things first, thanks for the video in general, I really hope your channel does well I just wanted to point out that in that equation you have(as units of measurement) W=V*A*s/kHz, when AFAIK there should be written Hz not kHz because Hz=1/s and W=V*A, with kHz the results would be 1000 times lower
Generally speaking i like your videos,however i have one criticism. your ones look like sevens. I was listening while dealing with my grandson and glanced up and saw 72 volts going into the circuit and thought something is very very wrong here.
The first time I've seen a MOSFET datasheet I'd been wondering what I was looking at. Now all these values make more sense
International Rectifier is one of the oldest semiconductor manufacturers and it's Californian. Infineon is the semiconductor shard of Siemens, Germany, and it was basically an international behemoth from the onset, running semiconductor manufacturing both in Germany and as a joint venture with UMC in Taiwan, and they've acquired things left and right over the years.
You mentioned that you didn't feel like you did I great job on this video... you're wrong. You did completely fine. I actually enjoyed sitting through the full 40 minutes (sped up at 2x of course, so I guess 20 minutes). The only thing I would mention would be to link some example data sheets you pulled your numbers from in the description. Thanks for teaching us a few things.
Didn't even go 2x
"Mosfets are as turned on as they can be" :D Keep up the good stuff
Hey coming to this video 5 years later along with your other series. Your videos are very insightful and well explained!
Holding CTRL will align your line
WOW
I don't understand VRMs or MOSFETS or anything electrical. Still watching the whole thing because I'm learning a thing or two which is awesome.
I covered the bare minimum of making a working VRM in another video on this channel.
I really like these theory videos, there are very few places where you can about power regulation from an overclocking perspective. Moarrrr
So my display card is dead, and i'm loking for solution, and woala and here i am on how vrm work series. Thx for free knowledge, very much appreciated.
I liked this video, one of the best you've done.
I don't mind the rambling its normally informative.
38:35 "you want really low that on high side, you want really low that on low side". noted
27:18 "you can't really replace the MOSFETs" - actually, why not?
FIRST!
Actually Hardcore Overclocking ooh stop
en.wikipedia.org/wiki/Factor_of_safety
there you go, thats what you do when in doubt about actual values lol
you can't be because I can post comments before the video is public :P
and even it is uploading
40 minutes :O i have to watch this in 2 sittings.
or speed it up
Nah, i will never speedup good content :D
Thanks for the explanation man, more like this video please :)
There are some sound problems at the end. 39:36 for example.
Also I recommend watching in 1:5 speed.
I would also like to see more of this types of videos.
nope that's just me losing track of what to say. It sounds like a audio cut but that's just me hitting thought process wall.
theirs static a few seconds after that.
I'm sure his brain generates static when it's overloaded. Look at his hair.
Only thing I missed was the graph, why does the voltage drop when you turn it on?
Wait... Could I hand solder a "external socket" to the bga or socket, and then have power coming from an external VRM?
6:47 no, that's called Ø.
Didn't Volterra made MOSFETS out of silicon? They were smaller than your regular MOSFET, included high and low side FETs in one package and were rated for 40A. I remember having those on my EVGA X58 Classy.
Yep AMD also used them on a lot of their reference cards.
i wanna know which chip is switching the frequenzies on wich the fets are switched on and of ... because im in gpu repair for some time now and wanna know why my vrm ic is blowing every single time its replaced and first fired up :D
Good
good video but I think you should make a script next time to make it a bit more compact
For the equation for the switching loss, shouldn't the voltage have its own triangle? This would result in higher current limits, which was the point of not using continuous drain current.
The equation 100% works I just didn't want to work the proof for it(I also did just copy it out of an article written by an Texas Instruments Employee). The basics of the proof go: W = (0.5 * V * A + (Tr + Tf))(this whole part converts to joules) * KHz(this converts the joules back into watts). Also the current rating using the switching + conduction loss is higher than the continuous rating. So this method does exactly what I was aiming for. High side FETs like the IRF6811 go from 32A at 125C to 45A+ at 125C(1.5V 500KHz)
So would it be wrong if I squared the 0.5, (Tr+Tf) and frequency and used the delta of input voltage and minimum voltage drop instead of just the input voltage? Would the higher current this results in blow it up?
I'm not seeing what you're trying to do with that.
That should be the voltage for just the switch times. You explained the current component of the heat with the triangle, but the voltage component changes the same way.
This is too long to explain in just a comment. But the equation works as is. I could do a proof for it in another video but I'm not a physics channel and really don't want to spend the time on that.
I have a good question... I think. You said on the other video that the way you convert 12V to 1.2V is with this inductor thingamabobby charging and discharging producing a zigzag voltage pattern. The 12V coming in is converted from 120V... but it's AC. Does this conversion in the power supply also use an inductor? My main question is does the main 12V coming in also have a zigzag pattern? Is anything done to alleviate this or does the CPU VRM inductor charging completely negate the need?
No it uses a rectifier, using diodes (one way gates) and capacitors. Google AC rectifier or AC to DC convertor.
It's on Wikipedia or in Google images.
Shouldn't the V in the equation for the measure of the W for the VRM be 2.2V instead of 12V?
And another question on Tr/Tf are them from the controller(like an IR35201) or from the power stage(like the TDA21472)?
Well done.
I don't see in the formula de internal resistance. With higher current output the mosfet gets warmer, what up the internal resistance and therfore more heat. With double the switching frequence expotentially the internal resistance should go up too You have in the conduction loss the I2 but i don't see that habit in the switching loss in your equation? Or is this common electrical circuit behavior covered bei the ROJC?
PS: Where does this C° originate from - I'm used to °C.
MOSFET datasheets tell you the RDSon for -50 to +150C
why don't they just stick out a probe - through the casing that one can use measure the j-temp?
Why do they make things so complicated when there are simple solution?
I have a question how do i calculate the efficiency of a VRM?
I'd like an answer to a question I just couldn't find anywhere. ASUS STRIX RX 470 4GB has such configuration that VRMs are quite close to one of the VRAMs. Tom's Hardware raise a question of dangerous temperatures reaching 96C on one of the Samsung VRAM chips that are, I think, rated for 95C max even while the GPU core is just hovering around 62-63C. In which scenario is this dangerous, and will slight overclocks, or power draw limitation adjustments hurt the lifespan of the card?
VRAM and RAM are generally really really hard to kill. The only issue I could see is that hot VRAM clocks worse and it might not be stable if it gets really really hot.
Thanks a lot for a quick answer! I calms me down to know this should have no lifespan impact.
ou, that was one of interesting math lesson in a looong time))well done)
but...still...how i can measure current of 6600k when it runs at 4.83Ghz under 1.425v up to 1.488v. passing prime...93C-cpu, ambient around 18C whitch is not so much comfortable)))) well...and i think it can a little bit more....pff)
Use a multimeter, or do POWER/VOLTAGE (on load after Vdroop) = Current
Hi Buildzoid I found out about you from Gamers Nexus channel. Your videos are great.
Anyway, I wanted to know if I could discuss some technical things with you? Eg. I have a project where I want to generate 10+ V potential as excitation signal for a device which then returns me the signal in milli volts as a function of some real world value. My board can only work on a cheap 6V lead acid battery. The excitation signal needs to be noise free as any noise in it will lead to noisy data signal (the milli volt signal). How to get 10+ V from 6V battery? Step up buck boost converters are noisy and costly. I was thinking maybe I could somehow generate a +5V and -5V signal using linear power converters and thus end up generating the 10V potential (peak to peak). Can you share your email or something with me so we could discuss? Thanks.
So that "Super duper low RDS MOSFET's !" - that all MB guys are so happy about on boxes - is only useful in Low side ?
I guess marketing guys didn't know what to do with Tf and Tr for high side...
Good to know RDS comes from "I squared R times duty cycle" part.
Super low RDSon is easy to market. RDS = Drain To Source on Resistance. Looks like I forgot to mention that. Basically it's electrical resistance of the MOSFET when it's turned on.
Good to know :)
Here's another question : You said that we can't measure silicone temp inside MOSFET's and can only do Tcase with special tools.
Q: How can a VRM controller (?) know that one of the phases (ie. MOSFET's) is overheating then ?
I would like to know why AMD AM3+ MB's can throttle themselves despite low-ish CPU temp (CPU being FX-8xx0 series).
I'm assuming that R9 290X or Fury VRM's can do this as well.
They don't know internal temp. They just have a thermal probe somewhere in the VRM section and when that detects a too high temp it starts to throttle. This temperature is set low enough that under normal conditions there should be no risk of the FETs dying. So basically they do a calculation that say at XC the VRM can't handle Y current and when that X temp is reached they throttle the CPU. However this is usually a temp only safety system and doesn't actual take current into account.
Thank you for explanation :)
Wish they (MB manufacturers) shared that probe singnal somehow with users.
Sometimes it's just tiring to explain, why there is a thermal throttle on performance when CPU/GPU temps alone are OK.
PS. Great video !
Nice video,learned lots from it,but even though you tried so hard and got so far to about 40min. it's just too much ramble in the end(it doesn't even matter).
I have a question,my RX 480 is a reference sapphire watercooled(just the chip,still have the default plate cooling VRAM and VRM with a fan on it),i'm currently bios editing it and want to lower the memory timings and keep it at 2000MHz,if i try 1500MHz timings copy pasted will it work fine generally?or even at 1625MHz timings i could do 2250MHz right?
I'll talk to myself a little,it's good,1500MHz no good,blank screen while flashing and crapped out,i reflashed with another card a modified clocks and voltage bios with settings that worked in windows,but no good,exclamation point in device manager,only default bios works,i thnk i can only change the power limit and maybe timings without exploding everything.
talk to myself some more.
what the hell?i just can't use polaris bios editor on this card,no matter what i change,it fails to work with drivers,is PBE only for aftermarket cards?
Are you on the new ReLive drivers? Because those re-enabled BIOS signature checks.
www.techpowerup.com/228536/amd-bios-signature-check-re-enabled-with-relive-locks-out-polaris-bios-modders
thank you,so it's going to be like this every time i update/reinstall my drivers?amd is starting to look like nvidia or apple with locking things up.
They will probably disable the singature checks again some time soon. Up until the 16.9.1 driver there were signature checks. Then they stopped. Now they're back again.
Question:
Do GPU Makers have better specsheets then you have? Because they also have to know that stuff and if i only know the typical RØJC i would tilt pretty hard and dont buy MOSFETs i cant trust.
yeah there are non public versions of datasheets that have more information than what I can usually get my hands on. The thing is if I design a VRM to do X. and the typical spec is capable of more like 3X then I'll just use the cheap FETs anyway.
What mouse do you use to get the ability to hand write like this? What mouse settings?
I think its a wacom drawing tablet or something like that...
For this video I used a drawing tablet
Shouldn't it be 150 °C (not 150 C°)
and shouldn't the frequency be measured in Hz instead of kHz?
KHz = 1000Hz. I don't want to spend time writing out 200,000Hz
Actually Hardcore Overclocking what I meant to say is to get the right result shouldn't it be in Hz?
5 * 200KHz = 5 * 200,000Hz
does that answer your question?
First things first, thanks for the video in general, I really hope your channel does well
I just wanted to point out that in that equation you have(as units of measurement) W=V*A*s/kHz, when AFAIK there should be written Hz not kHz because Hz=1/s and W=V*A, with kHz the results would be 1000 times lower
Thermal _resisty_
Generally speaking i like your videos,however i have one criticism. your ones look like sevens. I was listening while dealing with my grandson and glanced up and saw 72 volts going into the circuit and thought something is very very wrong here.
He's powering a forklift truck not a CPU, Obviously Duuh
LAST!
4th
and 8th like