Measuring the effects of the ASUS Direct Power bridge

hey guys buildzoid here and today we're going to be taking a look at something that I've been wondering about for a very long time and that something can be found on the back of this Asus GTX760 direct cu2 so let's take a look at at the back of it shall we and of course the thing that I've been wondering about for a very long time is this thing right here uh now I haven't really been wondering about what it is um it's pretty simple but I have been wondering about like what difference does it make and so uh recently I bought one of these cards on eBay and hooked it up to the osilloscope to find out uh yeah how like what effect if any does this have on this card's power delivery now then what is it I still haven't told you that um so this is what I would call a power Bridge um the reason I would call it a power bridge is because I'm not aware of any like official name for what this is um as for what a power bridge is it's basically like this one is just a big copper plate that connects from the V core vrm which lives over here to the back of the GPU core now uh basically this goes so this goes in parallel with the existing internal vcor power plane of the card um and it basically just reduces the resistance of the vcor power plane by you know adding more cross-section effectively cuz this is a very thick copper plate so yeah uh it has a very low resistance so um that's what it is um as for why is it on this GTX 760 I don't really know Asus could have decided that you know they're not happy with how much power they're losing in the power plane of this card and they didn't want to use like extra PCB layers to deal with that so the solution that they came up with was this copper plate that you know just sort of goes onto the back of the card and you get basically more uh Power plane cross-section without actually having to add any more PCB layers um CU I'm assuming that there there just isn't any more space for a wider vcore power plane in the board um and uh or they just decided that this would be a really good marketing stunt right like it could actually be that they decided that they do want to improve you know the power D like so I I don't know how much of this is like a practical design Choice versus how much of it is a marketing Choice uh this was featured not just on this GTX760 I think they also did a GTX 960 with this maybe a 970 um there's a bunch of 10 series cards that do have the pads but don't have these power Bridges installed so like the uh like padding for the connection is present on a bunch of Asus 10 series cards but the actual power Bridge itself was never installed so they eventually stopped uh doing this but interestingly enough say a 980 TI streaks uh which you know pulls a lot more current uh or even like a 780 Ti uh from Asus like those are both cards that pull way more current than a GTX 760 does uh those don't feature a power Bridge so I'm assuming the main reason that they put this on this card is because this also has a really low layer count in the PCB um and maybe it is just a marketing stunt though yeah I like well not not tooo certain about that then again like I don't know how it compares to other GTX 760's and now that I've said that that opens up some horrifying possibilities for questionable GPU purchasing decisions doesn't it um cuz yeah like I I know what impact the power Bridge has on this GTX 760 but like is this carard just like especially bad in terms of its own internal power plane or are all 760's like that I don't know um cuz this is the only one I've measured but anyway so what kind of effects could we expect from this like mainly I would expect this to reduce the power losses in the power plane right cuz this effectively just reduces the resistance between the VC core vrm and the GPU core um as for voltage regulation I wouldn't really expect it to achieve anything based on my past experiences with DIY power Bridges cuz I've been doing modifications to cards inspired by the Asus power bridge for a very long time now one thing that's always kind of frustrated me about those modifications is that while they would pretty consistently reduce the power losses in the power plane cuz it's not really that complicated if you have a lower resistance there's less power losses in the power plane like that's just how that works but the thing that I've been kind of annoyed with is that I was really hoping it would deliver some kind of like Improvement in terms of voltage regulation like less undershoot because the impedance of the power plane would be lower but in practice I've never actually measured a Improvement in voltage regulation due to a power Bridge so I was wondering is it just because my DIY power bridges are bad or is this just like like power Bridges don't actually improve voltage regulation they just reduce the power losses in the power plane so it's a yeah that's the reasons I wanted to measure one of these for ages is like this is the like proper professional implementation of a power Bridge so you know is it just my my power bridges that suck or is this just like power Bridges just don't do anything for voltage regulation so anyway um yeah uh that's that's what I've been wondering about and that's what I went ahead and measured recently now then um yeah here you can see the power Bridge from another angle as you can see it's a copper plate uh it's very thick uh which was kind of surprising to me I didn't I don't know what I expected the thickness to be but yeah it is a substantial piece of copper um and then here you can see it without the sticker uh it's not it like the reason it's so yellow instead of you know like copper colored is because it's actually like it's an old card and it's very oxidized so um yeah that's not like I I I think if you got the card uh brand new from the factory it this would have been like this color not this yellow color um cuz uh yeah uh this does seem to just be like an oxide layer that's built up over the years anyway uh here you can see where I decided to hook up the oscilloscope so I'm just running coax all the way to the card this is my preferred method of doing uh voltage measurements because uh it is very good at not picking up things like vrm switching Noise by accident cuz yeah that's a really frustrating thing is if you put a probe or if you have any amount of like exposed wire and you run it like anywhere near the like switch nodes of a switching regulator it has a tendency to end up picking up the uh SW like the actual like pwm signal from the switch node um so yeah and and using like coax basically as like all the way to the board does a really good job of avoiding that issue um so yeah anyway uh as for why I chose the voltage measurement points I did well uh behind the core it wasn't really like I I wasn't really like there wasn't really too much consideration for other than just I want a good like physically robust connection and so the uh polymer capacitors are the best option for that cuz multi-layer Ceramics are not exactly physically robust things um so like soldering coax to multil Ceramics it like I have done it when desperate but I tend to avoid it because if you like you pull on the coax badly then it'll like break the multi-layer ceramic capacitor so I don't really want to do that if I can avoid it and so here uh yeah I soldered it to this uh aluminum polymer capacitor um because well there's no way I was going to be connecting it to this one or this one there just isn't enough space to get the soldering iron in there so I went with this one um as for the uh capacitor for the VOR vrm I guess I could have used the unpopulated pads over here um but it doesn't really actually make a difference cuz the the main thing like I don't actually care about like what voltage like what the peak to Peak measurement at the vrm looks like cuz like the the core is over here so like if you know if we have a lot of peak to Peak over at the vrm that doesn't really matter that much um cuz it's not like the core can't see that right um so really this is more for getting an average voltage measurement um from from the vrm area which is the point of that probe so uh um yeah that that was just and it was just kind of convenient to hook it up to the capacitor though I did want to hook it up relatively close to the power Bridge because I didn't really want to be measuring like the uh too much of the effect of the power plane itself on the card right so um yeah so that's why I chose the measurement points that I did like this one's just there wasn't really a better place to connect the coax and this one was just I wanted it to be kind of near to the power Bridge so um yeah as for uh oh and the last thing to mention is that I am using a 50 ohm termination at the oscilloscope um CU yeah running coax straight into a one Meg ohm impedant like input impedance oscilloscope tends to cause uh a mess because especially if you're measuring like very fast transients those will actually just end up reflecting uh a bunch and then then your measurement is basically useless so I do have a 50 ohm termination at the oscilloscope for this now then uh here you can see the card with the power Bridge removed right and the the experiment here was literally just like connect coax to card measure it remove the power Bridge measure it again see what the difference is so not exactly very complex so here you can see what it looks like when when I remove the power Bridge also here you can see that the power Bridge actually like goes over a bunch of PCI Lanes which uh kind of nice to have a very like obvious visual uh indication of what this is bridging across um though I will say like this style of like PCI Lanes coming into the GPU core like this is very typical of low-end Nvidia cards like I think even a lot of GTX 970s have the pcie coming in through through through this area and actually even on Modern cards this is very common like yeah it it is very very normal for the PCI Lanes to come into the core through through this area so it's honestly I mean I don't I don't know if it's that weird that the power Bridge has like that you don't see the power Bridges anymore from Asus but uh yeah like here like it visually makes a lot of sense right because it's like oh we've got pcie Lanes on the top layer so you can't run the VOR power plane there so you can just go over it with this great big copper plate but um yeah anyway uh also here's the the power Bridge um when I desoldered it so that's all it is it really is just a copper plate with some tabs bent bent off of it also if you're wondering how I desoldered it I do have a soldering iron that's just powerful enough to heat the entire plate up to leadfree soldering temperatures so that's what I did I I literally just uh I have like a 6.6 mm uh chisel tip so I just sat the soldering iron at the middle of the plate until the entire thing just came off um alternatively I could have used hot air but that's I don't like I didn't want to use hot air for this because hot air is very like like um inaccurate in terms of how much stuff you're heating up whereas with the the soldering iron I can just heat the copper plate directly um so yeah anyway also the soldering iron that I use I think has a peak power of50 watts uh and the reason I bring that up is if you used something like a 30 watt soldering iron I imagine it would either not achieve anything or it would actually get stuck to the copper plate one of the two so uh yeah helps having a soldering iron that has basically the same TDP as the entire GPU um anyway let's take a look at the measurements so here we have the measurements from a ryol HDO 1104 uh I am using the full 100 mahz bandwidth um on both of the channels channel three so the the pink one uh is that's from the vrm right so that's uh where is it um yeah so the the pink line is this probe right so that's the pink and the yellow is of course the core um so that's this probe that that one's yellow that one's pink um so anyway uh the first set of I mean actually if you just look at these um it should be pretty obvious which which uh which oscilloscope captures have the power Bridge installed and which ones don't um cuz there's like a pretty big gap between the yellow and the pink line on on some of these captures and not on the others uh as for what test I was running this is all fire strike Graphics test one um so this is at the end of the test and I did run two passes uh each time so we have 349 samples into the like averaged measurements uh and the reason I've done two passes is just to make sure that there isn't some like weird oneoff uh like undershoot event which yeah you can see that you know pass one we get a minimum like work worst case like the worst undershoot that I measured was 1.0 485 volts uh and then on the second pass it was 1.0 485 volts how very convenient uh so yeah the the the runs are very or well I shouldn't say very cuz this is just two runs but relatively reproducible and like good enough for my purposes here uh as far as I'm concerned so anyway um I guess we're going to start with the like average voltage differences um also you'll know notic that I did try my best to get the voltage offsets of like the vertical wait is it the vertical yeah the vertical offset of each of the channels matched up it's not exactly matched the pink channel is actually at 1.09 n rather than 1.1 but um uh close enough um so yeah plus we're just using the measurements anyway so it doesn't actually matter if visually they they are lined up perfectly but I thought it would be nice if we could actually just have them like overlay on top of each other pretty pretty closely anyway so um oh also I should mention that I didn't let the card get up to uh equilibrium temperature as in I didn't let the card sit there running fire strike for like an hour before taking the measurements mostly because I'm lazy but the reason I bring this up is because uh copper has more like the yeah the electrical resistance of copper increases with temperature so uh if you have a card you know and the PCB is at 20° C which it wouldn't be by the end of a fire strike run but when when you first start the test if the PCB is at say 20° uh and uh like that's going to change how much voltage drop there is in that power plane compared to if the PCB is at say 90° right I don't know where the equilibrium temperature for the power plane on this card is I also don't really care because I'm not going to be running the card on sustained loads but that would potentially increase the differences between the like no power Bridge versus with power Bridge measurements that I've taken uh because uh OB well the power bridge I would like would reduce the the power like it would lower the equilibrium temperature right because there' be lower power losses in the power plane so the power plane would end up at a lower temperature that lower temperature would mean the power plane resistance would be lower so you have like this like stacking effect if I actually ran the test for for longer but uh I to just a very quick like so how much of a difference does this this really make uh like so this is kind of like a worst case scenario for for the power Bridge cuz if I did let it heat up it would probably be more effective at reducing the power losses in the power plane than with my very quick test of I I ran fire strike once I ran fire strike twice uh and even in that like short so like there was a like I didn't let the card cool all the way back down to room temperature um but I didn't uh didn't bother to let the card get up to like equilibrium um so that that is worth noting um that yeah at like if if I ran the test longer we would potentially see bigger differences between the uh stock configuration and the configuration without the power bridge but anyway so let's just talk about the uh voltage drop um between the VOR vrm and the back of the GPU core and so at stock uh so with the power Bridge installed um at the end of the fire strike run the average voltage was like 1729 volts or uh 1731 volts which functionally the same thing as far as I'm concerned we'll call it 1.73 um and then the voltage at the vrm was 1 Point uh you know like the average voltage by the end of the Run was 1.1 1944 versus uh 1 and well second run was 1.9 48 so we'll call that like 1 Point 94 I guess right yeah we'll call it 1.94 something like that so basically between the uh basically so we'll go back to this photo so basically between our two measurement points here we have a voltage drop of about 22 M volts which is very low um yeah that is that is actually just very low um so that's neat um potentially we might have seen a vol bigger voltage drop if I measured from here um but I wanted to measure the effect of the power Bridge so that doesn't really like if I measured from here cuz the thing is like uh like I don't know what the internal cu the card does have an internal vcore power plane right so I don't know what the internal vcore power plane shape is for all I know it might do something like this to sort of you know like it might be something like this so that it avoids the PCI lanes that are coming in through here um so maybe this would actually not be that like the voltage over here wouldn't be that different from the voltage over there but I don't know I didn't you know I I guess I could have used a third channel to measure like the far end of the of the VOR vrm but you know I I am interested but not that interested so anyway so the with the power Bridge installed the card has about a 22 molt voltage drop across the uh like between the core and the uh vore vrm um and that is quite low um yeah from my like I've measured a few cards and I don't think I've measured anything that low before um so anyway or is it no yeah no like I shouldn't have got it's low it's pretty low off the top of my head it's pretty low I unfortunately don't usually measure this for every card cuz it's not really that important usually like yeah um but anyway so about 22 M volts in the stock configuration now with the power Bridge removed um the average voltage didn't really change it did actually drop a little bit um but it didn't like significantly change and the I'm guessing the reason why it slightly dropped is because the uh measurement point that I'm using um is not the well so the um so the voltage regulator on the card actually has a direct connection to a like voltage measurement point on the actual package of the GPU so in the like the pin out of the GPU core There's an actual pin for like core voltage measurement um and that's what the voltage regulator is regulating against it's not regulating against um wherever the hell my coax is hooked up right so the core like the core voltage regulator actually has no idea what the voltage over here is so I'm guessing what's going on is that uh the average voltage is a bit lower because the voltage regulator is regulating against the core and there's basically some voltage drop from the back of the card to the actual silicon on the front of the card right so the average voltage on the back of the card has uh dropped by a fraction of a molt so now I feel like maybe I'm I'm just wrong cuz that's so small that uh that could actually just be well it is see I'd like to say it's like oh it's just measurement tolerance but no we get that twice right that is consistently about a molt or two lower um so that might be a real effect um yeah having an oscilloscope that measures voltages to four digits is is wild to four decimal point actually yeah to four decimal points is um a bit much because this is like this is actually just like one Molt of difference um so it might be repeatable but I'm not 100% certain on that but if that is actually going on then it's probably because the voltage measurement point is actually on the like the voltage regulator regulates against the Silicon not against the back of the card and so you know now that it's a bit harder for uh Power to basically get from the V core vrm to the back of the core the back of the core is at a slightly lower voltage than it used to be um anyway um but yeah so we we get a very small reduction in average voltage um that honestly if like it's it's a molt so yeah um and then we get a really big increase in the average voltage at the output of the vrm which is what you expect when your power plane resistance you know suddenly drastically increases and so actually with the power Bridge so the so run one right we get 1.72 versus um 1.234 and run two we get 1.72 again uh against 1.2 35 um and that's about 62 63 molts so uh go back here in this configuration right going from here to here we have a voltage drop of 63 molts um which basically means let's say you had 100 amps going from the vrm to the GPU core this would produ that would be like three 6.3 watts of heat um that would be getting dissipated entirely in the power plane and that's kind of the the sort of the the like thing that yeah like this that 6.3 watts is entirely in the power plane of the card now because there's no power Bridge right whereas with the power Bridge right where I measured 22 molts of voltage drop uh from you know here to there um if we had 100 amps flowing uh to the core from the V cor vrm which I think in the stock configuration the card doesn't pull that much power but when you start overclocking you can you can get these cards to pull quite a bit though at the same time if the card is pulling more current then the voltage drop is obviously going to increase so you get the idea but this the 100 amps here is for illustrative purposes only um yeah though it's probably not that far off what a GTX 760 pulls cuz the cuz it runs at like 1.1 something volts right so 100 100 time 1.2 volts it well it doesn't even run at 1.2 Vols like 1.1 like we're getting like 1.17 on the core so that would be 100 actually at the vrm we're getting 1.2 is so we'll call that 120 watts uh then some kind of power losses in the vrm let's say 10 watts or something let's say 12 yeah let's say 12 so that would be 132 Watts add the memory to that and like I think GTX 760's are somewhere around 150 watts something like that so actually 100 amps across the across the power plane probably isn't that far off of what it is in reality um but anyway so in this scenario we'd get only 2.2 watts of power loss across the like uh Power delivery Network to the core right uh and the crazy thing is those 2.2 Watts like some of that power would be getting dissipated in the actual power Bridge itself right so that wouldn't be heat being produced in the PCB that would be heat being produced uh in this plate and this plate obviously like on probably run has a EAS like is more cooling friendly than the power plane itself where you could kind of a silly idea that you could do is like stick heat sinks onto the onto the power Bridge not that I which would probably produce measurable effects under like sustained High load scenarios um that you could actually measure a benefit from strapping heat sinks to the to the power Bridge um to make sure that the power bridge is as cool as possible but yeah so honestly like uh so in terms of like the Improvement going from like no power bridge to with power Bridge we're talking like saving about four watts in the in the power delivery to the core um which is something I guess um It's not nothing it is measurable but at the same time I can I can kind of see why Asus decided to stop doing power Bridges at some point because yeah this really isn't this isn't doing much um yeah especially and also if you consider that by the like cuz the last time Asus was still putting the the actual like pads for power Bridges onto their pcbs that was 10 Series right and 10 series gpus basically don't pull any power at all um so it kind of makes sense that by 10 series Asus just kind of gave up on this idea because well the current drw probably wasn't high enough to justify the uh extra expense not that I think this probably isn't very expensive but um yeah I can kind of understand why they ditched it um anyway now so it does improve the power losses in the power plane um also if I had let the no power Bridge setup run to like equilibrium temperature I think it would have like we would have seen an even higher voltage drop at some point right cuz like we're dissipating like three it's not literally three times but something like significantly more power is being dissipated in the in the like power plane of the or power planes cuz it's a multi-layer PCB there probably isn't a like just one V cor power plane maybe there is maybe it's like a four layer PCB I don't know how many layers this board has um but yeah so I don't know like so we're going from like you know like the the thing is like I don't have measurements for the entire power plane so it's not really correct to say that oh the entire like from the vrm to the core it's definitely just 2.2 Watts because I don't know maybe there's going to be some power like potentially somewhat a higher voltage drop from like over here to here so there's going to be like a gradient of voltages across the entire like V core to V like core to vrm connection right so the fact that like across this section of it we're seeing 22 molts doesn't automatically mean that like this isn't maybe another 10 m higher and I also don't know how much current is flowing through each part of it so it gets complicated but um this is a pretty substantial increase in voltage drop um so we we'll say three times based on my measurements but my measurements are not all encompassing so there's some error in that probably but anyway um so you know uh yeah if we ran this to like equilibrium right this is now putting out like three times more heat and all of that heat is now concentrated in the PCB itself right because the power Bridge isn't there anymore so I'm thinking like yeah if if I ran the uh no power Bridge configuration for extended periods of time as in if I just let it get up to like you know equilibrium temperature um it would have potentially had an even bigger voltage drop across it uh and actually one of the reasons I decided not to try to run it to equilibrium was I wasn't enti L certain if the PCB of this card is actually design like designed to require the power Bridge cuz theoretically you could have a PCB where the internal power plane is so thin that if you ran it for like extended periods of time like the it would just keep increasing in operating temperature indefinitely like the PCB would never be able which is probably pretty unlikely but like you could have the internal power plane end up at some Insanity temperature like over 100 degrees or something and actually there's uh there was a z590 or z490 motherboard where I measured basically that exact situation where the vrm temperature was fine and the temperature of the PCB a couple millimeters away from the CPU socket was Skyhigh because the board had was like a six-star PCB and all of the current trying to squeeze into the CPU socket was basically just heating up the the power plane right before the CPU socket which is where it gets narrowest um also I guess the pins in the socket probably produce some some heat as well so that was like so you had this really weird situation where it's like yeah the actual like mosfets this the Active Components you know the switching components were running cooler than the board like way out in front of them um and so I didn't really want to test if like this PCB can actually like cuz I don't know maybe it it probably does it's probably actually like I I would guess that this is probably very similar to a reference GTX 760 actually no cuz this is an Asus card and Asus card Asus likes to do rather heavily customized pcbs basically as much as possible like a lot of other manufacturers will basically take a reference Nvidia card and then just slightly Shuffle some things around but Asus likes to make cards that look very distinctly different from ref from reference and video video cards so yeah I don't know I didn't really want to um yeah I I didn't really want to see if like the PCB ends up at some insanely high temperature if I just run it with no power bridge for a long time um also it would have just taken forever and I like you know I I am I do wonder about the power Bridge Effectiveness but I don't wonder that much um so anyway the other thing I wondered about is does this have any impact on voltage regulation and the answer is no no it does not or at least uh it doesn't have any impact on voltage regulation that I care about because if we look at the max measurements actually nah it doesn't cuz like technically like yeah you might be like oh the average overshoot was 1.29 volts versus uh 1 uh 30 one volts and it's like yeah so that's like 2 molts um so that's not a difference and then and that's across both runs right like here we have a Max of actually this is no that's absolute Max here we have a 1.3 versus a oh look actually here the average overshoot is actually slightly lower than in the previous test so yeah the maximum voltage did not change uh the overshoot is the same regardless of if the power bridge is there or if it isn't though I will say one thing I find interesting about the card when the power bridge is installed is that we get overshoot spikes behind the core that are taller than the peak voltage at the vrm um yeah which is super weird because I would have always figured that you'd get the highest voltage directly at the vrm itself rather than at the core cuz as far as I know the overshoot is caused by the excess energy stored in the inductors having nowhere to go but by the looks of things here we seem to have some kind of like extra like there's there's some like extra kick from like the power plane towards the core um or something because yeah like we get more Peak voltage directly behind the core than we do uh at the vrm which doesn't really make uh like sense to me in the within the like within my understanding of like the overshoot comes from the inductors and it's like but that like if the overshoot comes from the inductors why are we seeing more overshoot all the way at the actual core than at the vrm which is like right next to the like than at the than at the inductors right so um I guess there might be some kind of like uh like the inductance of the power plane itself is actually causing a even higher overshoot directly behind the GPU core um as you're basically cuz when the the core transitions from high load to low load um all of that current that was flowing through the power plane well it's still kind of wants to keep going probably um and that's piled on top of the current that's in the inductor and that would probably explain why we're getting you know spikes on at the core that are actually taller than spikes at the vrm uh which uh yeah that's actually like I did not expect that um so anyway but the actual like P like the overshoot level in and of itself hasn't really changed um right though it is worth noticing that when I remove the power Bridge the overshoot at the vrm is actually cons like yeah see pink is now above yellow um actually pink is consistently above yellow um so that's kind of interesting is like yeah when When there's less resistance in the power plane um we actually end up with the spikes at the core being taller than the spikes at the vrm so that's an interesting effect now undershoot which is the part that I actually care about because this right here is what causes the GPU core to crash when you're running a benchmark at very high clocks or just generally if you're overclocking you want as little overshoot as possible right so uh did the unders I mean little under you want as little undershoot as possible the overshoot is kind like it's obviously not great if the overshoot is super high but it doesn't cause blue screens so I don't particularly care about it um anyway the undershoot on the other hand uh so run one uh with the bridge we Min minimum is like 1.048596 without the bridge we're looking at 1.06 6 so slightly worse and then worst case worst yeah the worst undershoot that I measured was 1.04 85 so exactly the same um and actually 1.06 75 and 1.66 uh yeah 1.66 is only 1.5 molts apart so functionally the same and if we look over here oh look 1.66 1.05 27 so this is actually slightly better than uh either of the the uh uh runs with the power Bridge installed and um yeah though the average here is actually better than the average for either of these um by 2 molts or actually no that's like three molts so uh does the power Bridge improve voltage regulation no no it doesn't um it it just doesn't whatsoever so now I don't have to feel bad about all of my DIY power Bridges basically behaving the same cuz that's been my experience like I've had like there's been like I've done this with multiple cards as like you know like just trying to do anything to get the voltage regulation to be better and so one of the things that I've messed with was just like adding uh like really thick copper wires or copper plates going all the way from the like V core vrm to the core and very consistently the only effect that I've ever managed to measure is that it reduces the power losses in the power plane I've never seen it improve the voltage regulation and well evidently it's not supposed to because even this like you know Asus implementation uh which probably has significantly more engineering behind it or at least you'd hope it has more engineering behind it than my DIY attempts of just slap some copper somewhere uh yeah this doesn't do anything for voltage regulation like it it just doesn't the voltage regulation hasn't changed whatsoever like the only reason we can see a difference is because the HDO 1104 is a 10bit oscilloscope and so it measures to just like four decimals which which is like differences of less than 5 molts are not relevant um and even 5 molts I'd say is like okay that Pro like that's probably more dependent on like measurement methodology and like how many runs you've done than it is on like than it is an actual noteworthy effect so yeah this really like power Bridges evidently do not do anything for voltage regulation which is quite interesting though actually there is an exception to that uh and that exception is if you're e- powering a GPU so if you replace the vrm with a different vrm and the reason that benef that actually sees a difference in voltage regulation if you add a power bridge is not because of some inherent property of the power bridge but it's because when you're doing external P like external vrms um or when you're replacing the vrm with an E power the E power if you don't reference it to the GPU core uh the E power regulates against its own output so if there's a lot of resistance between the output of the E power and the back of the GPU core and you know and you have to consider that like so there's the resistance of the uh you have the resistance from the output to like your like the connection you made and then you have also the resistance of the the existing power plane in the board that you just owed right so those two resistances get added up and that's going to like mean that there's going to be like like the E power is regulating the voltage against its own output so the voltage at the core is going to basically be offset by that resistance depending on the current flowing through the resistance between the E power and the core so in that scenario adding a power bridge is going to change the resistance which is going to change the uh like the volt the voltage regulation at the core because there's like less resistance between where the power E power is regulating against and the core um but if you just reference the E power to the actual core uh you basically end up in this same situation that you end up with a stock vrm because the V stock vrm of the GPU regulates against the core which is why it doesn't really care if the you know power plane has uh like three times more resistance in it um it just kind of compensates for that because the vrm right like like the voltage regulator doesn't yeah like it doesn't know what the voltage over here is it doesn't it doesn't really care it just knows I'm supposed to reach um I think for this card it was like 1.15 or something volts right this is my target voltage I will increase the pwm signal at the inductors until I mean I'll increase the pwm signal until this voltage is achieved right so how much resistance is between these two points kind of doesn't really matter to the voltage regulator now there is like if the resistance is high enough then at some point it does actually become a problem but not inherent like small amounts of resistance is just like yeah the voltage regulator actually just doesn't care um so yeah that's that's why um the the power Bridge really just doesn't do anything for voltage regulation it just reduces the power losses in the power plane so uh nice to have that uh measured I guess and now uh I also took some like zoomed in uh captures cuz obviously all of these captures I I forgot to mention it but these are at 5 milliseconds pred division right that's my preferred way of measuring things so that I get a lot of spikes onto the screen at the same time um but anyway here we have some zoomed in zoomed in captures and yeah like this right like here you can very clearly see that like we have a load release overshoot and at the GPU core it is like a lot higher than it is at the vrm which is super interesting so I'm guessing like this extra voltage is like coming from the inductance of the power plane or something cuz if it like if it was coming from the inductors you would expect the pink line to be taller right or at least I'd expect the pink line to be taller um anyway uh oh you might also notice that the the pink line for whatever reason is like thicker than the yellow line and I think the reason for that is that if you look at actually no you can't look at it but the the way the vrm on this card is designed uh there is basically no multi-layer ceramic capacitors directly at the output of the vrm like all of the multi-layer ceramic capacitors on this card are uh like all the V core multi-layer ceramic capacitors I should say are here um right here there's like yeah there's nothing directly at the out of the vrm there's like some over here that are for VC core but the bulk of the like high frequency filtering is directly behind the core and then uh right at the output you just have some through hole aluminum polymers which have a ton of equivalent series inductance so they really don't filter out like high frequency switching noise uh basically at all which is probably why we see the like Peak to PE like the like in this like relatively steady state looking region right over here uh why this is like fatter than this um is probably because there's no multi-layer Ceramics directly at the vrm um so yeah but I do find it really interesting that like the overshoot is actually worse at the core than it is at the vrm cuz I would have never expected that um and I I guess like and like my guess for why that is is just the inductance of the power plane itself cuz like I don't like ad like I'm not an engineer so I could be completely wrong about this but I'm not aware of anything else that would be causing this um so yeah I'm I'm going to go with yeah like the power plane is doing that because I don't know where else that extra voltage is coming from um anyway uh with the power Bridge removed uh well there are some differences in the spike spikes again but um well now you know the pink spikes are taller than the yellow spikes though you can also see a slight offset in when when the spikes happen which is uh interesting right like where the peak is like this peak is obviously earlier than this peak um but that could also just be down to the difference in capacitor configuration um all right so yeah these are these are kind of interesting to look at but not really useful in terms of judging how effective the well how effect like like like these are cool to look at but there's not like a lot of like oh this is what like I like I'm not I'm not g like I'm not seeing any like useful to me information out of this is what I'm trying to say there uh which might be just because I'm too stupid to understand what I'm looking at here um but yeah like from from what I like like I I find this interesting but there's not really anything I can do about that it's just like well I'm just going to throw more capacitors directly behind the core anyway so like what difference does it make um right cuz that that's ultimately like how how do you make that Spike smaller more capacitors so that hasn't changed um and then and this measurement right here is just like well we're seeing kind of the same thing except now the pink line is offset upwards by by a certain amount so yeah anyway um that's it for this video hopefully you found it somewhat interesting if not particularly useful um but yeah this this is something I've been wondering about for a very very long time is like did this actually do anything and the answer is I mean technically it would give you slightly more performance if you were riding the power limit right cuz you're losing less power in the power plane but like it's like four Watts it's it's like not like it's you know like most benchmarks aren't like yeah I'd say most benchmarks literally aren't consistent enough to measure the performance impact that this would have in a power limited scenario unless we were well no the thing is if you have a really low power limit then the amount of current flowing through the board is going to be really low and the effectiveness of this is going to get reduced right so this is most effective when you have a lot of current flowing um and yet somehow you're and you're like riding the power limit um that would be when this would actually have like a performance impact from an overclocking perspective uh like from like should you do this mod to your GPU perspective is is pointless anyway this video is now way longer than it ever really should have been um but yeah hopefully you found it interesting um so that's it thank you for watching like share subscribe leave any comments questions suggestions down in the comment section below and if you'd like to support what I do here with actually hardcore overclocking I have a patreon there's a link to that down in the description below there's also the hocc tepring store where you can pick up shirts hoodies posters you know the usual YouTuber merch and I've also got a band camp there's a link to that down in the description below as well and that's it for the video so thank you for watching and goodbye

Channel: Actually Hardcore Overclocking

Views: 22,474

Keywords: Overclocking, PCbuilding, Buildzoid, AHOC, Actually, Hardcore, Hardware, OC

Id: JDw1Ul-nBhc

Length: 48min 36sec (2916 seconds)

Published: Sat Jan 06 2024