Gadgetory

hey guys build Zoid here from actually hardcore overclocking and today we're gonna be taking a look at the Radeon seven refunds PCB from AMD it is it's it's very interesting it's one of the first times I've seen a PCB and gone like I mean there's a few other cases where this has happened but this is like one of the more recent ones where I've gone like I have no idea where anything is because of how complicated this thing is well how many different voltage rails are on this thing well different regulators and rails you know you you need different regulators for your different rails so before that this video is brought to you by Thermaltake score p3 case the core p3 is one of the most unique cases on the market it can serve as an open-air standing chassis a test bench in vertical or horizontal orientation or as a wall-mounted showcase pc the core p3 now comes with a 5 millimeter thick tempered glass panel for its slide but keeps the front top and back open for air the core p3 is versatility as a display piece test bench or standard desktop is reinforced by its price of roughly 110 dollars on amazon you can learn more at the link in the description below anyway let's get right into it and so starting off with the largest and most important - erm on the card the V Corps which in typical AMD file not really so you know and in the past with like the Vega 64 AMD set a you know AMD had like the most powerful consumer the are like consumer GPU vrm right it was like on par with extreme overclocking cards and this vrm could have been that like in fact this vrm could have been more powerful than what you get on a 20 80 TI but you know that and that basically means it isn't so this is the the lovely V coral in this and it is still very very powerful and still very very efficient it's just not as ridiculous like I'm sad that it's not as ridiculous as it could have been okay I'll be honest with you I really wish that this was still this was still full populated and also we can kind of see up here you know in the PCV normally there'd be a little switch thing with a nub that sticks out in that little slot that they have cut so yeah that is where the bios which would have been if AMD had bothered to include it but anyway back to the vrm which the reasoning why these phases are missing is actually really really simple in typical AMD fashion AMD has opted for some very nice and very expensive power power company well they're not MOSFETs their power stages there's MOSFETs in them but you know that goes it's significant enough difference that I wouldn't call these MOSFETs so they've opted for some very very nice power stages which historically they've really just gone for like really nice MOSFETs and then well actually no they did use a lot of well yeah power stages as well so anyway AMD likes their high-end well like you know high-end power components so they've gone for TDA International rectifier or Infineon at this point TD a two one four seven two's these are 70 M smart power stages so at this point you might be like oh so it's like the same thing as the FD MF 31 70s that you find on like in video reference cards actually most of the r-tx 20 series well no these are about twice as expensive as an FD MF 31 70 these are about $4 apiece and that's even if you're buying 50 thousand of them so basically by not including these two phase like these four phases right here AMD is probably saving around $20 on the vrv core vrm which makes sense because I mean this card is supposed to hit a price point of 750 dollars retail and that right there is really really expensive too so you know it's just like well you got to start saving cost somewhere right and and this isn't gonna get cheaper that that has a relatively fixed manufacturing cost and I'm not even sure if it's possible to manufacture HBM GP without all of the HBM populated because if you look at like the Titan me right the Titan V only has three active HBM stacks but there's four of them on the chip so yeah that's that's one of the reasons why I'm wondering if AMD even has a choice about you know going for Less HBM with this thing because it's just like well it might cost more to assemble than it costs to buy the actual stacks and so the issue is that you can only assemble it as a four stack configuration not as a three or two anyway and a two would probably have bandwidth issues but anyway so you know trying to save some money on the vrm and then getting the really really high in power stages now what makes these kind of special compared to say the FDM f31 70 is that because these are you know 70 amps more power stages so that means they have all of the usual features like over current protection over temperature protection built-in current monitoring built-in temperature monitoring you know all kinds of cool features like that which makes them a smart power stage but because these are also made by international rectifier and extremely expensive they also come with features like a tolerant like they can run at switching frequencies as high as one point 5 mega Hertz whereas the FD MF 31 70s which and video likes to use those go only up to one mega Hertz not that that's a particularly useful difference because most of the time you will never run your voltage regulator at that kind of frequency it ruins your efficiency but these also include something called body braking mode which is a standard feature for international rectifier power stages which essentially is used for load transients on load release which basically minimizes your voltage overshoot on load release which is a really really cool feature but yeah so that's something the FT MF 31 seventies just straight up don't have so essentially what that allows these the the body braking what that does with these power stages is essentially if you go from a high current draw to a low current draw condition you can stop using the low side MOSFET on these and that forces all of the currents to give flow through the body diode and the body diode has a bunch of voltage drop so you can very quick we burn off the energy stored in your inductor without getting the usual without getting as high of a voltage spike as you normally get when when you go from a high current draw scenario to a low current draw scenario so that's a pretty cool feature that the these have and that also adds to their cost and they're also ever so slightly more efficient than the FDM f31 70s for a comparison because both are expect for like Intel CPUs at 1.8 volts the TDA 1.8 volts output the TDA to 147 - pushing 40 amps only dissipates about 5 watts with 600 kilohertz switching frequency whereas the FDM f31 70 would dissipate about 5.6 watts with 500 kilohertz switching frequency and then if you go all the way up to 60 amps output on these on a single one of these these would produce only about 10 watts of heat whereas the FDM f31 70s produce about 12 watts of heat so these are slightly more efficient they have slightly higher switching frequency capabilities and they have body braking mode and that's why they're twice the price so yeah that that's kind of you know am V for you they really like their high end power components unfortunately they can't actually have that many of them on this card I'm like they should do like a special extreme overclocking edition of this card with with a bios switch and voltage controls and no power limit and and no boost algorithm that would be great and fully populate the VR I'm in charge a thousand bucks and I'll buy it I'd actually probably buy that anyway so you know that's our power stage here the the phase count for the vcore PRM is one two three four five six seven eight nine ten phases and this is an actual 10 phase configuration it's not a true 10 phase because AMD is still going with their ever-so-popular well with their favorite voltage regulator I'm pretty sure that is built just for the them the IR three five two one seven which we've already seen on like Vega 64 and Vega frontier Edition this thing goes up to eight phases there's no datasheet on for this thing at all but one would assume that this being a 3 5 2 200 series chip it'll go up to 2 megahertz switching frequency if you so desire you know you wouldn't actually do that because it'll ruin your vrm efficiency but you can and yeah it goes up to 8 phases and here it's actually running in a 5 plus 1 phase configuration if the if the vrm was fully populated it would have been running in a 7 plus 1 but as it is it's only configured for a 5 plus 1 and then it is doubled into the 10 phase by these chips right around here so and those are of course because this is an international rectifier the RM design these are of course ir35 99s which are like the their main feature is that they can go quadruple or they can run as a quadruple ER so you can go from like 5 phases to 20 with these but here they are only used in doubling mode and the downsides of them being quadruple errs is that they do the most basic of doubling possible where they essentially take the PWM signal and just put it a split it between the 2 phases which basically means there's no current balancing and as far as the voltage controllers can concern this is a five phase but the 3599 does have the advantage that like you still get the reduced output ripple because your phases are all interleaving right so you do have 10 different PWM signals on the actual power stages themselves so you get the reduced output ripple you get the reduced input Ripple but you just don't get any like this since you like current balancing your vrm efficiency isn't quite as high as it could be if it was perfectly current balanced and all that kind of thing but uh you know that because as it is the the like the ir35 to 1/7 is doing all of the current balancing and it does it in groups of two so it'll like take this block right here which is two phases and it'll current block balance that against this against this and against this it will not actually current balance the two phases in each of the like the this individual phases within those blocks will not be current balanced at all but uh yeah still like as far as well like VRMs go that this is definitely not a design it's just like it's the limitations of going with international rectifier and the only company doing like fully current balanced voltage current balanced doublers are inter cell and they well they're really expensive to run so they're they're like even more expensive than than the than international rectifier parts because you'd be like Intersil I'm not sure interests I'll even makes a 70 amps smart power well they do make a 60 but I'm not sure they make a 70 amps more power stage yeah and the the doublers that they like the current balanced doublers from intersil i think are like 1.5 dollars a piece in bulk at least last time I checked them so you know yeah not really a great option in terms of maintaining low you know keeping your costs low know that International rectifier is much better but it's it's not quite as bad because at least the doublers are like half a dollar each instead of one and a half anyway so that that's a recurve erm another interesting thing to note is that we almost have a full you you know AMD was the first to come out with the l-shaped erm and video kinda in my opinion one-up them with the double I configuration where you have one row of phases on one side of the via via of the core and then the other row of phases on the other side of the core and essentially what that means is when you are pushing power to the GPU core what happens with a standard sort of erm layout where you just have one row of phases on one side of the core as you push current into the core you get a voltage drop across the power plane and so essentially this side of the core has less core voltage than this side and you can actually check that on GPUs while they're running if you measure the bypass capacitors on the back of the GPU core 4v core the ones furthest away from the vrm will tend to have a couple millivolts less voltage than the ones nearer sometimes as much as like tens of millions so yeah and on like a Vega with the l-shaped vrm you could actually measure like in this corner you'd be measuring like 1.2 and if you measure it in this corner you'd measure 1.17 right so you'd have a difference like that so this vrm I mean it's normal full you if they went for the full you then you'd have like this point right here that would be your lowest voltage but with this configuration that point will very likely be actually we should go like that that point will very likely be right here again but it's um it'll be interesting to see how much of a difference it would be from like here right like across there and what what kind of voltage difference you'd have it should be very low because this is getting very close to almost like a full surround I wonder if we'll ever see an O but anyway so that's kind of interesting that Amity has the vrm laid out like that and that does mean that you get a bit more uniform voltage on the GPU core which in theory means you'd have a bit better clock margin for any given voltage because ultimately your lowest voltage is also gonna affect like your lowest voltage going into the core is gonna obviously limit your maximum frequency so anyway and it does mean you also have a bit better efficiency because there's a lot less voltage drop across the power plane but that's really a minor thing anyway so that's the the vcore vrm sort of you know not mention not taking a look at the power efficiency so let's take a look at the power efficiency the TDA 200 472 s make this VRM ridiculously power efficient and for my convenience they're also SPECT at 1.2 volts so 1.2 volts out output 400 kilohertz switching frequency and a drive voltage of 5 volts which is standard which also means we have a 5 volts regulator somewhere on the card I'm not sure where there's because we have like regulators up there up here up here this thing that thing and we'll get to those eventually but those like then we have another regulator down here these are like the minor ones there's also like I assume this is a regulator but I'm that might be 5 volts fortunately like there's rails everywhere on this card so and I'm not sure which one specifically as 5 volts but there is a 5 volts RL somewhere on the rail because these I mean some on the card because these run off the five volts being a smart power stage that's pretty standard for them now then they are expected at 1.2 volts 400 kilohertz from my convenience and for the card does not run on 1.2 volts right it's seven nanometer it runs on a lot less voltage than that anyway so 200 amps output at 1.2 volts you're gonna be looking at about 15 watts of heat dissipation on this vrm which is crazy efficient like that is really really really great efficiency right there it's just about 94% efficiency actually it's not exactly 15 okay because if you actually go you know if you go without 15 watts figure at 1.2 volts 200 amps you're gonna get like 94.1 percent efficiency yeah that doesn't work out it's like fifteen point five or something because it's just barely under 94% efficient this vrm is at 1.2 volts output 200 amps so yeah but about 15 watts so you know like you don't you shouldn't need a heatsink wholeness you really shouldn't if you're only pushing 200 amps now going up to 300 amps output this vrm is gonna still only produce about 20 watts of heat which is again like there's a crazy efficiency like these td8 to one for actually most of these 70 amps more power stages are really good in terms of their efficiency but the TDA 2 1 4 7 2's are a bit better than most and then going up actually wait now I misread my notes that's only 250 Watts 258 and still 20 Watts well yeah that's still a really great efficiency right there going up further to 300 amps output you're gonna be looking at about 25 watts of efficient like heat dissipation which again like I mean I don't need to tell I like the efficiency is just out like up there and I'm really sad that they didn't go for a 14 face like the seriously it would have been insane how efficient this vrm would have ended up being but anyway going from 300 amps to 400 amps so I think ultimately like stock the card runs in that 200 to 250 amp range 300 amps up you're gonna be looking at overclocking and the higher the current you're going you know looking at like extreme cooling methods to hit the higher currency if the card even scales to that kind of current level anyway about 400 amps you're gonna be looking at about only 40 watts of heat and 500 amps you're gonna be looking at about 60 watts of heat so at this point the the vrm starts actually you know like we got a pretty major increase in heat dissipation going from 400 amps to 500 amps and then 600 actually my notes don't have 600 amps which is also like unfortunately the datasheet ends at 60 amps anyway so we couldn't really go past that and I doubt anybody's gonna run these on ln2 anyway so yeah and if they do I'm not sure if they'll even scale to 500 amps all right we still like well like we still have to see about that but still for normal usage this vrm is really really efficient so that's great for extreme overclocking who knows like it really depends if the card even needs that much current win on ln2 and I don't know that we'll find out because I'm not buying one of these I don't plan to and I it's not competitive with a twenty atti so I really doubt anybody's gonna seriously bench one of these so yeah that kind of sucks because this is a really nice VRM and it would be even nicer if it had all fourteen phases but anyway moving on to the other main prm's we're not gonna talk about the other like circled regulators because quite frankly I don't know which one of them is which and there's a bunch of them that have to be on the card so we have like 1.8 volts for the HBM there's gonna be a pecs rail so well there's a PCIe rail there's a display drive rail PCIe which usually those two are combined so PCI in display but uh yeah oh yeah there's also the 5 volts rail and unfortunately there's like more regulators on this card then I know like common AMD voltage rails for so evidently AMD has added a couple new ones but I don't know what they are so anyway moving on to the ones that I do know what they are we have down here HBM power which is a two-phase so that's our HBM vrm right there one and two and then down here we have VDD CI and VDD CI is normally on like AMD's gddr5 based cards is normally huge okay like if you look at like 7-9 70s or 7-9 50s there are multiple 7-9 70s and 7-9 50s with like two phase VD DCI because the VD DCI rail used to be so power-hungry and that's the the memory controller in fact VD DC I used to normally pull more current than the memory itself did so yeah that's kind of an interesting like difference between the older cards and the HBM cards where now with the VD DCI rail is this cute little adorable thing right here which is a single phase and that is a eye are three five four zero one power stage and that's a 20 amp part so like that this thing you know this is a really low current rail where previously this thing went like that like oh no seven nine seventy this wood straight up blow up because if I remember correctly seven nine 70s I had stock average like 15 amps on that rail so yeah interesting out of the HBM memory controller needs so much less power now that the card has four HBM chips the the HBM vrm itself has gotten a quite an upgrade right but so the HBM erm is on the again on the TDA two one four seven twos and that's going to be controlled by this chip right here which is another three five two one seven and then we get this VR i'm over here which is the SOC regulator so yeah Vega has a system on a chip portion and that's another two phase and that's also hanging off of this three five to one seven down here the VDD CI I'm assuming is running off of the three five to one seven on the back because this one is like that one's maxed out at seven plus one right and so this one straight up couldn't run a set like it can't do a seven plus two right which you need that plus two for the HBM and the sock rails so this three five two one seven here is running in two plus two configuration for HB m and SOC both of those are again on more TDA to on for seven to s which is probably ridiculous overkill for both of them because the HBM generally doesn't pull that much power I mean HB m the HBM power on a vega 64 was like a vega 64 has two HB m stacks or HB m two stacks and that one like quite frankly that vrm was a single-phase and it wasn't good enough for more than maybe if i remember correctly i was a good for like 30 amps max or something so the new HB m vr m is actually kind of overkill and kind of an odd decision in my opinion i do wonder why i'm d didn't opt for like a week or HB m vr m and then you know spend more money on the vcore like i'd prefer that but uh anyway here they decided that you know they're gonna go for two like a really really powerful HB m erm i guess it might be for the 32 gig of re into the card because logically the 32 gig HB m it has more like there's more memory chips in those stacks so they would pull more power so i guess that's what why we have the the two-phase HB m now whereas previously it was like an anemic single-phase so that's that's that's an interesting difference but uh and finally we have the SOC rail which is the same as the HP m so yeah that there is the Radeon 7 PC B it's uh you know like it is actually more efficient than a reference vegas 64 like even though it has less phases the individual phases are much better than what you got on a Radeon Vegas it like on the Vega 64's just because the TDA 2 1 4 7 s are absolutely stupid in terms of their efficiency I think these are awesome and super expensive as a side effect but uh the yeah so this is ultimately like even though it's a ten phase it's actually more efficient than the twelve phase you got on a Vega sixty-four also because I think thermally this vrm works out to be less dense it should actually be less hard to cool without a heatsink so that's kind of fun and then for HB m we've gotten a huge upgrade on the HB m vr m VD d c-- eyes got a bit of an upgrade as well previously it was a just a minor like it was basically that that used to be VDD CI on a Leica Vega and then SOC has like well I don't remember Vega I'm not sure Vega had an SOC rail I don't think it did so that one's new and I'm not sure how much current it needs so that's kind of that yeah but that's the Radeon seven mic it is very complicated right and the layouts rather unique and overall I'm a fan of it but then then again it's like when have I not been a fan of an AMD reference design so yeah that's it so thanks for watching like share subscribe leave any comments questions suggestions down in the comment section below and if you'd like to support gamers Nexus there's a store gamers Nexus net where you can pick up things like the mod mat you can see in the background shirts I think glasses mugs and that kind of thing and there's also the gamers Nexus patreon and also if you'd like to check out more content from me there's a I have a channel called actually hardcore overclocking where I do more PCB breakdowns and other overclocking related content so yeah if you'd like to check that out that would be cool thanks for watching and good bye