Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

Ask GN 98: What is FinFET? How to Read VRM Phases?

2018-08-27
hey everyone welcome back to another ask Jian episode we're still working on finalizing the set here we're actually about to rotate and go back to a 45 degree angle shot putting a shelf over here you'll see more on that soon but as always leave your questions in the comment section below for next episode filming three today one is going up for the patreon bonus episode that's patreon.com slash gamers Nexus to get access to that and then two here on the main channel so let's get into the questions for this week before that this video is brought to you by it the NZXT e series power supplies the e series PS use our high end power supplies with real-time digital voltage and temperature monitoring / rail wattage measurements and data logging functionality for power usage the E series PSU is also come with a 10 year warranty all the way down to the 500 watt unit and they run fully modular and with silent fan operating modes learn more at the link below first a quick note a lot of you asked about when we'd be getting these back in stock the G n cobalt blue beer glass with a gold trim around the outside top of it so they're back in stock you've been asking for a couple weeks now thank you for your patience the first round was kind of a trial we sold out in like 14 hours or something it's pretty crazy but they're back in now so you gotta store it on cara's access Nets pick up one of these if you like them we should have enough for at least a little while and we'll see how the demand is but they've been going pretty quickly so far but now they'll probably go even faster either way they're back in mod mats we still have a good couple left until the we'll hit a pause before the next production run so if you get that order in soon you'll get that as well anyway first question is actually a very good one and it is from Nick Kirin II Nick said hey Steven your recent interview with David Cantor you guys talked about how at one point the foundries added FinFETs to their process I was wondering what exactly is a FinFET and how is it beneficial to CPUs and GPUs so this question is one that I answered back and asked at number26 I had to go look it up quite old at this point honestly my answer I didn't rewatch it but it probably wasn't quite as detailed as as it can be today so we'll revisit this one this is something where David cancer would be excellent as a source but I'm going to put in my version of this first get you all the basics and then they will revisit it if there's enough interest for the future so FinFETs first of all this is a thin field effect transistor which can be thought of as 3d rather than planar or a flat plane the big shift was from 28 nanometer planar process nodes to 16 nanometer FinFET for things like Pascal for instance which is largely where Nvidia's power savings came from infinite process a silicon fin protrudes upwards and it acts as an electrode FinFET processes reduce power leakage by using a three-dimensional design wherein a fin protrudes to form the drain and the source with a gate circling around the transistors fins you can think of it like a literal thin as an ax dorsal fin or something like that like a shark fin that sort of helps to guide the flow of power and reduce leakage FinFETs therefore also increased per core performance per watt and moving to fin vet lowered voltage requirements as an example to 0.8 volts at the lower end rather than the previous 1.0 volts at the low end for the previous process they're still using double patterning lithography for FinFETs as we don't yet have extreme ultraviolet functional in any meaningful capacity but the gate design is significantly reducing power leakage and voltage requirements FinFETs use it more conducting channels than planar transistors leading to reduced switching voltage as mentioned in the 0.8 volt low end for Pascal GPUs earlier up to 1/2 the power requirement as well by the way with some FinFETs and reduced leakage overall so if in FETs are critical for the modern moves and in the most recent GPUs not counting what's obviously coming up at this point so not counting that stuff that we don't have eyesight to do at this point the previous GPU switches from 28 and then downward a lot of the power saving gains that was from FinFET a lot of it was just from process changes in general a lot of it was from architectural optimization with Pascal or with Vega there were obviously some changes versus previous architectures Polaris and PG so some of its architectural but a lot of it does come to the process technology so FinFETs are you can just the very simple version of it is going to fin fat you're coming from something else the other thing you're coming from is planar so you came from a planar transistor which is it's literally a plane it's effectively two-dimensional and with thin fat it's more or less three-dimensional and there are some images we should have shown by now to illustrate that so you can kind of think of the fin as almost being like just think of it like a bucket it kind of helps reduce leakage by acting like a bucket and and that's a large part of the power optimization has reduced leakage reduce voltage requirements stuff like that so that's a big thing with nymphets there's more here but that's that's kind of where I'm gonna stop at my current knowledge shuffle anyway so that's hopefully that helps you out next question is from TM he says can you not answer questions that are easily Google and verified within five to ten minutes of reading results in forums like the watt per meter Kelvin 1gb if they don't phase passive cooling heating the room blah blah blah that's all line is tech tip stuff all right so first thing mr. TM why don't you try posting one if you have such good questions that are better than everyone else's that no one has ever come up with an answer to and it's not on Google anywhere post your question that would be more useful secondly so a big thing here these videos the ask G ends it's not supposed to contain like groundbreaking revolutionary information that you could never find anywhere ever before that's not the point the point is you're getting information from a source that you already watch for other content or I get it from someone who is a verified expert like David cancer in the previous the ten nanometers discussion and that's where hopefully you get some value I mean the you talk about the watt per meter Kelvin one thermal conductivity so the whole point of that question it wasn't what is watts per meter Kelvin because the answer is thermal conductivity although some people seem to think that it was what does that mean so okay we know we know that their own conductivity watts per Kelvin same thing but what does it actually mean and so that's where I want spoke to ESG and got a great answer from VSG who is a thermodynamics expert and was able to answer it hopefully with a bit of fine tuning for our industry so that's the thing with these questions sure you can go google thermal conductivity and eventually you get to the answer you want or you come to our content where we're very focused on the specific application of computer hardware and boil it down to the important components and talk about those or maybe throw in some interesting side notes and things like that so yeah not really the point there is very little I could answer that you can't find on Google because there's a lot of stuff on Google so sorry to disappoint you leave your groundbreaking question below and we'll talk about it next time WG what is it the silicon lottery why does it exist how does it impact consumers well WG unfortunately I'm not allowed to answer this question because you could google it apparently from a previous expert so the the silicon lottery no no either there was a great comment though about does he look like a plastic surgeon I love that comment thank you the silicon lottery is so what is it why does it exist how does it impact consumers is the question it's the main immediate thing here with the silicon lottery is that you have to know kind of where this stuff comes from CPUs and GPUs they're not created equal the spec might be the same but the actual quality of the silicon you might hear the term ASIC quality with GPUs that refers basically to this and the quality of the silicon can vary from individual part to individual part despite being the same skew ie 8700 K Vega 56 whatever and it's true for memory too by the way that's why you have some memory that performs better than others despite being the same spec when you start overclocking so silicon madhuri then ultimately the cpu that you have or GPU it is diced from a wafer so wafer is a big circular disk that's those super colorful things that we see that is cut out of a silicon crystal it's a synthetic crystal it's grown it can be grown in a lab and they cut Louise into wafers horizontally and then from there you dice the sometimes we referred plurally to as dice to pay on who you ask but they cut the dyes and that's your silicon now getting into why the silicon lottery exists answer the question if you cut let's say we cut sort of towards the middle of the wafer as opposed to the outer edge of the wafer the middle of the wafer is actually prime real estate it's worth a lot more server memory typically comes from the middle of the wafer it is a obviously because it's a circle there's less real estate in the dead center and that's where the lithography is the most accurate so lithography when making these components when making the silicon components way more accurate at the center you end up with higher quality components higher quality silicon and for things that are very intensive or you have specific requirements again like server memory that's where they'll pull those from and then the consumer memory comes from the outer edge or as close to the outer edges they can get before it becomes just unusable so that's number one number one for why does the silicon lottery exist is because it comes down to manufacturing and if you cut stuff like towards the center versus the outside it does actually differ in quality so volt frequency is kind of a place where you can see silicon Lottery comment to play I believe part of the question was how does it impact consumers at the end of the day if you buy an 87 or kr 770 hundred it'll hit spec that's what they're meant to do but overclocking is where it impacts consumers and then once you start overclocking it your volt frequency curve is what looks different so if you have a golden sample as it's called a really good chip you might have a volt frequency curve on an eighty eighty six K that's something like at the very high end maybe you're able to hit 5.3 gigahertz that's only like one point two seven volts which is extremely good very uncommon and so that would be an example of a good bin or of just a silicon lottery when you you've won the silicon lottery at this point a bad example might be something like you're able to hit spec and you can maybe overclock actually we had one our 8700 K originally hit 4.9 gigahertz 1.4 something volts not very good kind of lost silicon lottery so some of that comes down to maybe it was just on a bad part of the wait was that towards the outer edge of the wafer or something like that or it's just some other manufacturing tolerances so there's still multi patterning going on here multi patterning for lithography can introduce error it's something that we talked about in our 10 nanometer discussion with David cancer if you're curious about multi or quad patterning or something like that we talked about it there and that's that's another area where error or just quality differences can be introduced some of them are bad enough that the chip gets thrown out and some of them are bad enough that it it still gets produced but it's just not going to overclock very well or might have might get pushed down the product line maybe instead of being a high-end part it becomes a low-end part and they just turn off some features so that's another thing that can happen with manufacturing tolerances some may be a core is defective they just disable it fuse it off and then you get something like an i3 or an r3 or whatever instead or they could all be perfectly functional and they just few stuff off to fuse stuff off so there's some kind of product segmentation so that it's actually a profitable to make all this stuff so hopefully that answers the basics of it the silicon lottery at the very heart of it when people talk about it they're talking about how high of a frequency can they get with how low a voltage so it's a sustained high frequency with a sustained low voltage that's what you're going for if you get really good numbers on both high frequency low voltage then you won the silicon lottery and it's not like it's a literal lottery anything like that it's just it's manufacturing tolerances end of the day so how well does it all come out of the line when you get your hands on it does it overclock better than expected or average or whatever so that's that's the hopefully a decent answer for you for that next to us from Applejack who said Steve are you open to the idea of selling a gamer's Nexus at mod matte mini actually yes we've been working on a medium mod matte for quite a while now like half a year probably production takes a while we've made multiple samples we're trying to get it as good as we can basically get it to perfect before we start selling it and list it so that will be 32 by 16 for the medium mod matte in inches freedom units so 32 by 16 as opposed to the current one which is 4 feet by 2 feet so we're cutting it down a bit little fit better on people who have desks that are a bit smaller so hopefully that that works out for all of you we're just waiting on some final sampling at this point hopefully this next sample is perfect and we can start making it and have it hopefully in production well before end of year I would like to see that anyway next question is from max VLA who says if we have vintage computer component posters we'd like to donate how do we do so so this is in reference to last episode I think where I talked about wanting to decorate some of the either the set or just the office in general with some older component posters as an option for now you can just I guess post a comment below or actually you know what's best is tweeting at us just tweet at gamers Nexus let me know what you have and we'll figure it out I don't have a peel box presently we have people ship stuff to FedEx or UPS location locally and I go pick it up so we can do that or if there are enough of you who just post in the comments or tweet at me where it makes more sense to open him he'll box I'll do that and I'll announce it in one of the next episodes in a couple weeks probably when I have time to go open one of those up and then we can just give you an address to send stuff to next one is from hsj I guess to close that out let me know the interest from multiple people from like all of you who want to do that that way I can gauge whether or not I need to just email one person and give them a FedEx GPS address whatever or if we need to open something that's better for for a lot of people HS Jason how to determine the phase design on VRMs on motherboards because you can't simply count the number of chokes and call it a day that's definitely correct and I used to think it was done that way and I think it's it's pretty fair that you might think that because if you're just getting into it and all you cease marketing language say in fourteen phases or whatever and then you kind of try and figure out where is that number coming from and you count the chokes a lot of times it might come out to 14 so yeah definitely that's that's where I started with my understanding of it and I think probably a lot of other people too but we learned and some of the things we've learned if you really want to be accurate with it and you don't have a better way to validate something that I've done for builds I'd he does the vrm analysis videos for our channel if he can't figure something out by looking at it because he often doesn't have the board from me he'll tell me to start in around at the components on the board with a DMA digital multimeter and figure out where they all go so you if you have the board in hand it's a bit easier to work with because you can start probing different legs on the on the components like on the inductors and try and figure out how many phases are able to communicate with each other so for example let's say we know there's B core vrm maybe it's an Intel board so you have si and IO its name the board you have V core and SOC so you kind of figure out how big do you think those miner rails are and then you just start probing around that area I'm like the far right of the VR I'm on the far left of the VR I'm because you're not gonna split Veet core in the middle with like VSA or something like that so VCCS a so it's you just find the edge left to right edge bottom edge whatever that's probably where one of your miner rails is and you can probe those and what you're looking for is figuring out which ones are connected to which vrm which which component legs SMD legs and to do that you're just doing a resistance check so you're just looking to see if they are continuous basically and build Zoid has a great video on his channel that's about how to do this and how to identify vrm phases and I would highly recommend you watch it go over to actually hardcore overclock and check out his video on this he did one for us too that was a bit of a slightly different topic but if you look up our intel hades canyon nook and you see no hades canyon vrm analysis probably just hate his canyon vrm analysis type that in he has a video explaining how he does some of the identification of different miner rails and major rails and things like that because that board was kind of complex it was difficult as non-standard for a motherboard obviously so that's a good video to get some ideas on what you can do if you're trying not to do actual measuring of stuff like taking resistance checks for example you can start looking at things like doublers so hopefully it's not a quadruple or they're pretty rare but typically you see those small chips that are kind of seated it's it's normally like on a GPU it'd be like capacitor bank conductors MOSFETs and then just very few sparsely-populated doublers if they have doublers and that makes it pretty easy / - but it could be a quadruple er it could be the memory prm's a bit bigger so if you're not trying to probe stuff you can look at oftentimes the memory of erm on like a GPU will change the inductor or the choke or sorry the inductor menchaca same thing the inductor or the MOSFET and very slight change is a lot of time so if you see the change in component you can read the text on it see that it's a different MOSFET than all the ones below it you know that's part of the memory of erm and so you can just start isolating from there so that's memory of erm with a different two different MOSFETs and the vcore vrm MOSFETs so we put that out to the side ignore it and then you're left with maybe eight more MOSFETs and some doublers and then you know you might have a four phase that's double there's something like that but otherwise resistance checks the best way to do it and check build Zoids video for more on that as he is quite the expert next one is from Kevin Cusick who says with the upcoming Intel CPUs potentially keeping hyper-threading as an exclusive for the i9 now with an exclusive price to match probably how many cores do you need such that hyper threading no longer brings benefits for gaming so years ago and still occasionally hyper threading was actually detrimental to performance and I'm talking like when it first came out basically so hyper threading had some slight overhead such that when enabling it you could actually lose a bit off of average FPS you're not talking a lot it was less than 5% in almost all cases but it happened things have improved a lot Rison went through the same thing Rison first came out disabling SMT simultaneous multithreading same thing basically disabling SMT would sometimes give you like 20 30 percent more performance in some games and that was largely resolved with later a BIOS update it's AG ISA updates all that stuff but the point is on both platforms there is history of being worse with their version of SMT which is sort of the generic brand of it so Tanner's the question these days rise in an Intel both have had enough improvement software developers have optimized enough that hyper threading in essence he actually do in a lot of cases benefit but there is definitely a point of diminishing returns that's the question obviously quick cut there I had to help with something so hopefully I can remember where I was but anyway getting back into it I'll recap a bit of it but applications like blender they definitely can make use of the extra threads even if it's hyper threading which isn't really a true core but it will show a difference for sure because blender just wants threads so it depends on what you're doing Cinebench is another good example set them to the same frequency let them rip and the one with more threads generally speaking if we're talking 8687 or EK same frequency there will be a benefit to having the extra threads at some point you start choking if if there are too many threads and not enough frequency in some applications like xeon cpus some of the older x79 Zeon's that were maybe something like what were they even with twelve core I think there are 12 core might have been ten but let's call it ten I know there was 10 cores young for x79 though is that like three gigahertz 10 cores 20 threads for something like premiere far worse than a four core with a higher frequency so it all depends on what you're doing but that out of the way that aside the answer to your question the majority of games are still stuck at four threads a lot of them especially with modern CryEngine modern Unreal Engine they're capable of using eight threads and will do so very effectively so for CryEngine as example this is the first one that really started moving towards eight threads it's kind of gone these days unfortunately but CryEngine had a dedicated thread for rendering for draw calls things like that dedicated thread for physics for game logic for audio for whatever else needed to be done a iowa's typically done under logic but you'd have eight threads and up to that point it actually did make a difference and it wasn't the ministry in returns and once you unpassed that it was diminishing returns because they just it wouldn't get juggled were scheduled properly same is true for Unreal Engine these days it'll it's capable of using the thread count just kind of depends on at some level what the game is and what kind of changes the developers have made outside of the engine but to answer the question though a lot of games are stuck at four threads but I would say to very directly answer this you said how many cores do you need such that hyper threading no longer brings any benefits for gaming you weren't asking art for production so for gaming I would put it at about six cores these days like there are a lot of games we've tested and and there are benchmarks on our site and on the YouTube channel destiny I think comes to mind spend where the 8600 K and the 8700 K over clock to the same frequency would more or less produce the same result it's not always true definitely not there are instances where the low-end performance like your frame time performance frame time consistency what we call 1% Louis point 1 percent lower just frame times in general and consistency of them there are absolutely instances we're having the extra threads and hyper threading there can help pick up some of the low-end performance and level out your frame time consistency a bit but if it's sufficiently fast as a processor and you have sufficient core count the amount that changes things is really heavily dictated by the game and in a lot of cases they're close enough that in 8680 700k I had a same frequency let's say 5.0 gigahertz if the games not using the extra threads they're basically the same in performance and we've shown that time and again I can't remember Final Fantasy might have been another one wasn't a great game to test in general but that's the real world not every game was perfectly optimized and that was another one of those so hopefully that gives you an idea basically I would peg it at about 6 for the answer for sort of the newer higher-end better built engines and then for for a lot of stuff that's still out there and you remember keep in mind games are in development for five plus years at a time so it could take a while for things to really start switching over but that's it for this one as always you get a stored on camera sacks it's done that to make F one of our new at GN beer glasses we've just restocked cobalt blue color with the gold trim and we do have plenty of other things there as well like the mod mats we've got this shirt the blue print shirt we have the graph logo shirt all that stuff go to patreon.com/scishow and XS to get the bonus episode and subscribe right here on this channel for the second public episode of sgn that'll be going up shortly if not already thank you for the great questions this week leave more below for next week I'll see you all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.