Gadgetory

hey everyone and welcome back to another stn this one we're shooting in the old studio just for the intro though because i'm joined by senior am the analyst snowflake that was impressive snowflake I thought you'd bail on me and if you're new here I just wanted to give you a look at where we came from just recently the last couple of months actually I know a lot of you were new from the Walmart video and other videos but if you've been a longtime GN viewer of course give it up for a snowflake in the comments below if you have questions for next week's asked well hopefully next week's ask GN episode leave them in the comment section below and I'll do my best to get to your questions for technical questions for the next acts GN and if you're new here this is a Q&A segment where we take your questions whether here or on patreon you go to patreon.com/scishow gamers next to leave a question there in the discord and we try to answer them to the best of our abilities and this is senior AMD analyst snowflake before that this video is brought to you by ether all take 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 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 so before we transport teleport over to the proper set the real set the new set these days this is what happens the old ones elude while still here you know the iconic wood wall there it is it's still around and I've mounted some bike frames to the wall for the side channel the GMCs GN steve side channel choice does that we'll be doing some more videos on that channel on this set at some point not sure exactly when but that's what's been going on here if you've been wandering it's been a lot of work but let's teleport over to the main set now where the audio is better and also everything else including the lighting all right so this isn't ask GN a lot of you already know the rules if you have a question as I said earlier post it in the comment section below or if you have access to our patreon only discord you can go to patreon.com/scishow cameras Nexus you can also post your questions in the ask GM channel it's been a little while since we've done one of these so we might end up doing two for this weekend but let's just start with the one and see how it goes first question is from Sauron and Sauron said so I just watched dare Bauer's video where he talks about the 9900 K stim issues and he sanded down the dot I was under the impression that the dye was the processor and all the transistors were in there but if it's okay to send it down then what really is the dye and where are the transistors located the PCB under the dye that makes me think again what the dye really is and what its function is this is a great question and fortunately most of those are pretty easy to answer at at least a service level so the most immediate answer I'll give you what is the dye well the dye is a much it's a smaller piece of silicon underneath that shiny layer that shiny layer is a diffusion barrier and I'll go over some more of that in a second with some notes from there Bauer as well the one you referenced and so under that diffusion barrier you eventually have the actual silicon die depend on which processor it is sometimes that's facing down towards the substrate sometimes it's facing up it depends on which one you're working with but the you call it let's see you said the PCB under the die so the that PCB that's a commonly called a substrate and that's just an inter connecting grid of traces to connect all the everything together on the CPU proper but the dye itself is a smaller piece of silicon that sits under the diffusion barrier so what is that diffusion barrier well this one eyes to do some research on myself and I knew the basics of it already because I've also seen that video and I spoke with Roman a bit in the past about this so I have Rowan being their power so I have some of his input as well alongside quotes from an old article of his written in 2015 that's to this day still accurate so let's go through my notes first then we'll go through his is written version so the the shiny layer is made of silicon nitride and it is a barrier that has worse thermal conductivity than just straight silicon the problem is exposing just straight silicon to really anything else can potentially cause corruption of the silicon you could end up with issues later on from materials diffusing into the silicon so there's a barrier there at diffusion barrier silicon quoting their Bower is about 150 watts per meter Kelvin thermal conductivity which is pretty damn high not it's a bit below aluminum give you some perspective silicon nitride again quoting their Bower depending on the application is between 30 and 100 watts per meter Kelvin that's a massive range but it really depends on the thickness of the material how they're using it what type of silicon nitride they're using but that's your range some atoms of liquid metal in his example could diffuse into the dye if you sand down the surface so there's some risk there inherently withstanding down the dye aside from the existing risk that you might sand a bit too far and when you're dealing with literally billions of transistors under that shiny layer at some point you go a little to the even one part but not the other you could kill something and and that that something may very well be the entire CPU so you have to be careful he ever do that kind of deal it in extremely liddie and wouldn't really recommend it we've done it but i wouldn't recommend it but anyway any kind of diffuse isn't diffusion into the silicon over time could damage it so as Roman said in his video one of the concerns with doing I sanded down die or sanding off that diffusion barrier is that over time whatever interface you put on top of the then sand it down dye could cause damage to the dye itself because you no longer have that protective barrier between it so on his 79 80 XE he removed between 30 and 80 micrometer zuv microns of silicon and 80 at the center so that was the deepest part now a die especially in tiles they tend to be they are not level all the way across he'll get kind of a concave IHS but that's not the same as the dye itself the dye itself have differing heights in the corners and the center as he showed the center is a bit higher on the die on that diffusion barrier and the one he was working on so silicon nitride layers are used to protect the dye underneath they are particularly good against things like corrosion interference caused by ions or other chemicals that you don't want interacting with the silicon you also use diffusion barriers between the thermal paste and the silicon die or in this his case liquid metal and thrown paste is topped by nickel-plated copper which is another material more chemicals that could end up plating or something like that so let me read what what Roman said to give you a a well researched answer as well so Roman in an old article he wrote in 2015 said quote nickel will act as a diffusion barrier to prevent any atoms to form an alloy with the copper indium also sticks to nickel but not really well so to improve the adhesion you have to apply another layer on top preferably using a noble metal because they provide some of the best wedding conditions examples would be gold silver or palladium and the melting point of silver and gold is quite similar at around 1,000 degrees Celsius melting point of palladium is at 1555 Celsius which makes it much harder to apply on the heat spreader so choosing between gold and silver gold easily forms an alloy with indium and has great wedding conditions before you can think of soldering you have to apply a gold layer on the heat spreader the gold layer has to be around 1 to 3 microns thick the diet is made out of silicon si but you can't solder directly to silicon this is another reason that the viewing barriers there otherwise indium would diffuse into the silicon which would result in a different doping characteristic or damage the chip over time so you need another diffusion barrier on top of the CPU the diffusion barrier is formed out of several layers made of titanium nickel and vanadium on top of the diffusion barrier you need another gold layer as wetting layer for the indium connection so if you want to read more about this that's kind of the whole section he had on your exact question but if you want to read more it's an old I think it's an overclocking guy is overclocking that guide as a site you can find it there or of course you can go if you don't know who he is to dare bowers channel we would highly recommend it because he's a great source for extreme overclocking and other material property information so hopefully that helps you out with the question I want to went ahead and quoted his response for the most detail that we could get on short notice although I have spoken with him about this question in the past so next one this turn was from sad panda so sad panda asks asked a question on your last live stream and you gave a quick thought but wondering a little more i'm overclocking a non ki v 6400 at 4.7 gigahertz 1.3 to 5 volt so i have to disable c states turbo boost full for a full time overclock three days power usage and heat I've been using level 8 LLC to ramp up voltage under load and drop back down to one point two eight zero at idle this seems like it has the effect I'm looking for but I'm worried that max LLC that's load line calibration you don't know maybe spiking higher than what is reported in hardware info is this something I should worry about and test with a no scope order the spikes normally report accurately as max B core in hardware info another good question so for this one it depends on the motherboard a lot typically hardware info at least these days this is a bit of an older CPU you're talking about but on a lot of the modern products we work with Hardware infos pretty damn accurate there is still some variance and we spoke with asus recently about this so what I have here is a great discussion topic from from your question but we spoke with a C's about this previously so the chart we're going to put on this tribute to the screen is from John Sandstrom aka Elmore you may know him from overclocking forums Elmore worked at Asus just up until recently and really didn't get enough credit for the engineering work he's contributed to a Seuss's products BIOS things like that I'm sure he got credit internally but externally wasn't wasn't wasn't that famous as he deserved to be or maybe it was too famous depending on if you're talking to him or not either way Elmore got hounded constantly by the community for questions I got hounded constantly by press with questions and one of the things he did was provide us with an example right before he left but the the level of detail he'd go into so tested at 5 gigahertz on one point three volts with a 9900 ki1 found the table on the screen now for the Maximus 11 hero and Maximus 10 hero and this is a quote from him he said as you can see M 11 hero should be at most 10 to 15 millivolts off on the on-die sense measurement and is primarily due to the SI o ADC granularity and error characteristics the Maximus 10 hero is using a simpler voltage sense circuit and is in this case up over 100 millivolts off under heavy load you should see similar values if you measure on other boards he continued saying there are no on-die sensors there's only on Dai sense what this means is that there are two pins on the CPU for measuring the quote actual voltage reaching the IC without any current bias on the Maximus 11 we've added a circuit for feeding this differential signal into the super i/o ADC or analog to digital converter BIOS cpu-z and hardware info will all read this value from the s io and Elmer also sent us some photos we can put up on the screen of measurement points for a V core andand I sense for those tomb other words the max was 10 and 11 so to really get into your answer then it is possible that a high load line calibration can overshoot and you might spike up instead of down like with a more conservative LLC we have video on the channel by bill joy where he explains how LLC works I think it's probably called how load line calibration works search it on our channel and that should help answer your question in greater depth but to really give you a straight answer here hardware info cpz things like that are not always accurate it varies board to board we've found that they are pretty accurate with ninth gen gen and with the Maximus 11 hero and that's helpful for figuring out if you're getting the voltage you think you're getting now you the best way if you really want them to be careful to measure this you could take a DMM you stand the board up maybe it's case and you have good access to the back you stand the board up somehow and you when it's off you find the capacitors in the socket the CPU socket backside that correspond to V core you can I mean his photo really helps a lot with making that easier but if you don't know where they are you don't have those boards what you do is you probe like a leg on a choke or a MOSFET from the V Corps v RM and then you probe all the different capacitors on the back of the CPU socket until you find one that is continuous so you're doing the resistance probe and once you find the one that's well one of the ones that goes to V core you can measure both sides of the cap this is dangerous so be careful the reason it's dangerous it's not not like physically going to harm you but it's dangerous probably but just to cover myself there almost certainly not going to there's really no there's there's no meaningful amount of of current going through there that's going to hurt you but the only reason it's dangerous is because you can kill the component and that's because you're working with a really small capacitor and so if you slip when you're probing and you bridge two caps or a cap and something else then they can obviously arc to one another and that might very well kill the motherboard so that's the only reason there's any danger to doing that if you feel really uncomfortable probing two sides of the same tiny cap there's the best way to get an accurate measurement for your actual voltage then what you can do you could either solder small wires to it kind of like Eleanor did don't really want to do that though so perhaps easier would be probe one side of the cap and then stick the ground into a molex cable you're gonna have variants here so we've measured it between 20 and 50 millivolts and that'll throw off your numbers but it's kind of close enough where you'll be in the ballpark and figure out if you're you're within safety for your voltage and that would that will vary based on the length of the molex cable and stuff like the vary on the power supply to basically you want the ground that's closest to that capacitor as possible and the closest is grounded on the capacitor is going to be the one literally on the capacitor but if you don't feel comfortable doing that grab some other ground nearby and a molex cables just easy you can you can shove it into the molex cable it's the probes like the perfect that pen and it's not gonna go anywhere but um to answer your question though it depends on the motherboard but typically typically they are pretty accurate these days just you know that's how you would check if you're the one to be sure and next what oh yeah and also check our video from builds right in the past on LLC if you want more information on that next one so this is from banana milk it says what makes some 100% CPU GPU loads more intense than others Intel burn test AVX workloads fuzzy donut etc fuzzy Donuts the best one I can speak to you because I know pretty well how it works we use it all the time so the fuzzy donut is a its fur mark and it's a great GPU torture test it's known as a power virus because in the past it was used to literally kill GPUs by way of pulling too much current or buy too much too much current load on the MOSFETs specifically so you'd have MOSFETs just pop and I know pretty much all of our core audience knows what this is at this point we have a lot of new people from the Walmart video so your MOSFETs are right there next to the inductor line those are the inductors it's capacitor bank and the the old versions of Nvidia AMD drivers ati drivers in the old versions of fir mark you could very well push too much power because it's a power virus and blow up a a mosfet or some other vrm component so that's been changed in these days for mark if you run it you'll notice the frequency is actually significantly lower than base and boost and that's why we always make a note when we do those tasks with fir mark that yeah it says 1500 megahertz don't worry there's nothing wrong with the card it's just that it's not gonna run it 19 100 when it's been absolutely tortured with something that's so thermally constraining so anyway what makes it thermally constrained there's a few reasons for this let's talk about the base of your question which is what makes different loads more intense than others a good example is times by extremes gt2 graphics test - this is a very memory intensive workload if you veer M intensive specifically so if you open up I don't know you could open up gpu-z for example and you look at the vram request it's not necessarily what's being used but it's pretty damn close so look at V requested you'll see it's fairly intensive and also if you increase the memory clocks in that test gt2 versus GT one you'll find that the maximum stable overclock will be different between them and that's because in GT one you're really straining the core and GT 2 you're really straining the memory so if you can boost the memory a bit higher in GT 2 your score will be affected in a disproportionate way than if you boost the core clock in GT 2 the difficulty is that in GT 2 we've seen the vram frequency is less likely to be stable so you can more likely get away with boosting the core frequency further it's just that the payoff is less because the workload is primarily on GT tip as on vram and GT 2 and the reason that you have these differences and we're ones more memory intensive and one isn't is really just the way the applications program then what kind of work look what kind of in this case textures you're dealing with and assets and things like that so to go further on this some CPU loads might strain the memory controller and core simultaneously which produces additional heat on the CPU because now you have multiple components active and really doing work so you have more heat in a smaller area an application like fur mark is stressful because in part it never stops loading the GPU and a lot of gaming or real workloads there will be contact switching that can give the cores a moment's pause to cool down where a moment is a very small period of time with stress testers like fur mark it's loading the GPU to 99% constantly and as unrelenting you'll notice that fur mark doesn't really load the memory the command processors and the GPU never really have to talk to the CPU via PCIe for the most part in a real gaming or workstation application your CPU is going to be constantly shoving textures and assets from storage into the GPUs command processor remember everything except for DMA everything pretty much has to go through the CPU at some point so it's shoving stuff into the command processor the GPUs command processor then shoves that data into its video memory and switching assets and changing the type of rendering workload like for example switching between geometry processing and texture filtering and streaming or copies transcodes and encodes rather decodes that will give small periods of time where some of those lanes on the vector aren't really doing anything that lanes on the vector here would be CUDA cores stream processors things like that so GPU utilization is measured in multiple parts one hundred percent utilization could mean that you're copying on copies you're 100 percent utilized on everything that handles copying or maybe 100 percent utilized on everything that handles video encode and decode or compute or maybe the graphics that the singular graphics thread in a game scene one hard percent utilization can refer to a limitation in one part of the pipeline but maybe not others so you might have other parts of the GPU that are completely unutilized meaning that only part of the GP is truly at 100% load and so your your thermal difference there is going to be less significant than an application like fur mark where it's constantly doing work so with something like fur marks fuzzy donut you can cram all of the data necessary to run the computation that it's that is running for fur into the cache for each cluster of shaders and then you burn those genuinely at 100% without ever letting up and giving them a chance to breathe so to speak so the best way I like to think about it is to imagine each component as functionally its own computer because it is at this point a CPU is more or less a complete computer within itself it can't exist without the rest of the computer core it can exist but can't really do much without the rest of the system but if you kind of look at the architecture of a CPU it's got some memory in it it's got a memory controller and it has cores and it can do all the functions needed of a computer it's just it needs a host system they even have voltage regulators and a lot of them these days you look at a GPU same thing so they these parts when you think of them as a full computer within itself it's easier to see how it can bottleneck in different parts of the silicon but not all of the cores or all of the the lanes on the vectors as as it may be so you may for example when you see a hundred percent load on your CPU Washington 4k video 60fps on maybe a laptop with just an IEP you might see video encoding decode threads if you really dig for it hitting 100 percent but 3d is not really doing anything or a copy is not really doing anything and you can see this even through task manager these days it's really useful for visualizing it so hopefully that helps you with that question though a very good one and then the next one is from CG Fuu who says Steve in the PC game and video game community more broadly we're starting to hear more and more about the problems with crunch culture as developers approach release of a game and are asked to work a high number of hours by management you yourself have talked about the hours you've worked during launch review periods in the brief window between receiving hardware for review embargo lifts my question is given your contacts in the industry does a similar culture of crunch exist on the engineering and marketing side that the companies that develop the products we all love to evaluate I recognize that the time horizon ie when the crunch happens is probably different for someone developing in RTX 20 80 TI compared to The Witcher 3 but does a culture of crunch pervade Intel AMD or Nvidia and other firms like Asus MSI EVGA ahead of the release of a new product not the short answer is yes it does exist so there was a period where for a long time we did a game only content no hardware anything and I worked I lived in the game industry for that entire for a while doing that and in fact there was also a period when I was interested in becoming a developer before I realized that doing this was more enjoyable to me more suitable to me and in that time researching everything and talking and interviewing with developers we have a lot of them on on the website before we really did any of the YouTube stuff the type of crunch is a little bit different in my experience so with time horizons one thing like you said but the type of crunch with I don't know GPU development you kind of you don't necessarily have everybody in the whole company under crunch it's more like people in media relations and technical marketing and testing and validation those people are really putting in the hours once the product is in that final week stretch before it releases and that's because they're dealing with often hundreds of media finding all kinds of problems or thinking they find problems and then technical marketing's job is to find the truth PRS job is to filter the truth so that the company doesn't get damaged by whatever that may be and that kind of that that week-long period hopefully before a product lifts is definitely a crunch period for those that team I know people who work at some of the companies you mentioned I guess I won't name any of them specifically but I know plenty of people at all these companies who in speaking with them at events and before launches and things like that they're just like walking zombies a lot of the time which isn't too different for media we get it bad too and it's really for us it's it's more of what you make of it I mean we can get away with skipping an embargo if we really had to and I've done it but these these people can do this they'll be fired so yes crunch exists and the extent of it I think it's a little different so in games I mean it's gotten a little better these days from what I understand but I know companies that it was not uncommon where he'd be everyone in the company be working like a hundred hour week leading up to going gold or something and that's obviously not really healthy for the company I we've interviewed Richard Garriott famous for Ultima we've interviewed Chris Roberts famous for Wing Commander and now star citizen whatever you make of that doesn't really matter for this point but speaking with both of them especially Richard Garriott when he was at origin I mean they had like bunks and couches in their offices because people would stay there and sleep or nap and get back to work and I remember speaking with him this is obviously years after origin closed and he told me about when they eventually got rid of that stuff because they realized that it was just burning through people too much and the turnover you get from people not getting a break was worse than you know driving those people to finish the prod to to the standards you want because if you lose that person than your quality on the next thing will be lower because you have to train someone up to speed but I would say that's that industry is to my knowledge a bit worse than what we see at the companies and hardware but absolutely there's still crunch and hardware I I can think of individual people at every one of these companies you've named who work way too many hours there's a guy well I don't need kind of specifics I guess but there are people who have had like serious injuries from accidents and things like that or they they're working from the hospital there's people who are working just I'll get messages at like 2:00 a.m. their time because they're trying to follow up make sure everything's good for launch day and everyone bust their ass and worked really hard and we all love it for the most part so I mean I like what I do and it worked again where it's like 105 hours for RTX launch so yeah we work a lot but on the media side at least we can control it I didn't have to work that many hours if I didn't want to and it is fun so it's not just work on the company side it's it is a bit of a problem but there's not a lot you can do everyone's got low margins with GPU you sell a video card you're making like 3% margin you can't really hire enough people to double up so that your your main person didn't get away with you know a ten-hour break or something on what today so it's a problem when margins are that's limits it's competitive world last two so I wanted to bring this one up I've seen it in the past didn't see a specific person asking this recently but in the past I've seen people asking do you turn down ad so what are your ad standards and I did want to bring this one up because I just turned a couple down so just so everyone's aware the reason you see like GM store ads sort of towards the end of the month is because we've served through all of our advertisers for what we think are quality ads or ads worthy of sort of being attached to our name and so serving GN store ads at that point helps because it helps and sauce from selling ads to less scrupulous companies that we don't really think deserve to be attached to our name or don't make a good product and it also means that we can distance ourselves from reliance on advertising money in general and instead push something like our store for example or patreon and that just means we can be more independent and and keep the reviews material as financially distant as possible from the companies that provide the review products so wanted to bring up because I just turned down an ad from a marketing company that specifically wants hit up an influencer which we are not I we don't we don't call it there's a specific connotation for that word and we don't call ourselves influencers review outlet and they wanted to influence here they wanted like to pay for a review which is absolutely wrong no painful reviews is is extremely ethically bad they want it to pay for like a I don't know a product overview or something and some other biessed here so I just wanted to know why you see the store ads because we've posted like four in a row at the end of the month so I figured I'd mentioned why it's because we turn down ads that are we think low-quality or that compromise integrity and then the last one here there are some people one guy specifically one person one person asked so AMD is responsible for its board partners on the RX 580 2048 review where I was bashing AMD for launching a misleading GPU and I think there was some stuff lost in translation there because it was a China only product but that data lands card isn't the only one so there's like all the vendors in China have the ability to make this rx 580 2048 the reason for that is because AMD made that GPU so I want to make that really clear because I don't know if I thought I made it clear enough but I guess not that is an AMD product the rx 580 2048 you can go to Andy's website and it's there so it's not like data land went rogue and made their own silicon or made their own cut out rebranded five seventy whatever that wasn't that that was an AMD product and this happens we're like the board partners are not going to rebrand GPUs they will no longer be a board partner if they do that and their business will die so justice everyone understands how it works I think about everyone pretty much does but GPU supplier Nvidia AMD works with the board partner they sell them the GPU often the memory attached to the PCB the board partner is responsible for figuring out the PCB if they don't buy the reference one and they're responsible figuring out the cooler maybe a custom fan curve stuff like that but they have nothing to do with making the GPU itself and the only time you really see scams is typically on ebay from a no-name person who is maybe flashing BIOS or something in spoofy and the cards name and we've shown that in the past before but that's not the same thing here sorry that's all for this week if you have more questions post them in the comment section below try to do more of these and you can go to patreon.com/scishow gamers axis to get access to a special STM that has a couple of extra questions thank you for watching subscribe for more and go to store that gamers access net to help us out there as well I'll see you all next time