Gadgetory

everyone we're doing an episode of ask GN it might be a little bit of bad code because we're in a cable car we're a gondola going peak to peak at Whistler this our last day here after PAX West drove up and we'll put a timer somewhere in the corner because I'm on a clock for this before we get to the other side before that this coverage is brought to you by the core g21 enclosure from thermal take a $70 case with two four millimeter thick tempered glass side panels and a power supply shroud with top mounted SSD sleds learn more at the link in the description below okay so STM first question enigma one asked Steve what's the latest on hardware Lux is Vega 56 under bolt validation so that comes from the livestream Vega 56 under volting kind of works but sometimes it doesn't so in that live stream you can see that I wasn't able to successfully replicate the results that doesn't mean the results didn't happen it means that my car didn't like those numbers and specifically here's the thing it looks like it accepts the under bolt but if you take a DMM to the back of the card multimeter and check the voltages it'll actually be a bit higher than what you tell it to run it so that's expected but the voltage wasn't dropping as much as we were telling it to so it looks like it's working but it's just I'm not fully confident in it basically I think the software it's kind of lying to us like a lot of other things with via the Vegas software so I'm not sure about the results they could be completely valid but we can't replicate it under Bolton's just like overclocking some silicon likes it some doesn't next question for him says Steve I have a question regarding memory bandwidth and GPUs different graphics cards can have different bandwidth like 200 300 400 gigabits per second or gigabytes per second rather what does that specifically mean how much data the GPU can move per second or something else because higher bandwidth doesn't really mean faster cards so as quickly as possible this is the speed at which the GPU can maximally move data to and from memory some GPUs care a lot more about memory bandwidth than others you can look at the Vega GPUs they get choked on bandwidth which is provable by overclocking the HBM and basically what it comes down to is well let's do the calculation first so to calculate many memory bandwidth do you take the memory bus width divided by 8 which converts your bits to bytes makes a bit easier and then you multiply that by the memory clock the actual clock drop the effective 1 times 2 for ddr times 2 for gddr5 and some architectures are more bound than others so depend what number you get they may not mean anything that may mean a lot with Vega we tend to be limited on bandwidth with Furio tend to be limited this is getting very high up and then with some of the Nvidia cards it behaved a bit differently Nvidia doesn't need the bandwidth quite as much as the bigoted architecture does part of that is the color compression memory compression so as you push and pull data from memory if you can compress a lot more of that data you can compress the colors 8 to 1 even stuff like that then you really reduce your reliance on that pipe being wider the memory bandwidth being larger and in numerical value so the more compression you have the better off you are higher memory bandwidth doesn't mean better card but it is important for some architectures especially all eventually at some level a quick example though when I wrote down here we take a gtx 1060 that's 192 gigabytes per second memory bandwidth 192 bit bus the equivalence between this two numbers is basically a coincidence so 192 divided by 8 equals 24 times 8 gigabits per second effective you get 192 gigabytes per second or if you're working from base frequency which makes more sense that would be 2002 megahertz so 192 divided by 8 24 times 2002 gives you 48,000 48 times 2 for DDR 96 gigabytes times 2 for gddr5 192 for a second so that's a quick explanation of memory bandwidth think we're about halfway through this video oh yeah compression matters a lot but that's that's the very basics maybe we'll do something separate when I'm like not in a car doing this as Street guru these are all from discord today says should Andy add an automatic under boat utility to wat man we're in the GPU would where and the GP would render a scene at a user-specified frequency say 1600 megahertz or Vega then gradually scaled down the voltage until it sees instability where it hopefully stops itself to prevent a crash all the data could then be output to a file where the user could set States and white man accordingly this is a fantastic idea I hope AMD takes your idea so vote frequency tuning exists in precision and in the NVIDIA Pascal and onward cards there is a volt frequency curve for AMD also it exists a lot more in Vega than in previous article textures so it's possible for them to do this and I would think you'd basically run it through a bunch of checks at different and take a while run through different frequencies and voltages see where it's not stable and output your results from there and then load them if you want to load them and if you I don't know I think the challenge is how do you recover from the driver crash how do you save the stage do you put it out to a text file for the user or do you try and do it all for them that's the challenge but they have plenty of other things to work on on the driver side so not sure that'll come anytime soon but I next question is from contractor three one six who says Steve do you know if there's any qualitative performance and/or power consumption difference between using a single split PCIe cable or two separate PCIe cables to power the graphics card and what's the effect on the power supply so not really so this is just you don't really want a daisy chain cables that's what you're talking I was where you get one that splits we did that in the cooler master C 700p built that we did at PAX this past weekend and I did it because it was cleaner but I mean if you're trying to be technically correct using two separate cables would be it's just a matter of are you exceeding the amount of power that the gauge of that cable can handle generally the answer's no you're probably going to be fine but if you're working with say an overclocked tin pin card or high overclock lightning Z something like that you're pushing enough power where it's probably time to use two separate cables reduce the heat load on the cables so that you're not burning through your wire by pushing 300 Watts through it or something like that and they can handle more than the PCIe spec says that they can handle but there's really there's no reason to to beat the cables up like that if you can run an extreme overclockers high overclock you'll improve your stability a little bit as you approach the upper limit of the cables you'll reduce the heat load on that specific cable rails come into play here to 12 volt rails maybe we can get builds or to talk a bit about that sometime the future and do some demonstrations of what happens when you do use a daisy chain cable can you make it catch fire or melt or anything like that you can it just depends on what kind of hardware you're working with and if you are well the demonstration parts are because you don't want it damaged the high end card obviously I can produce that much current and power so that's the concern with that but yeah I think it just comes down to like if you're working with lower wattage cards you'll be fine with a daisy chain but just try to do two separate cables so that's mostly questions I had here for today and this thing is still going so we can add a couple more let's talk about the under bolting a bit more and so the backlit card under bolting the first question that I talked about will have a I can put a photo have Andrew put a photo in the video of the voltage checkpoints on the back and you can check for HBM to voltage that you can check for vcore so take the you know I'm the DMM take the black and stick it into a ground on a molex cable that's probably the easiest you want to be as close to the sword as possible and then take the red and probe the back of card you can check the HBM to voltage you can check v core they will be a little bit higher than what software tells you because just because we're you're checking of a stage but that will give you a better idea of if your voltage is actually changing from the one under bolting this is probably more accessible than the other solution which is take a current clamp clamp the cables and check the current ideally you do both to make sure that the power actually goes down which tends to coincide with voltage because volts times amps equals watts so you see one go down probably the other one went down but just to really make sure check the voltage on the back of card you'll see a bit higher than software but it should go down each time you step it down and software relative to each other so if you drop 10 millivolts you should drop 10 millivolts in the back of the card readout or roughly there about so that would be something I'd suggest we'll have a photo to show you where you check those yourself if you want to remove the backplate and you're good to go on that we're approaching the end here I think so if you want to check out content that was from Whistler like a bike park check the G on Steve channel we'll leave a link below you can subscribe to that one thing we're gonna start building that one up a little bit more for non-gamers Nexus content stuff that's like kind of related like the riding down the mountain while talking about coffee lake and things like that that's probably the most related we'll get so check out the GN Steve channel for that hope the echo isn't too bad for you all thank you for watching and subscribe for more patreon.com/scishow and was an access to help side directly I'll see you all next time you