Gadgetory

next time he asked me to fix the car oh I'll fix it it will be an interesting one we're back with ten doing another XOC lab video or just engineering lab video in general for this one so this we have you've got like a scope hooked up of a current clamp and we're gonna show is impacts of things like load line calibration on power consumption and impact of the or actually the measurement difference between like a software voltage measurement for GPU voltage and then an external voltage measurement before that this video is brought to you by Dollar Shave Club and their starter sets get your Dollar Shave Club starter set for five dollars by using our link below available individually for shave shower or oral care or all together in one package the kits can be customized to your needs with options including toothpaste and a toothbrush hydrating shampoo face cleanser and scrubs body wash or a razor with cartridges and shaving cream convenience is key and care packages can be scheduled to send when you need a restock go to dollarshaveclub.com slash gamers nexus or click the link below to learn more yeah that's one of the reasons why we implemented so much controls on the kingdom card so like you can play and find the best setting which fits your specific case you specific game or the benchmark so it's not limited to only one setting which would be for everything and so it's it's gonna be a dense video with a lot of information but this is it's it's some good technical insights to the cards we'll have a separate video on the physical hardware side of things like what tin does to get the cards up to the level that they can eventually advance them into like a KP design or something like that yeah pretty much some pieces on or oversimplified piece of the cloud we'll go through the design of the vrm and make the custom PCBs including like erm design and stuff like that ok so what is this demo going to be what are we showing oh that will show the load line difference and actually how the power is consumed by the cart when the bench who's running okay so I'll have the precision x1 open we can see the current GPU clock it's just running on essentially no overclocking and then all the power is maxed out and memory is not overclocked so we can see the GPU temperature and we can see their power on the drum from 12 volt from using this big program clamp yep see okay so what are we let's talk about each of the lines here so which line should be looking at right so essentially there are three lines here we have voltage the Purple Line and then we have the yellow that is the current from this probe okay but to get the power we actually need to multiply voltage by current and that's what the scope is doing and this is the red line so we have read out this is the RMS power so this is more like average RMS current average setting and then we have the red that is reported kakuhen power so there are two values one is maximum one is RMS maximum is the peak power and then RMS is average all right so right now we're just idling yeah well I then and because this is XOC bias we don't want to have power capping for this illustration so you we can see the power and is actually quite high so it's 90 watts RMS and when we start the benchmark and the benchmark is 3dmark times pi just gt1 test but the similar happens with any benchmark just the different levels okay and that's something we've noticed with times pi extreme for example when i was really desperately trying to get more power out of just chilled your cooling I would do hot key is for increasing the frequency based on the low level of benchmark at a time yep and so for that we kind of noticed that with like gt2 there's certain scenes where you'll see the power consumption spike maybe 40 watts or something like if you actually really engine for the best score you will have the voltage and frequency can apply in the using the profile in the precision so you can actually run one one part of the benchmark with the one level and then another part of the benchmark with different frequency and voltage levels so that's why you can maximize your score in the benchmark but like especially when you're running nail and - you need to handle their own - container and you need to apply the settings right time and all from there if you apply the settings then the benchmark and catch it and will crash out back into the windows yeah you cannot have it for every benchmark right so what what types of what's happening in the benchmark I guess or what kind of load has been applied to the GP where you do see those spikes it like indicative of an increase in memory utilization or well the power will take your measuring here is 12 volt input so this is essentially the complete power taken by GPU and by the memory to get okay but most of the country abuser is of course GPU memory I would say maybe like 20 percent will have total power that goes to them we're looking at a power before the VR MT with the clamp yes yes so there is also if efficiency effect from the vrm itself so like which in the modern vga card can run around 90% efficiency so 10% actually power is dissipated into heat by the vrm controller and there is also some drop on the traces but that would be like couple percent when the power is delivered from the RM point to the GPU die itself right so right now we are running with the load line at default so we can check that level 10 that's the default pre-programmed voltage and the power region is around 420 watts okay maximum and then around 300 watts in the average so I can stop this capture and we can actually see you can guess like the waveform is very repetitive yeah like it's you can almost count actually how much FPS you have running the benchmark just looking at the power that's why I like using the scope for this kind of measurement is very good illustration as if you just use the mm you see all just the number which is average and yeah and see what's the peak or what's the low power that yeah what's what is the waveform look like that actually making the number you see on the screen and when you run the benchmark which is have lighter load then of course this frequency will be faster as well and that the peak and transient of change in the power will be faster so like it's important to test on both low loads and silos right so look at 430 watts or 300 average and this was with the default levels and we can also run it again to take a look on the bow of the temperature 2080 I can pin card so you're able to turn it a bit yes we have extra controls we can flood play with the load line we can play with the different erm settings to find you own not too much power so not too much heat and also enough voltage for GPU to run the maximum frequency we can get and the power is important because it have direct correlation to the temperature GPU is heaven so like we can see like idle is 33 see this is all : one water cooler is essentially the same as containership whatever just no cover and we can keep an eye on the three the hit run I get honest pretty leave it a paused run it so the power is about the same and temperature goes like 40 C this is xoc by so the frequency is bit lower but we can see like this locate like around 40 C and we can see how the changes with the different load line so I can run much slower time scale so we can actually see that in the like recording view so you have the power current the yellow is the current and then when the benchmark stops you go I see I don't condition and then when it start again but let's say we use load line level 4 so that was level 10 yes level 10 is the default settings for kingpin card we can run the same test again and then just to clarify for the audience the closer you get to zero I guess the more aggressive the load lines yes zero essentially GP you will get what you set the voltage if you measure directly at the voltage stands points if you measure at the pro point from the connector on the PCB it will be still small difference and we will talk about that in a minute later okay so we had for DC before and will run little bit longer so go back to the fest scale so we can see the power and it's 416 465 and 308 305 around 300 Watts right average temperatures it'll be hard for you to see Parkinson's a lot higher yep it's 475 320 so it doesn't look like much but actually when you do if you overclock higher and especially when you apply current voltages you will see much bigger difference so the only thing that's changed between this test and one we just did just the world LLC okay and we already free CD cover and we came down for us okay so we came down from all ten to whoever 404 so that's a pretty immediate impact then that's that's very noticeable and then we can do the same thing with level zero run the same test so there are for every piece of silicon be GPU or CPU like there are three main factors that affected them linked together that's the power temperature and the voltage so if you have changed even one of the conditions then two others will change as well so that's why even if you if we chill this GPU to let's say zero C using the liquid nitrogen container the power will be also lower even to the same load not same is same voltage right straight like 480 now 480 watts yeah it's 40 like 490 Peaks 488 494 and we are running level zero now and then you also see the different little bit different so on their power depends on what part of the benchmark is running right but some pieces of the benchmark actually stressing more on the shaders other piece may be like using our TX course this is the not the our TX benchmark not the Port Royale but there's still a lot of different units inside the GPU that work together differently so we are up to a 500 watt peak and power was I see like three hundred twenty watts or so for average yeah and then we're around 40 40 if we set voltage to the fixed level say 1.2 then the card should down okay because I didn't for doing this recipe so we just yeah we just had a V V DD o CP I guess that was okay so we can just set the reference line okay easier to see compare the difference between the load line at aha okay this will be so the line a zero our zero reference and then the B is twenty three point seven amps so you will need to multiply that by 12 to get their power you so the benchmark is over and now I can save to level zero so you can actually see like different parts of the benchmark like right Bart changed I store in slow mode now that way we can actually see differences yeah they're all this one block is twenty seconds so the whole screen would be around hundred seconds so and you can see already power and a much higher level zero is such a good illustration and so that was that level 10 versus zero yes okay and so we're at your reference line is twenty three point seven for B yes and then there we go and power now has become thirty-seven the current reading is around thirty seven point two so that's just from the word line and then the money temperature right 27 yeah is that important to it that's pretty big increase so a little under four amps increase just from changing only the LLC nothing else yep because what happens is when you change the LLC their real voltage that the river to GPU is changes as well and when you have more voltage you have more heat more power because this line is zero what okay and then we have the first test which load line 10 that was 284 watts and then the in the end on which benchmark it dropped to like 263 watts that is Error error Marion so that's not t Caribbean because we on this scale we don't see the peak readings it's like too small to raise what the peak reans starter just follow the same trend or is there any more significantly they happen very fast and very short so they will be actually you will have much higher in much lower but in average like you can like scale them up so in the peak with the load line 0 was 340 watts so we have difference between 280 and 300 for trite which is also putting you closer to like if you're working with a car that has power limits yeah then it that's a consideration of if you start lowering the load line or tighten or you know run a more aggressive load line yes your power you will run into the power limit earlier and your frequency will be actually lower and that's why what you see in the end result as the lower benchmark score rank you would expect like okay I have more voltage more frequency but the score is worse right one of the reasons okay that's another reason why we are running actually xoc bias which doesn't have the power limit because we tallow us to just see the effect of the load line readout being kept by the power right because otherwise we would see different load line but the power is the same because it's all to be limited by the bias and driver settings I think I can show one short thing it has essentially take the reading on the card like because even like on the GOC form we see like some people as connect the DMM to the probit connector and they see difference between readout on the DMM without in the precision and also without in the other muharram or injuring tool like right hardware info or gpu-z and like why why is that and i was where it comes from right so you want to do a manual measurement for that yeah so could my proper riser does the real high speed one which like certified and everything for that which are expressed yes shielded coax wire I know there's like 3m ones are supposed to be the best or something like yeah well this is this is a real deal this is actually 3m is actually very good and there for the PCIe gen3 that will be like the one which I would recommend to reasonably-priced I would say like around 100 USD if you use by 16 right and this one is around like 350 so this is the same tag what is it Sam tag company they specialized on high speed interconnects like the backplane connectors for those plates over oh how hard is it to clamp like the PCIe sorry the the 12-volt lines coming out of the Reiser just like split them out yeah a lot of the clampers yes essentially I would need to do the same as here so I have the ground bypass just go directly to a cart and I connect only the 12 or wire so maybe we can actually talk about that a second okay essentially they've had their current measurement works use the big clamp probe you can buy this is the probe for the scope it's rated 450 amps but there are also like DMMs which made like a clamp so the the probe this progress for the scope it's designed to interface and readout with the current with the high speed so that's why it's like so bulky and big right now a clamp in the five volts in that you are just 12 no this is just 12 volts okay three cables from the power supply I say to the card okay and like if you want to have hundred percent accurate measurement for the power we would also need to help clamp the 12 volt coming from the PC East London but because on the KP card the power input is not drawing much power for the sake of the demo we mean gray otherwise I'll need to cut this wire and run through the clamp as well or you use another clamp connect to the second channel of the scope to measure the board so what what I want to show here is how the voltage changes depends on what settings we applied so we have the reading from the precision which is this GPU voltage it reads 1.17 2 volts this is we're just idling the 3d mode and then we applied 1.15 to the erm controller and I have the DMM cooked to the probit connector this is just reading m dv/dt doesn't read anything else so we read around one point one three four okay and what what is the okay so one point one three four versus one point one seven five and one point one seven three and it's all supposed to be the same thing ultimately it's just different places of being measured correct so the difference comes from how the measurement is done so Riggin which we see in the precision is taking by the hardware monitor and it pretty close to the GPU die itself and that's like a reason why we see the higher voltage because controller compensate for the voltage draw up on the PCB and on the GPU connection and we see essentially voltage closer to what's what's actually GPU getting and reading from the DMM is little bit lower as it reads from the vrm side okay and satan is somewhere in between because we are just idling and its effects also for the load line so the setting is just fixed value for no matter what now yep right so if i run the same benchmark three mark times by gt1 then we can see how this voltage is also change I have illustrated the differences in OSI guide for the kingpin card mmm it's public on the X - comm kite section so there are little bit more details on that gt1 and window mode and we'll keep an eye on our voltages so what's the the most realistic or practical way to get an accurate voltage measurement like out of all these numbers what do you trust most DMM or I will trust the current the reading from the precision okay because actually we spend some time to fine tune and make sure that voltage as close as GPU actually gets as possible so what does the fine tuning only apply to the EVGA cards I mean if I plugged in some other card I think it applies universally but like the level how much effort is actually engineers put into that aspect is very what because you like in the normal condition for the everyday user or gamer like you don't really care about my voltage is 0.1 percent of what is right so you don't even know the voltage most of the time because you don't you cannot monitor it easily on the reference cards there are no like yes you can connect the probes manually probe the point but you also get the different voltages from depends where you measure them like right now we are reading the benchmark is running and we're reading zero point one zero five and you can see the voltage in precision actually fire yes we've gone between well that's so going back to idle still 3d clocks I guess but idle 1.13 and then when it was running it was like one point one or one point as your five one one zero five and that is with level ten so we have the most blood lead rope now and we can change the load linings you have the changes as well so you will see less difference in the voltage change from the precision without them actually from the DMR because EMM is more like vrm centric and the reading the precision is more actually representing the type of course and check okay 11:23 1.05 on the DMM americans level zero and i've gone twelve twenty ish and precision and one point two five so they're both the voltages goes up by one point one two five because they're on level zero we are actually over compensate for that so that'll cover us for the voltage demonstrations that LLC demonstrations i think ten was worried that we were starting to look like two old men sitting down so we decided to stand again this definitely was not shot out of order that would be weird so thanks for watching the video subscribe for more could have a chicago / gamers nexus tells out directly and to check back for our other content with 10:10 thank you for joining me sure no problem we'll see you all next time