How to check bad memory on a graphics card GDDR5 GTX 1070 fix

hello there we are about to take a look at another 1070 from evga i've already spent a little bit of time fixing these burned beads i [Music] also replaced the memory face um mosfet controller turned out to be okay but the um card still does not boot so and this video is going to be a little bit about memory so in order to test the card this particular one that uses a reference board and by reference board i mean this board is kind of universal you can have up to six phases here for the gpu and you can even have two memory mosfets that they're all gonna be feeding the same coil um in any event this particular board will not run mats without the external power connector even though under normal circumstances you can power on the card if everything works it'll give you an image and it will complain about the external power supply so but you gotta keep in mind that without the external power supply the memory test will not run so with that said we're just gonna turn that on and as usual the monitor gives no picture and it takes a while you can see there that we are stuck on code 71 and the reset if you look at the light over there the reset has been released which is good and we're just kind of sitting here and chilling waiting for the monitor to kind of light up and there we go the light stopped blinking the display lit up but there's no image and we're waiting for the mats to run in the background and i've configured my test to shut down the computer at the end of the test and right now we're just waiting for a test to complete and it'll shut down the motherboard automatically by the way i think i'm testing about five megabytes five megabytes so we don't really need to test anything more than that and there we go okay turn that off okay so now we're looking at the report and it complains about the bank one and a bank one in case you didn't know is going to be this chip right there so we're gonna try to get it off the board and see if there's any soldering problems and we're going to measure its resistance to make sure that it works or it possibly works we're going to try to re-ball and resolder it back on and if that doesn't work we're going to replace it and see if that solves the problem [Music] so [Music] [Music] [Music] okay so there are a couple of ways to check for a ddr5 memory gt05 first and the most obvious test to do is to check for resistance on the actual uh voltage on on the vcc and that's going to be your first [Music] first pin on pretty much any corner so you can look on this chart that every every outer most corner is ground so you can put one on the ground and then the one right underneath it so it doesn't matter which which corner you're going to be doing this from either this corner that corner this corner or this corner is always first pin and then the second one down so and that's you're going to be measuring that for resistance and for a good working chip on a gddr5 by micron and let me give you a specific marking because some different different revisions of this chip will uh have a slightly different uh reading but this one is a most common one and it's reading is d9tcb so basically we're measuring the resistance from here to there and uh that's your memory resistance that you will be measuring on your card like so like you typically check your from free memory business like this so you will be getting around uh 10 kilo ohms um 10 to 11 kilo ohms is maybe maybe nine is what i would consider a working chip so this chip here actually gets let's see what this chip gets this particular one 9.5 so i get 9.5 uh let me let me show it to you so you can see it yourself there so we get 9.5 kilo ohms so this chip is supposedly alive however we're getting a data rate errors so there's also another check that you can do we can check the data lines and the data lines are checked in the diode mode so you flip your uh multimeter to a dive mode and then you proceed basically watching the voltage drop so you put one probe on the ground and then you start probing these uh these gray areas here and they all should they all should have uh more or less the same uh voltage drop and uh i'm going to show it to you now but you got to keep in mind that each multimeter um sends a different voltage when it's in diode mode in order to measure the voltage drop some multimeters send three volts this particular one i believe measures uh it sends out two volt signal so and you can literally hook up another multimeter to your leads while this multimeter is in the uh diode mode and actually measure how much volts it sends and that way you know so i believe this one sends two volts so i would place like i said one probe to the ground and then the red probe i will start poking all these data data lines this this this this until i find one that deviates from the rest of them and it shouldn't deviate much um and if it does deviate quite a bit then we know we have a problem so i i already went ahead and i scattered across all of them and i did find one let me show you one one eight and one three zero 1 1 [Music] 1 and 120 so 1 121 1 300 and 1 121 so basically these three these three here right there so measuring these three only the middle one does not match to either one of them about above or uh or below so that tells me that there's error on this data line and therefore this chip is bad so we're going to not going to bother with reballing and trying this chip so i'm just going to insta uh i'm going to re-boil another chip which i think is working and we'll take it from there do it's [Music] so okay go into a microscope so the challenge with this method is as soon as you start warming things up balls start to move around and that's because of the flux and surface tension that it applies so you can see some of the balls are already starting to move so i would have to manually carefully put them back where they belong or at least move them out of the way and then of course the key here is to heat it up slowly and then the most important element of this is the actual flux application if you apply too much the surface tension of the flux will pull the uh pulse together and if you apply too little the the minimum airflow that you set on your hot air gas station will simply blow them off so you need to you still need to have some plugs on there almost almost dry that's why you saw me kind of tapping i would wipe my finger and then i would wipe my finger dry and then i would kind of go tap um on the on the chip so that way it's all uniform and then slowly heating it up and you will eventually you will start seeing these bowls uh that they will start sticking to the pads so once you see your balls starting to move um but they don't align completely all the way at that point you can add a little bit of flux to help them move because they're already attached to the path just not 100 so i would drop a couple and then that should help them move to their final destination you can kind of see both they're starting to get glossy and that's an indication that you're reaching the proper temperature and when they do get glossy they start to actually move around and the problem i'm having right now is that i have this large aluminum block that kind of acts as a heatsink so it takes a while for me to heat this thing up [Applause] so [Applause] so [Applause] [Applause] so okay so the card cooled down we put it back into our test test bench let's uh power it on see what happens again the uh the flash flash card with the with mats is on there and the monitor is already responding that's good and we have a picture we're just now going to let the mats finish its test see if everything went well i'm not seeing any errors on the screen and i think we're gonna pass good so now all that's left to do is to uh get this card assembled and run it in the firmware and hopefully everything works well [Music] [Music] [Music] [Music] [Music] and unfortunately the driver fails to fails to install and when it does i get artifacts all over the screen in which case it means that the gpu is probably dead and there's nothing else i can do about this card so that error was those weird lines they were caused by this chip this guy right there it's a u505 this is a phase controller it controls all of our phases for the gpu so after after i had replaced that chip i can fire up the test bench to show you that it actually works okay there we go okay so we're inside windows now i'm going to run gpu-z and we're going to look at a couple things so first and foremost we want to make sure that our clock displayed correctly and um this particular board is does not support pci express 3.0 so it says 2.0 that's normal so if your board does support 3.0 you want that to make you want to make sure that it does say that um important number here is the power consumption uh five six percent all the way up to 18 tdp depending on really nothing's going on right now on the screen but if something were to happen that tdp is gonna it's gonna go all the way up like um if i were to start a stress test this tdp value is gonna go really high and i can't really see anything here on the screen that's because because of the stupid hdmi switch so let me let me turn that let me turn around and we'll actually have to look [Applause] through the camera unfortunately so right there tdp right now is 60 51 99 so that tells me that the card is being utilized fully and uh the other thing to look at is that the memory clock and the gpu clock are right where they're supposed to be you don't want to see like a 400 megahertz stuck on any one of them that would indicate a problem so i'm going to fire up the memory test and what that will do it will load 95 percent of the memory so that way we can see that the memory is actually functioning properly there we go and then we can monitor two different temperatures here one is monitoring the temperature of the gpu and the other ones monitoring temperature at the hot spot or basically a temperature where the memory mosfet is located and we're just going to let it run for a while and see how well it performs i hope you guys learned something today i thank you for watching and be sure to subscribe and like this video to help me with the view count and i hope you find this video helpful and see you later thanks for watching bye

Channel: northwestrepair

Views: 47,182

Keywords: how to repair, how to fix, how to diagnose, how to identify, how to find out

Id: IBW5alE6K0w

Length: 27min 45sec (1665 seconds)

Published: Thu Jul 21 2022