Hello everyone, I have a 3070 and a 3070 ti (both MSI Gaming X Trio) with strange problems that I have never encountered, having never worked before on 30 series cards I am a bit lost, both cards have normal resistances.

the 3070 Ti has all the voltages and is detected, but there is no image, I tried to use mats 455 with a secondary card, but when I use the command "skip_rm_state_init" an unusual error appears or "Memory addres training failed" and when I launch mats all banks of memory show errors in reading, I tried to flash the bios I also tried to change the chip of the bios but nothing has changed

the 3070 has the core voltage of 0.400v consequently the memory and PEX voltages are missing, the strange thing is that the voltage rises to the normal 0.750 (and the other 2 voltages are also present) but only for a second if the card is connected to the pci-e, if I connect the card through a raiser without connecting the riser to the pcie of the motherboard the voltage remains at 0.750v for a little bit longer.

has anyone ever encountered such problems? thanks in advance your help,

I am attaching the mods screenshot and mats results of the 3070 Ti.
let me know a translation in russian can be useful even if through gogle translate.

Best regards
Roschach

Hello, nice to see a new member, welcome to the best forum in the world

3070 Ti black screen with so many errors in all banks suggests the GPU failure. Probably broken contacts between crystal and substrate. It looks like a major failure. It is highly unlikely for all memory chips to fail at one.

I am planning to test 30-40 cards in my trash with similar symptoms. I plan to make some attempts to repair it. First I plan to drill the glue in 2 opposite corners and pour some flux and reflow. Who knows what will happen, but these cards are already dead but all voltages and resistances are good.

The problem is that the glue can expand during heating, during reflow and small balls below crystal may not touch each other despite of applying flux. Removing the whole chip for this procedure is also not good idea. You may not solder it back properly. I am building a spring pushing the crystal to the substrate without removing the whole GPU from the card. Spring will be mounted on the GPU's substrate corners. During reflow the glue becomes soft and a small push may reduce the gap by 0.01-0.02 mm and it would be enough to repair joints. This glue is softer but not soft enough to collapse in 240'C and should not pop up balls. Definitely 0.01mm should do the job.

I am planning such tool based on the principle from the attached image, strapped on four corners of the gpu. Sometimes reflow helps sometimes not but with a little help it may work.

What do you think ladies and gentlemen?

---

Posted after 25 minutes 49 seconds:
About 3070.

After successful initiation of the card the GPU receives POWERGOOD signal on one its GPIO and then it asserts PWROK to the PWM controller. From this point GPU will send commends on the serial bus to the PWM controller. GPU will request certain voltages.

1. GPU may send wrong commands or
2. Commands are wrongly interpreted by PWM or
3. Commands are properly interpreted by PWM controller but the PWM's reference voltage or measurement circuit are broken.

You need to dig in this area. There are several voltage dividers, voltage sensing circuits, check them.

Thanks! i have already learned a lot from this forum by reading the experiences of other members, i just noticed that the pcb of the 3070ti is quite bent so i will try the reflow in the hope that there is a problem with the tin balls under the chip.

the reflow has failed the problem remains so I suppose the GPU is gone, maybe in the future I will try the method you suggested.

as for the 3070 i checked the main voltages of the pwm controller they seem correct, i quickly tried to change the controller with that of the 3070ti (the circuit is the same) but the problem remains, i'll dig deeper comparing every single voltage with those of the 3070ti

Yes it may work as it works at 50/50 cases. I have universal profile 110->150->200->235->245->255 and I stop it manually when I tough the chip and I feel it is floating then I press the red button. Usually the chip floats in 235 area but after 20 seconds or 40 seconds, depending on mass. Maybe Vega needs zone 245'C for some seconds. But definitely I control the process manually and I apply only so much heat as is needed plus some seconds.
Do you have thermal camera? There can be a capacitor on the edge of the GPU itself, this very tiny around crystal. Once customer replaced paste and destroyed this cap, broke it a little. It was not knocked out. Easy repair, some euros almost for no effort I always first make measurements if the card is safe to power up, then PCIe pins integrity and next thermal cam when I can't find something obvious.

The second card may be challenging. It can be PWM chip but try to find PDF and measure whatever you can deduct. We can discuss your case when you supply more information.

yes I have a Flir One pro LT, not particularly good I know, however with the thermal camera I did not notice anything relevant, I checked both cards, and I had already checked all the data pins and also the capacitors on the back of the pci-e all normal

I have seek and a ZnSn lens from the laser cutter. It is very good. Lens is in a piece or rubber, self made.

Finally I had some time to work on the video cards, in the end after checking all the voltages of the 3070 without finding anything strange I decided to try a reballing but under the chip there were many missing pads (I attach the photo) and I am sure I have removed it correctly as I have done dozens of times, I guess the 3070 TI is in similar condition too what a shame...

very strange. same thing happened to me today when I tried to reflow a 3080 Asus. It looks almost like yours. the PCB was also bent, I guess this is the problem.
Did the pads get damaged when you pulled the chip or by itself? Mine started to "dance" when it reached 225 degrees. The "dance" brake the pads.

are we doing something wrong? this never happened on 1000-2000 series.

---

Posted after 8 minutes 58 seconds:


are you sure connection between crystal and substrate is also made by balls? then maybe vega 64 can indeed be repaired when having memory problems...
can't wait to destroy a chipped gpu tomorrow :)

I think the pads were damaged before the reflow due to the bent pcb, probably the reflow has only made the problem visible

Rorschach315
If you have the desire, the equipment and the necessary skills, this is not a problem:

where can i buy the pads? I can't find them anywhere, what are they called in English?

nevermind found them from china

Rorschach315
Here:
https://aliexpress.ru/item/100500251945 ... ec70&tt=MG
soldering lug

in my case I guess the bga station also did it's job, for large chips I keep the IR like in the left mark, 1cm from chip, because otherwise they don't heat enough.
I had no problems with 1080ti or 2080ti, but 3080 is larger.
I also think that the pads were broken, because the card came with some work marks in that lower left corner

Maybe you can test your machine also.

I do it from time to time but first I glue it by UV solder mask and then I solder it and again I cover exposed traces. I did it maybe 5 times under VRAM but there was 1 or or maybe two pads missing.

What I see under GPU in above posts is very, very strange. Pure soldering would not rip off so many pads. What force could cause it? Maybe the author did not heat the board enough from the bottom?
I have universal profile for lead and unleaded solder. There are steps to higher temperatures. I stop soldering when I feel the chip is floating for some seconds.
If you apply not enough heat from the bottom the tin will cure before you remove the chip.

in my case i am almost certain that the damage was prior to chip removal, i removed several large chips without any problems, recently removed the chip of a 3090 without damage, the two cards were both from the same vendor, both with the bent pcb, the 3070ti shows the same damage to the pads even if less severe, in the same lot there were 2 3060s both with damage to the backplate and damage to the memory near the pci-e connector, anyway strange cards... never opened, perhaps coming from some accident in a warehouse?

This is damn strange. We can repair some pads but not so many. maybe at least the GPU is in good condition.

Luckily many of the pads are not connected, as soon as the replacement pads arrive from china in about 15 days i will try to fix at least one, the problem is that i don't have a microscope so it will be hard, in the past i managed to fix some missing memory pads from the pcb with copper wire, I hope it is much easier with pads

Use UV solder mask before soldering. Clean it first, put a blob of solder mask, you can immerse a new pad in it. Use UV light. Next you can scratch the solder mask by surgical knife from the top. Solder wire, cover with solder mask again.

You don't need mats here. Mods on the screenshot already reported error status only on one of the total 4 memory channels. https://imgur.com/a/g2bVoWD

thanks for the tips guys.



indeed for that card the missing pads are from the memory bank A1, i think i can repair this one
now the long wait before the pads and the stencil arrives from china...

Good news! Today the stencil arrived and i didn't want to wait for the pads so i tried with the copper wire, and it worked! its alive!

Congratulations I wouldn't be so brave with wires I would wait for pads

I have the same errors on a gigabyte eagle RTX 3070 ti as yours.
I have the same errors as the guy in the first post.
but I have an error in one of the channels


Photo: https://imgur.com/a/g2bVoWD

Is there a stencil for the RTX 3070?

Hi everybody!

I have an RTX 3070 Ti. The mats run goods. But artifacts gpu.

The nvmt show faulty B0 and D1.
I changed are two memory, but not good again.
What is the broblem?
Maybe GPU?

Друг модешь посмотреть сопротивление 27pin шим контроллера питания gpu ncp81610 на msi rtx 3070ti