2 Artifacting 2080 tis, Determining faulty GDDR6 modules?

[inspirashun]

I have 2 faulty 2080 tis on a known good bench.

Dell card boots 100% into artifact mode. Error 43 from win10. Installs drivers okay. GPUZ shows 0 mb GDDR6 but sensors in GPUZ show ~2500 MB in use. Nothing recognized in MSI Afterburner. No significant sensor data in HWinfo.

Rog Strix OC on cold start shows no artifacting and is recognized with 11GB in GPUZ and HWinfo picks up sensor data. When up to temp, card locks up pc and upon restart will artifact from bios and is recognized to the same extent as the dell card, as a 2080 ti in limp mode.

This screams memory failure to me. Any thoughts?

 

[mtcn77]

You could try rewriting the memory bootstrap. Generally that is what I would inquire apart from a resolder.

 

[Athlonite]

Why bother shouldn't these cards still be covered by manufacturer's warranty so why not just RMA them both for good cards in return. Unless you really want to faff around with them

 

[inspirashun]

They were both purchased as parts for an educational purpose. The Dell has bent fins on the blower heat sink, I don’t see an RMA possibility. The ASUS will be going in for RMA.

I will try bios for the Dell to see what comes of it. If that doesn’t solve it, how would one go about determining the bad individual modules? I would hate to heat, remove and replace all 11 in one go.

 

[Athlonite]

They were both purchased as parts for an educational purpose. The Dell has bent fins on the blower heat sink, I don’t see an RMA possibility. The ASUS will be going in for RMA.

I will try bios for the Dell to see what comes of it. If that doesn’t solve it, how would one go about determining the bad individual modules? I would hate to heat, remove and replace all 11 in one go.

with some million dollar test equipment but not having that you'll unfortunately have to do it to each one and then there's the risk that it's just a dead die and no amount of resoldering is going to fix it unless you have a dead donor card to riffle for parts

 

[Ferrum Master]

Freeze or heat them up. See if it repeats the problem count.

Rinse and repeat till you find the weak spot.

I would like add... if one part is faulty... great chance for another one... and you have two faulty cards already. A professional would insist to change them all.

 

[inspirashun]

Freeze or heat them up. See if it repeats the problem count.

Rinse and repeat till you find the weak spot.

I would like add... if one part is faulty... great chance for another one... and you have two faulty cards already. A professional would insist to change them all.

Freezing them would be an interesting approach. I have a Kraken G12 bracket that would allow access to the GDDR6 modules while the unit is live on the bench. I could use that method to determine which module(s) are drawing heat more rapidly, and if it is a significant difference perhaps those are the failed modules. From what I see on the FLIR images posted of the cards, this ram gets HOT.

Is there a better method than an upside down air duster to generate the freeze?

 

[eidairaman1]

Get white sticker from card backs or side of card heatsinks and gpu-z screenshots please.

And no gpu bios since Geforce 1000 and AMD Vega can be modified, they are digitally signed so they break each time. (Thanks china/miners )

Cross flashing is only method to resolve issues now or detuning using rivatuner, afterburner, trixx etc etc.

Artifacting can be timings/volts or a defect in ram chips or die itself. Always check the video cable too

 

[inspirashun]

Get white sticker from card backs or side of card heatsinks and gpu-z screenshots please.

And no gpu bios since Geforce 1000 and AMD Vega can be modified, they are digitally signed so they break each time. (Thanks china/miners )

Cross flashing is only method to resolve issues now or detuning using rivatuner, afterburner, trixx etc etc.

Artifacting can be timings/volts or a defect in ram chips or die itself. Always check the video cable too

I will get the pictures shortly.


*********
I was trying to PM you on this exact topic of BIOS editing. I have my answer now but here were my thoughts.

"To create a scenario, on scam cards from various websites, there are cards flashed with BIOS that report incorrect GDDR amounts, typically to a higher degree. I am wondering if the opposite is possible, can you set a GDDR to a lower amount and would the graphics card only use the lower allotted amount?

For the purpose of testing GDDR6 modules, could I create a bios for a card that for instance, has (8) 1GB GDDR6 modules, M1-M8. For diagnostic purposes, one configured BIOS to only use say M1-M4, another to test M5-M8, another to test M1 individually, etc. Does that seem like something that could be done, or would the BIOS not be able to configure on/off at a chip level? If an 8GB card is set to a 4GB capacity, can it be instructed to use 100% of only 4 chips instead of 50% of 8 chips via BIOS?"

DELL 2080TI

http://imgur.com/a/tDA1PIr

ASUS ROG STRIX OC 2080TI

http://imgur.com/a/KDzfvqF

The ASUS posted today cold and without artifact however it went black loading into Win10. I did take note that the performance BIOS initiated all 3 fans, while the quiet BIOS only initiated the 2 outer fans.

 

[EarthDog]

They were both purchased as parts for an educational purpose. The Dell has bent fins on the blower heat sink, I don’t see an RMA possibility. The ASUS will be going in for RMA.

I will try bios for the Dell to see what comes of it. If that doesn’t solve it, how would one go about determining the bad individual modules? I would hate to heat, remove and replace all 11 in one go.

with these cards, nothing was pinpointed, their issues. So we dont know know if it is. Memory controller, or ic thing. No real way to isolate things.

 

[inspirashun]

with these cards, nothing was pinpointed, their issues. So we dont know know if it is. Memory controller, or ic thing. No real way to isolate things.

I got as far as the power stages on the Dell card, the mosfets pins were almost under the chip so I had to order some thin angled probes to get readings. They should be in this week to resume getting readings across the PCB. Youre correct though, there is far too much on the table to isolate anything yet. This dead time has me brainstorming a pass/fail test of GDDR6 on card 100% known good with no other variable than GDDR6 modules.

 

[eidairaman1]

For guidance ask one of these 3 or all

Eli Tech

My instagram https://www.instagram.com/tailoveleven/

www.youtube.com

Zettabit. lab

www.youtube.com

Actually Hardcore Overclocking

www.youtube.com

 

[inspirashun]

Thanks for the leads. I think Eli has moved on to other projects, I will reach out to buildzoid and zettabit.

I also came across this gent here:

At around 10:30 he starts measurements with a multimeter and has a method that appears to diag a bad module by pulling off an inductor and uses a power supply to send 1.3V to the GDDR modules, one heats up more than others and he pulls it.

I unfortunately dont know Polish and there are no subtitles on the videos. Im not sure the cards initial problems and I dont know if he ends up completing the repair after the part 3 video.