RTX 2080, atypical memory layout or schematic error?

[Burnedoutplacebo]

Working on an MSI 2070 that has mats/mods errors on memory D0. Tried re-flowing with no luck so likely to swap with another chip.

However, I noticed on a schematic for an MSI 2080 that the D0 is located prior to the D1 spot on the counter clockwise rotation.

Is this a schematic wrong or do some cards actually not follow the D1 then D0 layouts?

Another question if anyone can help, once I do figure out the correct chip, would it work swap this bad Micron memory with an equivalent Samsung chip rated for the same speed? Samsung is what I have on hand.

Thanks for any suggestions

[schlagbaum]

Re: RTX 2080, atypical memory layout or schematic error?

From my point of view the arrangement of the FB chips around the GPU has not changed. The picture shows the correct layout.

If you have Micron GDDR6 with date code prior 2019 at the board you should replace the memory chip. Reflow experiments with the old Micron's will be seldom successful.

In my case (an ASUS ROG STRIX 2070 O8G) I was replacing one memory after another. Every time a chip was replaced the next one was showing errors in MATS. So I replaced 6 chips for FB[B], FB[C] and FB[D]. FB[A] was not affected.

If you are in the same situation it would be a good idea to change the whole FB with chips from Samsung. Or you buy the cheap Micron's at aliexpress (as I did @Shenzen Hong Ming electronics co. LTD)...

[ktmmotocross]

Re: RTX 2080, atypical memory layout or schematic error?

check what revision of board u have

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Thanks for the help. The chips start with 8, so I think this means 18 and likely need replacing. Board revision is 373 v4.0.

As far as the chip arrangement, was following this image so still confused, but will study it some more.

If I swap a single Samsung chip for a single Micron chip, will it work or do I have to swap all the chips change straps to get it to function?

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

I removed what I suspected was the D0 chip, following the usual high low pattern a shown in the picture, and not the schematic.

The picture and typical high (1) and low (0) pattern is correct. I verified running mat/mods and indeed the D0 chip is missing. Not sure what is up with the schematic.

Next try will be utilize a Samsung chip to replace the now missing Micron chip; maybe it will work?

[ktmmotocross]

Re: RTX 2080, atypical memory layout or schematic error?

no it will not work. u must replace ALL chips and switch resistor to change to another brand.

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Thanks for explaining it won’t work to mix and match memory brands.

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

A bit of an update on this. I got a replacement micron chip from aliexpress and it worked, for a bit until another chip failed (A1).

I ordered a few replacement chips again but decided to play some with it and installed a samsung chip for the failed A1 chip. This did not work, so I moved the strap to select samsung memory. To my surprise, it then passed MATS/MODS. I thought micron chips can work on the samsung selection, so I tried loading windows, but it went to black screen.

Out of curiosity, I then moved the straps to select Hynix memory, and this is where things went off the rails. When trying to run mats for the 30 series cards and older (from Kings Overkill) it went to black screen. Moved the straps back to the original Micron, and still goes to black screen when trying to run mats for 30 series cards and older.

Tried to run mats as secondary GPU, and get an error message - "Found an NV device at domain 0 bus 1 device 0, but its memory cycle enable bit in PCI configuration space is turned off" - Anyone have a clue what this points to for diagnosing?

What I find really strange is that mats when run from the 20 series card and older menu item, works fine, still fails with about 20 or 30 errors (when using straps set for samsung), but does not go to black screen.

Scratching my head.

[schlagbaum]

Re: RTX 2080, atypical memory layout or schematic error?

The GDDR6 from the different manufacturers have the same functionalluty, in principle - but different timing configuration to be selected by the strap resistors at PCB. MATS tests the FB with a low frequency (e.g. about 400MHz) directly controlled by CPU with DMA per shared memory access mechanism over PCIe bus. So different timings have no differences in the test results. The frequency is to low...
A MODS memory test (e.g. 94 or 178) will run at full speed. For performing such tests its important to have selected the correct memory strap.

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Wow, this is interesting. Thanks for the insights. I don't know the commands for running MODS memory tests, so will have to find instructions for MODS.

I guess the memory timings for samsung allowed the micron memory to work at the low FB frequency in MATS, but then failed when loading windows. The micron timings, did not allow the samsung to function even at the low FB frequency in MATS.

Wish I could play with it more, but need to figure out what I broke when playing with straps as I no longer can run MATS (for 30 series cards) without it black screening no matter what straps I use.

Still confused why MATS for 20 series cards runs but not MATS for 30 series cards. Must be something related to "Found an NV device at domain 0 bus 1 device 0, but its memory cycle enable bit in PCI configuration space is turned off"

Again thanks for the help.

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Whelp, maybe the end of the road for this one. Must of done something stupid when playing with the straps.

This card card is not initializing. Displays picture until trying to load drivers. Code 43, GPUZ shows 0 MB for memory,. Replaced bios chip, triple checked all straps. All voltages and resistances seem normal.

Out of ideas for what to check next.

[Sandhyahad]

Re: RTX 2080, atypical memory layout or schematic error?

let check what revision of board

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

let check what revision of board

Thanks for the interest. The board revision is V373 Ver: 4.0. I have been working off of a board-view for V373 Ver: 1.0 and it seems very close. I don't have a direct schematic so using a Ventus RTX 2080 schematic. Not all that close for component labeling, but good for understanding functionality.

Out of desperation, I did re-flow the core with no effect, but no harm thankfully, and issue remains.

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Another thought, maybe related.

During all of this around the time it quit working, I did once have the card seated in the PCEI slot, and plugged in the 8 pin power connector, while the power was on, card fans went to 100% until i switched the power off. When turned back on, all appeared normal.

Did not think much of it at the time, but maybe this damaged something?

[Burnedoutplacebo]

Re: RTX 2080, atypical memory layout or schematic error?

Update on this. Turns out the whole "Found an NV device at domain 0 bus 1 device 0, but its memory cycle enable bit in PCI configuration space is turned off" was a problem with my motherboard or my CPU in my test rig. The thing fully died, and when I switched over to another junk rig I had lying around, problem is gone.

Have not fully fixed the card yet as I cant seem to get a good solder connection on my replacement chip, and another chip (B1) now is giving me error codes in mats.

Seems like the very first response to my initial query is spot on the mark. Multiple chips are failing one after another on me due to the crap allotment of chips that came on this card.

Will be be practicing my re-balling solder skills for awhile it seems.