Здравствуйте!

Yesterday I repaired rx470 MSI V.341 2.1.
I replaced 2 vram chips 7 and 8 and the card started. I left it for mining and after 12 hours the card passed away.
Tserver after making memfa gives me an error:
[error][Framework] Exception during attempt to execute test: TcoreSiliconTarget: Failed to load ME microcode using toollib on Tonga/Ellesmere FF and both tests fail AK600.001 and this previous one.

Error 43 in Windows as a secondary card.
Dmgg gives all banks dead.

Probably another memory failed because I can flash the bios from windows but I can't diagnose it now.

Any idea?

check the VRAM voltage

It is almost 1.5V, core 0.9

Someone here solved this by GPU reballing

Someone in the private message sent me information that I should check the memory number 1. I will do it today or I will try to reflow the first two memories first.
This is hard to tell what to do when diagnostic software doesn't work.

hmmm.. maybe try vbios without memory autodetect (i.e. with one type of supported mem) instead?

Change your motherboard

Hi crocodile70

Change motherboard into what? I have older model with Nvidia chipset and Amd Athlon x 2, one PCI slot for the post card. Motherboard from Gigabyte but I don't remember the model now. The older one, I will tell you later...

Gigabytes are not the best choice for a test motherboard

Change your motherboard

Especially mats, mods

So what motherboard do you prefer?

H170 pro gaming Asus

m4a785td-v evo Asus


my tests motherboards

OK, I will explain

I had 4 cards with almost identical symptoms. All AMD 400/500 series. First card I treated by unsoldering vrams 1 and 2 as suggested and there was dirty soldering under them. I reballed them back and this card is mining again.

For all others I used IPA (FLUX OFF or similar) pressurized under vrams and I made a reflow of all eight vrams and these cards came back to life like they were never spoiled.

It means that this error "TcoreSiliconTarget: Failed to load ME microcode using toollib on..." likely means there is something wrong with vrams. Remember to clean under, bake for some time and reflow memories.

There is also a technique I developed for my test setup when the Linux loads up despite of a black screen. I put the card to the x16 slot and I log into it by SSH from a second computer. There is much higher chance that dmgg.py will show correct results.

Sometimes a card can oscillate do b2 and initialization restarts. It may be visible in Linux as a secondary but tests will not work, neither tserver nor python script. This is still the case of VRAM. Do not be afraid to clean under them and reflow them without reballing. When they are floating I tap them pushing down a little and it worked.

Thanks - very important information !

ну тут я уж чисто по приколу решал пошатать чипы, все равно менять не на что, эту то как раз брал на память, но без низа даже на 480 китайских градусах не смог, а греть больше 20 сек я чет не захотел, без низу.

но после остывания она заработала)

так что имейте ввиду load ME microcode - еще не приговор ГПУ.

From another topic - quote

viewtopic.php?f=48&t=100822&p=915841#p915841

Low resistances are more convincing as far as GPU is concerned. It may not always be the case too. 1.8V is used as a reference source for memories. This is in general very delicate matter. If you have some thoughts just share them. Don't forget about correct grammar and spelling because translator gets confused