Really makes you wonder, if a ROM can go bad really fast on a GPU, and reflashing fixes it, this may be a proper troubleshooting method on a card you really want to keep…
EEPROMs could get some leakage current on the erase pin. But that would just point at some other component beeing bad.
So far, I had one case where the EPROM forgot its programming in regular use and never had a case or a ROM going bad.
Yeah, but that would require thousands of hours to point out and would require probing each component, something no sane person would do. A reflash every once in a while sounds like a good band-aid solution for a old card.
Just an osciloscope and one or two multimeters. Probably less than 10 hours when the pcb isn´t too much of a mess.
Or just cut the erase pin
I just had a weird brain fart when I saw the topic title. For a split second I totally forgot that Jayz is a YouTuber and I was like “WTF is a multimillionaire rapper doing reflashing old GPU’s?”
Still not as bad as Jayztwocents getting tagged instead of JayZ on twitter because of something political related.
Wow whats with that silk screen its a mess!!!
dont want todo that. youll leave it floating and only make the solution worse XD
I have a GTX770 for this reason. The board isn’t warped, the components look good, the ASIC looks good, I just need a way to reflash it and its probably fine.
This is a total edge case problem and fix.
EEPROM’s going bad is not a common failure mode and if it does happen it hints at something much worse going on. During normal operation the EEPROM is only read once at boot before being shadowed (copied) into RAM for faster access. At no point is it ever written to. (Unless the Nvidia driver/Vendor tool tried to secretly install a fix for GPU’s with a bugged BIOS config) But I have never heard of Nvidia or a vendor tool trying that.
I do however know of GPU firmware rootkits.
To carry on.
These 64K and more commonly 128K EEPROM’s are typically rated for several hundred thousand full read operations and rather much fewer write operations.
So if you think you can prevent EEPROM corruption by periodically reflashing your GPU firmware you are dead wrong. If you’re lucky it will do nothing, in most cases it has every potential to make it worse and you will end up with stuck EEPROM bits that will no longer reset to ground (0) state.
You’re best option is to make one dump of your EEPROM then Sha256-sum it.
And just later compare it to the sha256-sum of the ROM dump you get with GPU-z (which is a shadow ROM copy).
And no this isn’t a standard troubleshooting tip. DON’T FUCK WITH YOUR EEPROMS!
This belongs well into the section of BIOS Modding and undefined behavior where
- GPU is actually still POST’ing and inits as a basic VGA device.
- You’re actually able to still access the EEPROM via the PCI I2C bus.
- The Driver is clearly unable to setup a normal PCI register state.
- Most importantly! You actually know what you are doing and aren’t just following a troubleshooting list.
I’ve been doing GPU and Mainboard EEPROM modding for years now and every week do I get messages on twitter from people that roasted their GPU by flashing it or trying to fix it by flashing it.
If you gonna mess with your GPU firmware be prepared to totally lose it and your mainboard if you mess up.
A bad flash or bad vcore/memory/pci-e bus settings (even from just an undocumented change in a hardware design/modding tool) can kill hardware in a real hurry.
You have been warned.
Nightmare fuel addendum:
GPU based Firmware embedded keyloggers and boot-rootkits are a thing. They’re able to interrupt the normal boot process, inject themselves in the middle before then continuing the boot process. Old BIOS and modern UEFI GPU’s can all be modded to enable that.