• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

[SOLVED] HD4850 weird monitoring values after bios flash

neelrocker

New Member
Joined
Mar 16, 2022
Messages
1 (0.00/day)
Hi there,

After my 980ti died I wanted to try to use an old gainward hd4850 golden sample as temporary replacement.
I knew this card was having problems (artifacts then freeze) and I already tried to lower frequencies in the past (without success), but wanted to try that again.

My previous attempts to lower frequencies were using ati tray tool as the card was used under win7.
Now the card is run under win10 and ATT or Afterburner are not compatible so I wanted to do it through a bios mod using RBE.

Unfortunately I haven't noticed I was using a too old winflash (v2.6.7) to flash back the modded bios (under win10).

The result are weird monitoring values like:
- gpu 200Mhz (was set to 500 in modded bios) - default still seen @700
- memory 6394Mhz !!! (1000 in modded bios)- default still seen @1000
- memory size 4096MB (it's a 1024MB card)
- fan speed -1500000rpm
- temp 32797°C
- gpu usage 200%

Those values are taken from gpuz (v2.44.0), Rivatuner (v2.24), HWinfo (v6.02).
The values read in those monitoring softwares were correct before flashing the bios (the card is installed with driver v8.970.100.7000 - catalyst 13.1 under win10 x64).
Also before the flash I was able to manually set the fan duty in rivatuner and this doesn't work anymore now (the fan is active but at low RPM given the lack of noise).

Do you think flashing with this too old winflash has damaged the card or it may just be the monitoring readings that are wrong now for some reasons ?


Below is what I have done for modding the bios and also what I have done to try to put back the original bios:


BIOS MOD:

- first note the card is set to "disabled" in win10 device manager and no display is connected to it (I'm currently using intel iGPU).

- I have first made a backup of the original bios using winflash (v2.6.7). As the checksum was not corresponding I have made another backup using gpuz (checksum was correct) and I have used that one to create the modded version.

- the modded version (with RBE) consisted in setting all the 10 clock states to: GPU @500, mem @1000, voltage 1.044v (original was 1.046v which is not in the voltage list, and some states with no voltage set at all)

- this modded bios was flashed using winflash (v2.6.7), with the results described above.

At this time I have realized how old this winflash version was, so I have tried to flash the modded bios again using amdvbflash_win_3.31 but the weird values in monitoring softwares were still there.


ORIGINAL BIOS FLASHED BACK:

I have tried to flash back the original bios using several methods:

- amdvbflash_win_3.31, atiflash-4.17 (freeDOS bootable stick), and even winflash (v2.6.7) again.

I have tried to flash back:
- the original backup made with GPUz
- the original bios for this card from the techpowerup database (exact same size than my gpuz backup. I have not compared both yet using hexcomp to check they are 100% the same)
- the original backup made with winflash (v2.6.7) (it has a slighly bigger filesize, 64KB instead of 62KB).


What do you think I could try now ?

I have not activated the card back in win10 device manager (still disabled) and have not plugged any display neither.
If I save the bios again (after flashing to original) the values are good when editing this file in RBE.
So the card should currently have the correct values with the loaded bios, but the values in the monitoring softwares are still the weird ones, and I still cannot manually set the fan duty anymore with rivatuner.

I have not played with the forced flash parameters in atiflash (-f, -fa, -fp) nor forcing the flashing algorithm (by default it has used M25P05/c)
Could the weird monitoring readings have to do with the win10 gpu drivers ?
Should I reinstall them ?
Should I try to activate the card back in win10 device manager regardless the weird values in the monitoring softwares ? (3 different softwares see the same values for gpu/mem clocks)

----- EDIT -----

Just managed to solve it.


- I have first tried to flash the original bios using amdvbflash.exe (command line ran as admin). I have used the -fa parameter to force flash (this is to bypass the "already programmed" when trying to flash with a bios similar to the one already loaded onto the board). Flash worked but issue was still the same. BTW the flash was done with the GPU power cable (8 pins) disconnected.

- then cleaned the AMD/ATI driver (DDU safe mode). The card was then detected as standard GPU (microsoft win10 driver) but the issue was still the same in the monitoring softwares.

- tried a bios flash again with amdvbflash.exe -> issue still there.

- then reinstalled AMD/ATI driver (catalyst 13.1) and immediately heard the GPU fan accelerating a bit, which was a very good sign. And indeed everything was back to normal in the monitoring softwares (gpuz, rivatuner), clocks are correct, all other datas are back to normal (memory size, and many other that were 0 or "unknown", like ROPs/TMUs, shaders, pixel fillrate etc), and I can manually set fan speed again with rivatuner.

So this was only a driver issue it seems. I'm going to try to mod the bios again to try to achieve the downclock I want to do.
 
Last edited:
Top