Kreyren/kreyren

AMD Vega 56 - GV-RXVEGA58GAMING OC-8GD rev 1.0, 4/E/LH1702 00842 V17149

Opened this issue · 1 comments

Kreyren commented

I have AMD Vega 56 (GV-RXVEGA56GAMING OC-8GD rev 1.0) that doesn't have signal nor fan spin on system startup.

https://www.igorslab.de/en/custom-rx-vega56-benchmarks-review-2/2/

image

image
^ opamp (LM358 SMDR)

image
^ VBIOS location
vbios
^ VBIOS - AH172 25Q41BT 79172 -> GigaDevice
GD25Q40CTIGR (https://datasheetspdf.com/datasheet/GD25Q40.html)?

Initial diagnostics

GPU connected to a healthy system results in no fan spin nor display on startup.

Hypothesis

Short to ground preventing the VBIOS boot.

Diagnostics

Seems like bricked VBIOS

Cooler

The cooler is fine i connected it to 9V battery and the fans happily spin.

Connecting the GPU without a cooler to a system doesn't give me a display.

Conclusion: The cooler is not causing the short

PCB

I checked all the "bigger components" for conductivity and i couldn't find anything faulty.

Requires more diagnostics for other components, but i don't know what are the expected voltages and the boot sequence (help-wanted)

Discovered manufacturing defect on the highlighted area
image

My hot air was set on 260C, but it seems that something broke inside it and set it to 400+ that damaged the PCB.

This damage goes 2 layers deep and exposes one of the board traces, for permanent fix this has to be covered with insulation and create new traces with pads for the pins.

Alternatively, the reference AMD RX Vega 56 SEEMS TO have vBIOS on the back side:
image

Which is same place as the Gigabyte AMD Vega 56 that is not populated. When testing the pin 8 (vDD) doesn't have the expected 3.3v which would point to board design having two vBIOS locations where only one is used, pressumably to make the second location work the board needs reverse-engineering to know where to add the components to establish the connection.

Temporarely this has been solved through jumper wires on traces.

PWM (35217-C704P-AV8H)

Seems to be custom-made for this GPU

On IR35217:
Pin#24 - SDA
Pin#25 - SCL
[https://elmorlabs.com/index.php/forum/topic/5500xt-ir35217/#postid-173]

Probing from 12V in to VDDCI inductor gets me 0.22 ohms

vBIOS

Pin 8 on vBIOS has 3.36V on power up plugged in PCIe and to PSU 12V2A rail -> vBIOS has power and the main rail is working.

Checking pin 8 on the VBIOS while the GPU is in PCIe + PSU on a system that boots gets me 3.36 VDC -> VBIOS is getting power.
image

Getting 0.0VDC on the bottom of these mosfets (other side reports 23.03 +- 0.05 VDC) -> Assuming corrupted VBIOS
image

Documentation: http://ww1.microchip.com/downloads/en/devicedoc/s71271_04.pdf

image

Decided wiring:
ground - black
power supply - red
Serial Data input - white
serial data output - orange
serial clock - green
Chip enable - purple
write protect - yellow
hold - brown

^ Standard for this coloring doesn't exists, developed what i find to be the most sane inspired by other coloring codes.

The on-board flashing intended to be kept permanently on the PCB didn't work as the PCB is not optimized for this flashing and introduces noise.
image

Debugger reports the VBIOS as 512kB while the known AMD RX Vega 56 VBIOS size is 256kB.

Found GigaDevice flash chip "GD25Q40(B)" (512 kB, SPI) on ch341a_spi.
Error: Image size (262144 B) doesn't match the flash chip's size (524288 B)!

image

Requested brainstorm in flashrom/flashrom#196 - Resolved

Chip has been flashed with VBIOS 016.001.001.000.033736 (https://www.techpowerup.com/vgabios/214191/gigabyte-rxvega56-8192-190723).

Now i am getting fan spin with mosfets still switched off and 0.9V rail shooting fireworks. -> Excluded issue with vBIOS.

Common failure?

Found https://forums.tomshardware.com/threads/vega-64-no-display-signal.3471991/ and simmilar articles all over the web -> Is this common failure mode of this GPU series?

Unlikely to be the case outside of windows failure.

Relevants

  1. Benchmark review https://www.igorslab.de/en/custom-rx-vega56-benchmarks-review-2/2/
  2. The PWM is 35217-C704P-AV8H
  3. PWM on octoparts https://octopart.com/ir35217mtrpbf-infineon-74192668
  4. PWM code decryption (assumed)
    image
  5. PWM documentation for 35201 (not same as the used in GPU) https://www.infineon.com/dgdl/Infineon-IR35201MTRPBF-DS-v01_00-EN.pdf?fileId=5546d462576f347501579c95d19772b5
  6. Relevant? https://elmorlabs.com/index.php/forum/topic/5500xt-ir35217/g
  7. 25 pair color code https://en.wikipedia.org/wiki/25-pair_color_code
  8. VBIOS database for this GPU https://www.techpowerup.com/vgabios/?manufacturer=Gigabyte&model=RX+Vega+56
  9. VBIOS info from manufacturer https://www.gigadevice.com/flash-memory/gd25q40c/ OR https://www.gigadevice.com/products/memory/flash/spi-nor/3-0-volt/gd25q-family/gd25q64e/
Kreyren commented

Fixing the VBIOS

  • Order new VBIOS chip, because it got damaged during the PCB overheating event.
  • Fix the damage that some retardrey have caused to the PCB.

TODO

  • Figure out the replacement chip