AMD Polaris GPU Diagnosing Guide

From Repair Wiki
Jump to navigation Jump to search

This guide is applicable for most Polaris cards from 460 to 590, some vendors may create different PCBs or use different components but the general working principles for all of them should be the same unless specified. This guide uses a reference RX 480 as an example.

The Card Layout[edit | edit source]

RX480 reference board front view (Figure 1)
RX480 reference board back view (Figure 2)

*PCB Image courtesy of TechPowerUp*

Before doing anything, it's a good idea to inspect the card for physical damage. Especially cards that have no backplate. They can easily lose some components on the back due to poor handling.

After making sure there is no physical damage to the card itself you can now move on with a multimeter to check the resistances of the voltage rails.

Step 1: Base Voltage Rails (12V, 3.3V)[edit | edit source]

The base voltages are the ones that get supplied to the card through the motherboard and the external 8pin power connector(s). What are the Base Voltage rails for GPUs?

12V rails[edit | edit source]

The card gets supplied 12v through the PCIe slot and additional 6-8 pin connector(s)

Start by measuring the resistances of the 12V rail coming from the PCIe slot (first 3 pins, see figure 1).

After that, measure each inductor for external power connectors (some cards have multiple external power connector and each of them have their own inductor you have to measure each of them individually).

The resistance varies from card to card and the value itself doesn't matter but it should be in the thousands+ Ω range.

3.3V rail[edit | edit source]

The card gets 3.3V from the PCIe slot only, from the 4th pin going left from the PCIe key notch in the front, and 2nd and 3rd pins on the back going from the notch again. (Figure 1 and 2)

If you get less than 50Ω on one or multiple base rails then you have a card with a short. The computer might not turn on in such a case because the power supply is protecting itself using OCP (over current protection). Repair Guide: Base Voltage Rail Short on Polaris GPUs

Otherwise, if you have no short then you can continue troubleshooting.

Step 2: Minor Voltage rails. (5V, 1.8V, Display Rail, VMem, VCore, and VDDCI)[edit | edit source]

Minor voltage rails are the ones created by the card itself using the base rails through either Linear Voltage Regulators or Step Down Buck Converter

Check the resistance of the output of those rails and compare them with Figure 1 and 2. VCore has such a low resistance that you won't find it useful to measure its resistance. A more helpful way is to measure its resistance against the 12v rails not GND.

If you get lower resistance on one or more of those rails, head to their pages linked below.

Otherwise, continue with the guide.

Step 3: Powering on the card[edit | edit source]

Assuming you have no shorts anywhere you can go ahead and plug the card into the motherboard and start testing (Alternatively, you can use a Lab Bench Power Supply and a riser to test the card with. Safer for the motherboard and gives you more freedom to move the card around and you get to know the current draw of the card if there is a short).

Switch your multimeter to DC Voltage mode and start by measuring the base rails first, if they are present continue to the minor rails.

Minor rails turn on in series, if one doesn't start, the next ones in series will not turn on.

Power sequence[edit | edit source]

The order in which they turn on in most polaris GPUs is as follows:

5V→ 1.8V→ Display Rail → VMem/VDDCI → VCore. (In some cases, VDDCI turns on after VMem, while in other cases VDDCI and Vcore are controlled by the same controller and they turn on at the same time)

For example, if 5v does not turn on, everything else in the chain won't turn on either. Hence no fan spin if you have a problem with 5V or 1.8V.

If you're missing one of them, check their respective pages for more details on how they work and their potential problems:

Step 4: No Video Out[edit | edit source]

Everything is present but still no video out? You may have faulty Memory, BIOS, PERSTB, Crystal oscillator, GPU chip itself, or in some cases a problem with straps.

Memory problems[edit | edit source]

If you've reached this point, the most likely culprit is the Memory. You can confirm this by testing the memory in Linux. Here: AMD Memory Testing Guide. is how to detect faulty memory chips.

BIOS problems[edit | edit source]

If the memory is okay or the card is not even being detected in Linux then the problem is highly likely to be the bios. Check: BIOS Problems on Polaris GPUs.

Crystal Oscillator[edit | edit source]

Often marked with Y followed by a number, crystal oscillators sometimes fail which will lead to the card not booting up.

In most if not all Pascal GPUs, the frequency of the oscillator is 27MHZ. An oscilloscope or a multimeter with Hz function that can go above 27MHZ is needed to test it.

PERSTB signal[edit | edit source]

PERSTb signal schematic view. (Figure 3)


U100 is a multiple function gate, the output of which is determined by the 3 inputs shown in figure 3. Since input A is always high (direct connection to 3.3V rail) according to the datasheet, the output will go high only if the other 2 inputs (B & C) are also high.

Sometimes this gate can fail which results in the output Y "PERSTb_BUF" signal not getting generated and prevent the card from working.

Start by checking all inputs, if one is missing trace it according to the schematic in figure 3 and replace the faulty components. If all the inputs are present but the output is low, check the Vcc pin and make sure the gate is powered, after that, replace it if all inputs and power is present on the gate.

If everything else is working as they should but still no video out then unfortunately you have a faulty GPU core. Best use for that card is as spare parts since getting hold of a GPU chip by itself is very hard and expensive and replacing it is a very advanced procedure that requires a BGA rework station and it's out of reach for many people.

Step 5: GPU outputs a picture[edit | edit source]

perhaps the card does output a picture but it is not working properly, here are the common problems and their potential fixes.

Artifacting[edit | edit source]

Artifacting is most often caused by memory problems, check AMD Memory Testing Guide

if you do not get memory errors even after 100+MB test in MATS then the core is very likely to be the issue.

Error 43[edit | edit source]

just like artifacting, error 43 can be caused by faulty memory or core but also BIOS and straps.

Start by making sure the memory is fine as shown in the guide above, then check if the BIOS is not corrupted/modded (flash original bios from either TPU BIOS Library or manufacturer's site) and check the bios circuit as shown here: BIOS Problems on Polaris GPUs.

After that if the problem persists, check the strap resistors, they can either get knocked off or change in value which will trigger error 43.

If everything is fine but the error persists then the core itself is damaged.