Hi, I have a XFX RX580 8GB that shows problem at 400MHz memory speed, all the other memory speeds are ok.
I tested with "tserver" with memory speed set by "./agt -mem=...."
If I set the speed at 400 (or similar 462.5 or 487.5) the memory tests is FAIL and also if I afterwards test at any other speed without a reboot.
If I set the speed at 400 (or similar 462.5 or 487.5) and then I change the speed to any other value without doing any test, all the memory tests fail at any speed.
If instead after a reboot I set any other speed (I tried 300, 600, 900, 1000, 1125, 1250, 1375, 1500, 1625, 1750, 2000) the test is PASS
So the card has no problem with Furmark because it sets the memory speed at the max value of 2000 but it is unable to start for instance the Radeon software or the Unigine benchmark.

Does anyone know what the problem is and how to fix it? Thanks

the memory controller is a IR35201 and I can measure Vmem_ctr between 900mV at min load to 950mV at max load while in the BIOS it is written between 750mV@300MHz to 950mV@2000MHz
I know that with Polaris the minimum Vmem_ctr value has to be always less than the Vcore, but it isn't in my case so this could be the problem.
To check it I increased the Vcore at low frequencies to be at least 900mV (but not at 300MHz that cannot be changed and still has to remain at 750mV) but the card still doesn't work properly. I will try to change the BIOS to have always at least 900mV on Vcore

The VRMtool would have been usefull here but it is not compatible with the IR35201

Hi,

I have exactly the same problem but on CORE, with Asus Rx580, but the difference is that any speed above 400 crashes the card and all voltages go down but only 5V remains and LDO from 12V to 3.3V powering the DIGI+ PWM is present. I don't have idea what is going on. I don't have replacement PWM chip. But I can increase memory speed and the card passes tserver's memfa test. There is a problem with a core clock only. I have written a post some days ago here: viewtopic.php?f=48&t=118167
The bios is updated/reflashed to the one taken from techpowerup website and it is rather good now.

Both have the same in common, increasing frequency doesn't have effect on voltage and the PWM goes into defensive mode and shuts down everything. Power off is required. I will investigate it, maybe communication is bad between GPU and PWM, maybe PWM is bad or something similar, maybe a resistor somewhere is dad.

I already conducted tests on the scope. The duty cycle does not change at all when increasing core frequency and in one moment there is a total freeze.

Hi, is your PWM core and memory controller also a IR35201?

Marked as DIGI+ ASP1300 by Asus which is probably Infineon IR3567B but it is a factory programmed chip so I can't replace it and I don't have a donor.
But the behavior is exactly the same but on core, not the memory. I did not have time to investigate it further, I don't have PDF for IR3567B. Do you have one?

here is a partial datasheet of the A version, it should not be so much different to the B version
https://www.datasheetarchive.com/pdf/do ... erm=IR3567

the Voltages are set by the GPU via serial bus, why do you say it is pre-programmed and you cannot replace it? You can buy it

Does the voltage of the memory controller change? It is also regulated by the same IR3567B

I know exactly that it gets commands from the GPU on the serial bus. At the beginning it decodes 2-bit state of the line and clock and after handshake of powergood/powerok it starts to passively listen to the commands. The problem is that this chip IR3567 has a firmware and if you buy a blank it will not work and probably from a different donor neither. This firmware is responsible for setting voltages on specific phases with a certain gain/aggressiveness and many other features. It is secret to the manufacturer so we need right to repair movement.

In my case it sets initial voltage to 0.900V with a duty cycle of 7.8% on all phases and it stays there. This is only a partial datasheet. Phases don't change duty cycle when they start on default power-on values and nothing more happens. This bus goes to all other chips responsible for power, fans, even sometimes on HDMI on the board. This is a common bus for the power management.

Does the voltage of the memory controller change? Just to know if the communication with the GPU works

I will test communication and voltage on memory later I will come back to this card in a free time. I connected oscilloscope to the GPU power only and checked memory if the shape is ok, I will make more comprehensive tests, maybe even tonight. But I was observing the card through thermal cam and memories were hotter at 2000MHz and memory test run faster. But who knows Wait for more news

OK, I tested everything.
uP1540 is for RAM 1.5V, a separate VRM. Duty cycle increases between 300 and 2000MHz but voltage is constant 1.510-1.512V. Memories are hotter only during memfa test which passes of course without errors.

PWM changes also on the main 6 phase controller, commands are being received, clock and data arrive to the chip properly.
In my case GPU 3d test and switching voltage and core frequency kill power of the board without reason. I don't have replacement PWM controller now as a last chance.
It just disables itself.
Here is a movie:
https://gofile.io/d/Lt3bGU
Scope is in average(16 periods) mode, every VRM looks the same with this leading edge overshot. This is high frequency scope 500MHz so every spike is visible. This is measured on the inductor..
Maybe these leading overshots, spikes trigger some kind of feedback overvoltage protection. Anyway I have a second card with failed memory and I will compare it. It looks like a feedback is taken from each inductor and all transients look the same so it means maybe gate drivers? They are quite hot almost 70'C, all three.

the VRM supply is ok to be 1.5V fixed
4 phases of the IR3567B are cpu core the remaining 2 are memory controller (that is inside the GPU) between min ~750V to max ~950V
You should check it if it also doesn't work properly. Did you also try the VRM tool?

Memory controller VRM is on the left of the PCB. There are 3 of them, memory, memory controller and PEX. On the right there are 6 phases to the GPU.
This is exactly this PCB but 8GB :https://www.ixbt.com/video4/images/polaris20-6/asus-rx580-scan-front.jpg

This leading edge which is ringing is strange to me like very low phase margin. It can cause instability in supplying power. It happens the same on all 6 phases so it is not a single transistor. Single transistor or a power stage can be easily spotted on the scope.

The power disables itself to such point that I have to power off for 5 seconds to discharge the board. It is completely latched up.

ok, did you also try the VRM tool?

Do you mean from tserver package? Yes I increased voltage from 0.9 to 1V and back and then to 1.1 and back and it hangs during this operation.
This card will hang windows so I can only test it under linux. It gives normal picture, no artifacts. I don't need to use secondary Nvidia.
So this may be a problem with power instability, faulty controller or GPU behaves like that.

I mean this one,
https://www.overclock.net/threads/vrmto ... s.1605757/
it is compatible with your controller IR3567B but not with mine unfortunately IR35201
Be careful using it!

Ho, I didn't use it, thank you, but I can't boot into windows. This card immediately hangs computer with Windows and this is a problem. Maybe it will work with Wine under linux.

boot windows 10 without the RX580 and remove and delete the AMD driver (maybe using DDU), then reboot windows 10 with the RX580 but without internet so windows will use the Microsoft basic driver that should be enough for this purpose

I will gonna make it hehe I have a second identical card and must measure it first later today.
Here are points of a serial bus between GPU and PWM for controlling voltage. GPU sends commands to the controller here. Here there is schematic for such similar card with the same controller as mine.

---

Posted after 20 minutes 2 seconds:
In both cases, yours and mine it can be dead GPU anyway. People with practice say that this behavior is well known.
Sometimes when it happens I see only one phase running and a picture is normal. The second from the top works all other phases are down. This happens every some crashes.

yes I agree in my case it could be a damaged memory controller but the fact the the voltage supply of it doesn't go below 900mV makes me hoping that the issue is the power supply :)

ASP1300 is available on ALI, shouldn't they be preprogrammed, configured by a programmer? People say they don't work, only from a donor boards.
Posted after 7 minutes 49 seconds:


Hehe the same as me Who knows what is bad. It hangs to almost complete no voltage. I can't tell what happened because all voltages are off. Only +5V on the left corner and 3v3 LDO are working. All others are gone.
There is a chance that this chain of power enable logic gates and transistors is sensitive, something is on the edge of good and bad and it disables the main power chain. It must have been early, just after +5V rail, maybe 5V itself gets blackout when GPU starts... This is the first such half damaged card.

well all our controllers are on sales on ali, there should be a way to make them working, there is an IFX programming tool
https://www.mouser.de/ProductDetail/Inf ... Os0MIHoA==
and software to download
The programming tool is expensive but I suppose there should be a much cheaper equivalent on ali, maybe the CH341T

I have some information. I measured the second identical card. This second starts only with 1 phase, the second from the top. The PWM controller is only warm 34'C instead of 50 but only 1 phase is running so who knows, doubler is also colder but drives only one phase.
We could even obtain this tool but we don't know what to program. Is it a software or configuration only? Who knows.
What do friends Chinese say? Maybe they could program a chip in their shop for some bucks... There are so many these chips so it will be easier to pay rather then buy a toll for each family. We don't have enough expertise.

Hi, after so a long time I got another RX 580 with damaged GPU so I did the transplantation from the XFX RX 580 to the other one of another brand and the issue happens again, so the problem was definitely in the GPU itself. So the GPU works perfectly with the following memory speeds 300, 600, 900, 1000, 1125, 1250, 1375, 1500, 1625, 1750, 2000 MHz but it doesn't work with memory speed between 360 to ~590 MHz.
Since the memory speed configuration belongs to the memory controller it can be surely changed in the BIOS. So I need to modify the BIOS in order to avoid the memory spped between 360 to 590 Mhz, is there anyone that has experience on this?