Yet another Chinese RX 580 help request thread

[darkphoenix16]

Hey there. I am sure you are tired of seeing posts like these. I got a Chinese RX 580 (for free) on Facebook marketplace. The seller said that only one port worked and he wasn't sure if the card would ever work properly so it was given to me at no charge.

I was pretty excited because I consider myself somewhat savvy and thought I could figure it out. Alas, I think I don't have all the knowledge. I spent a long 8 hours on it, and here is what I found.

1. Both DVI and Display Port connections seem to work. The HDMI did not but I didn't put much effort into troubleshooting it.
2. The RAM chips are Elpida memory. Many of the ram chips are faded and you cannot see where they come from, but I assume they are all Elpida
3. The CPU has a code 31939 on it
4. Severe driver timeouts when using an AMD driver. With a fallback driver the card works "fine" although appeared to crash my entire PC when watching a youtube video
5. I have tried a few 2048SP vBIOS roms with Elpida memory support. They all "work" in the same way as above in #4
6. In GPU-Z it looks every typical EXCEPT that the GPU and Mem clocks report 0. Same as the temperature sensor, reports 0.
7. If I flash with a 570 rom then both GPU and Mem clocks report 0 as well as the DEFAULT gpu and memory clocks
8. Booting into linux with this card as the primary display device will cause what appears to be a crashing cycle. I cannot get input into the machine to check a console.

I have since removed the card from my system. Its a bit of work to put in as I have this big GPU in the way that I have to remove, along with this annoying support bracket, to use its PCI slot.

I guess my question is, has anyone seen GPU and Memory clocks report 0 before? I felt there was something wrong with my bios because when I put 570 on there both current and default clocks were 0 instead of just current clocks. Is there any point in me continuing to try and get this card working? That cooler is really cheap, the board looks cheap too. I'm looking for some cheap upgrades for my nephews for xmas and I thought this card would reduce those costs if I could get it working.

Thanks!

 

[LabRat 891]

In UEFI, force the slot it is in to Gen2 speeds.
^This has fixed several varieties of problems w/ these 'import' Polaris cards.

Also, you may want to investigate RdN.ID / AmernimeZone drivers.
(note: if you use a VRR/FreeSync display, you may have to do a quick regedit to get it detected and working)

You may have an incorrect vBIOS, yes. However, wrong/no drivers will report 0/0 for clocks in GPU-z, IME.

IMO:
1. Force PCIe Gen2
2. Try out 3rd Party Drivers (that widely-support non-standard and non-consumer-facing variants)
3. If still crashing and/or improper GPU-z reports, check out another vBIOS.

Edit:
I'm assuming it's an 8GB model.

AMD RX 580 VBIOS

8 GB GDDR5, 1180 MHz GPU, 1880 MHz Memory

www.techpowerup.com

AMD, RX 580 2048SP, AMD Radeon RX 580 2048SP, (1180 / 1880)

Download Now

Memory Support 8192 MB, GDDR5, Elpida EDW4032BABG

AMD RX 580 VBIOS

8 GB GDDR5, 1050 MHz GPU, 1850 MHz Memory

www.techpowerup.com

AMD, rx 580, AMD Radeon RX 580 2048SP, (1050 / 1850)

Download Now

Memory Support 8192 MB, GDDR5, Autodetect 8192 MB, GDDR5, Elpida EDW4032BABG 8192 MB, GDDR5, Samsung K4G41325FE

 

[darkphoenix16]

Hello. Thanks for the suggetions. I've got through all three steps and unfortunately no change. I also tried 2020 Adrenaline drivers. I get a code -43 in the device manager for the card. Windows just doesn't like something about it. I could try some more vBIOS roms but I'm not sure how to find them efficiently. The filter does not seem to give me any results. The best choice I had was using google search to find some roms but I'm not sure if thats how you guys find them.

I did buy some gigabyte 580 8gb cards off ebay just not for 38 CAD each. They have 120mm fans strapped on them so obviously mining cards. I plan to take the shroud off this chinese 580 and put it on one of them and leave the other in its ghetto state ... but would still like to get this 580 working for the fun of it

 

[VuurVOS]

@darkphoenix16
Can you make a photo of the full PCB?

1850 MHz is not a normal memory speed for a RX580 2048SP. Can you try the following bioses?

AMD RX 580 VBIOS

8 GB GDDR5, 1244 MHz GPU, 1750 MHz Memory

www.techpowerup.com

AMD RX 580 VBIOS

8 GB GDDR5, 1244 MHz GPU, 1750 MHz Memory

www.techpowerup.com

 

[LabRat 891]

Code 43 is often an indicator of bad hardware.
I have caused that error w/o a "real" hardware failure but, it was still technically a component-level fault. AFAIK, my case is unique to Vega(s) and how it may 'act' after a hard-crash.

This card doesn't have a vBIOS switch on it, does it?
'Could be wrong but, I 'think' a CSM/BIOS-only vBIOS on a UEFI mobo w/ CSM disabled (as it should be) may also cause a Code 43.

When convenient, some pics of the nekkid PCB would be helpful.


BTW, it makes me smile seeing others willing to spend a lil $$ on 'tinkering' and 'playing with' what-otherwise would be considered eWaste
If you especially enjoy this kind of tinkering, there's other finds out there too.
Mining/'Import''Polaris cards have better community support but you might want to look into re-using and modding Radeon Instinct cards, too.

 

[darkphoenix16]

I have these two previous photos, and this one of the back - I put the card back together since but can take it apart again if necessary

I did not see a bios switch on this particular card - seems a bit barebones for that but I will check again. Is there anything that the Rdn.Id driver installers can diagnose?

 

[VuurVOS]

Code 43 is often an indicator of bad hardware.
I have caused that error w/o a "real" hardware failure but, it was still technically a component-level fault. AFAIK, my case is unique to Vega(s) and how it may 'act' after a hard-crash.

This card doesn't have a vBIOS switch on it, does it?
'Could be wrong but, I 'think' a CSM/BIOS-only vBIOS on a UEFI mobo w/ CSM disabled (as it should be) may also cause a Code 43.

When convenient, some pics of the nekkid PCB would be helpful.


BTW, it makes me smile seeing others willing to spend a lil $$ on 'tinkering' and 'playing with' what-otherwise would be considered eWaste
If you especially enjoy this kind of tinkering, there's other finds out there too.
Mining/'Import''Polaris cards have better community support but you might want to look into re-using and modding Radeon Instinct cards, too.

Code 43 is also when the bios is modded

@darkphoenix16 Can you please try the bioses which I posted earlier?

 

[darkphoenix16]

@VuurVOS - Yup! I am planning on trying those bios links you provided. I was kid-occupied last night so this evening I will flash both and let you know how it went. How do you find these by the way?

 

[VuurVOS]

@VuurVOS - Yup! I am planning on trying those bios links you provided. I was kid-occupied last night so this evening I will flash both and let you know how it went. How do you find these by the way?

Google since there is no searching function on boot message strings in the Video BIOS Collection database

Searching on boot message string "D00033 Polaris20 XL A1" and then some other things like "1750 MHz", "8 GB", EDW4032BABG
https://www.google.com/search?q="D00033+Polaris20+XL+A1"+"1750+MHz"+"8+GB"+EDW4032BABG+site:techpowerup.com/vgabios/

If those two vbios dont work, we need to check with D00034 instead since the board looks a little different.

 

[darkphoenix16]

Ok - Update

With the 247845 rom the device is reported as working in Device Manager. There was no change with the 253554 rom.

I knew there was a change prior to looking at the device manager because the experience reverted to a stuttery mess. So basically I see less than a second of activity, then the display locks up, blacks out, and then comes back a second or two later. Rinse and repeat. On this particular Windows 11 machine I eventually get a BSOD prior to being able to capture any information.

I am still using the 2020 Adrenaline drivers I had previously.

I am able to capture information now because I am running off of my integrated graphics with the RX580 2048SP connected. Here is what I see

 

[VuurVOS]

Ok - Update

With the 247845 rom the device is reported as working in Device Manager. There was no change with the 253554 rom.

I knew there was a change prior to looking at the device manager because the experience reverted to a stuttery mess. So basically I see less than a second of activity, then the display locks up, blacks out, and then comes back a second or two later. Rinse and repeat. On this particular Windows 11 machine I eventually get a BSOD prior to being able to capture any information.

I am still using the 2020 Adrenaline drivers I had previously.

I am able to capture information now because I am running off of my integrated graphics with the RX580 2048SP connected. Here is what I see

View attachment 318159
View attachment 318160

View attachment 318162

View attachment 318163

View attachment 318164

The GPU and memory clock are no longer 0 in GPU-Z so that is a good sign. Biosses with boot message "D00033 etc" in it are the ones you need likely need.

Can you use DDU (https://www.wagnardsoft.com/forums/viewtopic.php?t=4699) and then install Adrenalin version 23.8.2 or 23.9.1?

 

[darkphoenix16]

Here is what I have with an incomplete Rdn.Id install. I tried this first because I already had it downloaded. Not sure which 23 version I was using, the beta release. I say incomplete because it would not finish during the driver install. I had to restart the PC manually. That said, windows is apparently loading an AMD driver for the card and there are no display issues, but 3D is not working. I'll do another DDU and try a vanilla installer and report back.

Note, I still have integrated graphics enabled but my primary display is attached to the rx580. Maybe furmark is using the integrated gpu ... not sure. There are no "run with ..." options so I guess its using discrete card

 

[VuurVOS]

Here is what I have with an incomplete Rdn.Id install. I tried this first because I already had it downloaded. Not sure which 23 version I was using, the beta release. I say incomplete because it would not finish during the driver install. I had to restart the PC manually. That said, windows is apparently loading an AMD driver for the card and there are no display issues, but 3D is not working. I'll do another DDU and try a vanilla installer and report back.

Note, I still have integrated graphics enabled but my primary display is attached to the rx580. Maybe furmark is using the integrated gpu ... not sure. There are no "run with ..." options so I guess its using discrete card

View attachment 318169
View attachment 318170
View attachment 318168

I never see a "AMD Radeon RX 660" before. Can you tell us more about the system specifications?

 

[darkphoenix16]

Hey sorry! I captured the integrated graphics device there by accident (from a 7950x processor). That is what it shows up with 2020 drivers. With the 23.9.3 drivers it shows up as Radeon(TM).

Ok, so my report with 23.9.3 drivers is that its the same driver timeout loop. The same visual/freeze characteristics as previously reported but the driver seems to catch the error and ask me if I want to submit a report. This happens as soon as I connect the primary display to the RX 580 card.

I did not try the specific two driver packages you mentioned but I can this evening.

I also tried it with PCIe 2.0 mode and there was no change. I am not sure what causes a driver timeout.

Does this seem like something that can be fixed with another rom? This is basically the behaviour I had with the stock/factory vBIOS so maybe the card is just busted and was not tested properly at factory.

 

[VuurVOS]

Hey sorry! I captured the integrated graphics device there by accident (from a 7950x processor). That is what it shows up with 2020 drivers. With the 23.9.3 drivers it shows up as Radeon(TM).

Ok, so my report with 23.9.3 drivers is that its the same driver timeout loop. The same visual/freeze characteristics as previously reported but the driver seems to catch the error and ask me if I want to submit a report. This happens as soon as I connect the primary display to the RX 580 card.

I have some bad experiance with the newer drivers since AMD split the driver package into RDNA and Polaris & Vega. Please use 23.9.1.

Does this seem like something that can be fixed with another rom? This is basically the behaviour I had with the stock/factory vBIOS so maybe the card is just busted and was not tested properly at factory.

The card can be busted or it is a wrong rom. Its likely busted. Do you see a sticker on backside stating it is a 8GB or 4GB model?

Lets try an even slower clocked bios:

Mingzhou RX 580 VBIOS

8 GB GDDR5, 1206 MHz GPU, 1500 MHz Memory

www.techpowerup.com

 

[darkphoenix16]

Unfortunately the same performance / driver timeout issue with 23.9.1 drivers, before and after the rom provided. I did have gpu-z up and running prior to switch to the rx 580 as the primary display device. I saw activity on the core, up to about 800 Mhz. Maybe the core is bad.

 

[VuurVOS]

Could be the VRM is not working properly and not delevering the power it need. It seems this card is busted

 

[darkphoenix16]

oh well. thank you for your time - it was fun trying to get it to work. I'll take the fan shroud off and use it on another card that has a fan zip-tied to it. I can keep this card as a display adapter if I ever need to return something for RMA

Maybe I will look into how to test a VRM and repair

 

[darkphoenix16]

Hey! So I got the two Gigabyte 580 8gb gpus in the mail today. They both work. I flashed one with what I assume is a retail bios and I left the other as is. The one I left as is seems to have a 90watt power limit but only loses 10-15 perfect performance from the card that now runs at 125 watt. I was getting BSOD in windows and have been trying these cards out in linux instead as my entire windows partition table was corrupted and that entire ssd is now reported as empty ... I am not sure if the BSOD I got was from adrenaline drivers not liking the mining bios or what. I ended up flashing through Linux

here is the flash report for one of the cards

 

[VuurVOS]

Hey! So I got the two Gigabyte 580 8gb gpus in the mail today. They both work. I flashed one with what I assume is a retail bios and I left the other as is. The one I left as is seems to have a 90watt power limit but only loses 10-15 perfect performance from the card that now runs at 125 watt. I was getting BSOD in windows and have been trying these cards out in linux instead as my entire windows partition table was corrupted and that entire ssd is now reported as empty ... I am not sure if the BSOD I got was from adrenaline drivers not liking the mining bios or what. I ended up flashing through Linux

here is the flash report for one of the cards

View attachment 318693

You flashed a different vbios to a graphics card without knowing which memory is actually on the PCB

GV-RX580GAMING-8GD/F10/058A = Hynix H5GC8H24MJR
GV-RX580GAMING-8GD/FY0/06C2 = Micron MT51J256M32HFB
GV-RX580GAMING-8GD/F91/068B = Samsung K4G80325FB

 

[darkphoenix16]

The way I wrote the message is confusing. Prior to flashing anything I got a BSOD in windows. I assumed the cards were not working. The BSOD was with the mining BIOS they had prior and the BSOD happened when I instealled amd drivers. It also lead to my entire SSD being wiped - must have been during the "Automatic Recovery" loop that windows was stuck in

Then I went into Linux, tried to pull card information, and flashed a retail bios as I assumed something was wrong with the cards. They both work (retail bios and mining bios) but the retail card consumes 50% more wattage for 10% performance gain. The clock is capped as 1215 or there abouts and the retain is capped at 1340 or there abouts

I can flash another bios on them, its not really an issue - but was just curious if there are any long term reports of these minin bios not working with official drivers as I am thinking I might leave them on since they have better power vs performance characterisitcs. The mem clock does not list itself as > 2000 in rocm-smi