AMD Radeon Pro v540 Research Thread

[Catch2223]

Seems the buy it now one is gone already

Damn, someone was quick!

 

[PixelButts]

Damn, someone was quick!

I took another look and it seems the auction is now ended. Seems like they only had 1 but made 2 listings. Why not just make a listing that also offers buy it now if you wanted? There's plenty of those.

 

[PixelButts]

I have good news and bad news. The good news is I'm the one that got the recent listing, which is actually why I finally hit you up on that fan offer @Catch2223 . The bad news is I'm the one that got the recent listing and I am definitely not the best person to have it.

In the mean time here's some closer photos of the backplate, serial differences, and such. Some components on the back seem absent or changed slightly. As for what that means can't really say. I'm currently setting up a simple Windows 10 install just to mess with it.

What I find interesting is the part numbers have changed slightly, as has the serial number range, so there's definitely more than one of these out there given this one is 127.

For testing this I thought it would be really funny if I also had my WX 8100 as the second GPU in it just to get video signal so enjoy a really weird photo of 2 unreleased GPUs in the same board.

 

[LabRat 891]

I have good news and bad news. The good news is I'm the one that got the recent listing, which is actually why I finally hit you up on that fan offer @Catch2223 . The bad news is I'm the one that got the recent listing and I am definitely not the best person to have it.

In the mean time here's some closer photos of the backplate, serial differences, and such. Some components on the back seem absent or changed slightly. As for what that means can't really say. I'm currently setting up a simple Windows 10 install just to mess with it.

What I find interesting is the part numbers have changed slightly, as has the serial number range, so there's definitely more than one of these out there given this one is 127.

For testing this I thought it would be really funny if I also had my WX 8100 as the second GPU in it just to get video signal so enjoy a really weird photo of 2 unreleased GPUs in the same board.

View attachment 340374View attachment 340375View attachment 340376
View attachment 340379
View attachment 340377

Made in Taiwan

After you've had your fun with it, I'd certainly be interested in buying/trading it from ya.

Regardless, I'm looking forward to more info from your experiments on this Production variety V540.

 

[PixelButts]

Made in Taiwan

After you've had your fun with it, I'd certainly be interested in buying/trading it from ya.

Regardless, I'm looking forward to more info from your experiments on this Production variety V540.

Well, to say I am making progress would be hard to say. Same issues as before, though I tried to dump the vbios from gpu-z with the AWS drivers installed and it refused to let that happen. But with the ones from the LTT forum it let me dump it which was certainly odd.

I do not know which one is which for the roms, so take the A-B with a grain of salt. They came out identical though so perhaps not worth looking at yet. The bios revision IS different though so that's a good start. 017.003.007.013559

 

[LabRat 891]

I tried to dump the vbios from gpu-z with the AWS drivers installed and it refused to let that happen. But with the ones from the LTT forum it let me dump it which was certainly odd.

IIRC, the LTT Forum drivers are from a Chinese firm.

It does kinda make sense for Amazon to request or modify the driver to prevent offloading of the vBIOS. From what I can gather, these cards (and Navi12) were more-less developed entirely for 'special partner use'.

Which, wouldn't be the first time that's happened:
The 1st gen Steam Deck uses an APU that was co-developed for an 'Industrial' AR/VR device.
There's a ton of inactive silicon in 1st gen Van Gogh, and completely missing from the die shrink in the OLED Steam Deck.



Has anyone managed to tryout the "bootcamp" driver options in R.ID's driverset?
They're now much-better labeled as "for AMD Radeon Pro 5600M"

 

[PixelButts]

Has anyone managed to tryout the "bootcamp" driver options in R.ID's driverset?
They're now much-better labeled as "for AMD Radeon Pro 5600M"
View attachment 340391

Gave it a try, and nothin. The Device ID doesnt even exist in the current set of R.ID drivers according to them so they have been alerted to this. I have offered to test whatever they want to get it working so we shall see.
Was also going to see if the linux solution we have works with Intel Arc first but the .3 ISO is gone now and kernel defaults to 6.5 not 6.2 and I kept breaking it entirely so I'll look into it another time.

In the mean time if anyone wants me to try something I'll gladly do it.

 

[mczkrt]

Well, to say I am making progress would be hard to say. Same issues as before, though I tried to dump the vbios from gpu-z with the AWS drivers installed and it refused to let that happen. But with the ones from the LTT forum it let me dump it which was certainly odd.

I do not know which one is which for the roms, so take the A-B with a grain of salt. They came out identical though so perhaps not worth looking at yet. The bios revision IS different though so that's a good start. 017.003.007.013559

These roms are only 256KB, rest is padded with 0, while other working dumps have a valid data past 256KB.

If you manage to get Linux working, you can try this:

cat /sys/kernel/debug/dri/{x}/amdgpu_vbios.rom > vbios.rom

where {x} is card id as reported by switcherooctl, for example.

AMD drivers should not be required for this, KMS driver in 6.5 might be enough.

 

[PixelButts]

Well the good news is there was some fast turnaround on the R.ID driver stuff. It is now added. Bad news is it's still returning an error code 12 but at least now more people can try stuff and not be dependent on the AWS drivers

These roms are only 256KB, rest is padded with 0, while other working dumps have a valid data past 256KB.

If you manage to get Linux working, you can try this:

cat /sys/kernel/debug/dri/{x}/amdgpu_vbios.rom > vbios.rom

where {x} is card id as reported by switcherooctl, for example.

AMD drivers should not be required for this, KMS driver in 6.5 might be enough.

Found an archived .3 ISO for the sake of consistency, and made a backup of my current setup for windows to reflash quickly as I swap between them.
I'm also sorry that this took so long I was not fully understanding what I kept breaking so I retried to set this up 4 times. Got it working finally. The result of this is that the rom files are both 40kb which is clearly not right. Do you have any other suggestions for dumping this? They're also identical, which we know is not correct, but I did dump each one.

Edit
I tried with amdvbflash and it only found 1 ID, which I assume is because its mentioned explicitly in grub? This came out to 1mb

 

[mczkrt]

Maybe check with amdvbflash 4.71

sudo amdvbflash -i

to list adapters.

sudo amdvbflash -s <id> vbios.rom

where <id> is adapter number from the list.

EDIT: I see you got this already.
File is padded to 1MB but looks otherwise complete, with data between 256KB and 512KB.

 

[PixelButts]

Maybe check with amdvbflash 4.71

sudo amdvbflash -i

to list adapters.

sudo amdvbflash -s <id> vbios.rom

where <id> is adapter number from the list.

That's the exact thing I did actually. The second one isn't being listed. 0b:00.0 and 0e:00.0 are their respective IDs currently

 

[mczkrt]

In my experience, without correct drivers 2nd adapter tends to crash fairly often if any process touches it.

register aperture not available

That's some clue, maybe there is corresponding error in dmesg?

 

[PixelButts]

In my experience, without correct drivers 2nd adapter tends to crash fairly often if any process touches it.


That's some clue, maybe there is corresponding error in dmesg?

I have no idea how to even check this. My linux game is pretty weak. I can also drop in the engineering sample and see if its any different too?

 

[mczkrt]

Reboot and run amdvbflash -i
then

sudo dmesg -H | grep amdgpu

look for timestamp when you ran amdvbflash.
It's possible it crashed during initialization.

 

[PixelButts]

No dice, same result, heres the dmesg output from that

 

[mczkrt]

Hmm, so it does not crash, just can't be accessed by flash tool. Weird.

Reading data from the rom, build date is 2020-01-30, while ES rom is 2019-12-17, so not that far apart.
PPT limits are identical.

 

[PixelButts]

I swapped the cards to just check if the same happens and sure enough they do match with amdvbflash, cant read the second one. though the p/n for the vbios is different on the ES card.

I'm open to more options if you have any ideas for dumping that second one's bios. The method with cat SHOULD work but it kept saying permission denied even with sudo
In the mean time if anyone wants to try the R.ID drivers with Windows this is the one to grab: https://sourceforge.net/projects/ra...n10-Win11-PolarisVegaNavi-Nebula.exe/download

 

[mczkrt]

Another weirdness, after installing R.ID drivers, card reports using Large Memory Range, even if Resizable BAR is disabled in BIOS.

 

[PixelButts]

While in Linux, I also enabled bifurcation (x8/x8) as well as resizable BAR and it had virtually no effect. On windows the same was true it didn't seem to matter for some reason with and without the R.ID driver. I've let them know about this and will see whether they can fix that or not. Does yours report error code 12 and insists theres a resource conflict like mine does?

 

[mczkrt]

No, only error 43 so far.

 

[PixelButts]

Well thats interesting I guess. What components are you using in your system? Perhaps it's looking for something in an HEDT/Server system? Anyone else wanna take a stab at it?

 

[LabRat 891]

Well thats interesting I guess. What components are you using in your system? Perhaps it's looking for something in an HEDT/Server system? Anyone else wanna take a stab at it?

SMbus/GPIO, maybe a remote management module?

I've never heard of such a requirement, but: Hot-Swap support is one thing most server boards do, that ours do not.

Another weirdness, after installing R.ID drivers, card reports using Large Memory Range, even if Resizable BAR is disabled in BIOS.

Can a GPU be programmed to request/use a 'non-standard' BAR? Like, a 'workaround' to pre-ReBAR support? (possibly needing a modded mainboard BIOS?)

Also, might be related to UEFI GOP certificates being pulled by msft. (Part of issues I've seen w/ Vega acting bricked is it getting 'stuck' in some kind of non?UEFI 'safe mode'.)
IIRC, my MI25 and Vega 64 are listed as 'insecure' UEFI devices after a certain AGESA on my X570. These V540s are 1st gen Navi; I'd expect similar.

 

[mczkrt]

I'm still on the same platform, Ryzen 5 3600, Asrock B550M Pro4. I briefly tested on X58, that one showed error 12 with every driver, but I never investigated further, just assumed it's old age/incompatibility issue.

I don't think GOP has any relation, the card is not recognized as a display adapter by the mainboard. If you put another graphics card in the system, that one will be used to display POST every time.

There is still a mystery of Microchip FPGA, no other dual GPU AMD card I researched have such feature.

https://www.microchip.com/en-us/products/fpgas-and-plds/system-on-chip-fpgas/smartfusion-2-fpgas

 

[LabRat 891]

If you put another graphics card in the system, that one will be used to display POST every time.

There is still a mystery of Microchip FPGA, no other dual GPU AMD card I researched have such feature.
https://www.microchip.com/en-us/products/fpgas-and-plds/system-on-chip-fpgas/smartfusion-2-fpgas

Same thing occurs w/ an un-crossflashed MI25, too. However, it is possible to get an un-modded MI25 'working' in Linux (no video out)

The MI25 (and WX9100?) has a similar/same Microchip FPGA. I'm not sure what the intended use was on the MI25 either.
Regardless, it's something somewhat common to AMD's Instinct cards.

My guess:
the FPGA is for inter-PCIe Device communication. No reason to have the same-similar FPGA on a single GPU MI card, unless it's for functions external to the card.
dGMA, perhaps? Or, maybe something application-specific and related to 'synchronizing' work across multiple cards?

 

[PixelButts]

Well here's what I've done so far
Setup A

• TUF B550-Plus, Ryzen 3700X
• Windows: Result is sometimes error code 43, sometimes error code 12
• Linux: Works through that workaround

Setup B

• Lenovo P520 Stadia Dev Node (listen, I'm desperate), Xeon W-2135
• Windows: Result is error code 43 only with whatever I try
• Linux: Haven't tried because I dont wanna set this up rn, but it probably works

Seeing as it works under linux albeit with an alternative card to handle the video output, it's not necessarily a hardware issue, it's a software issue then?