• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Radeon Pro v540 Research Thread

Joined
Mar 31, 2023
Messages
131 (0.22/day)
I have once again followedup on that ticket, and I inquired about that potential mispoken model and the answer was as follows
  • No update yet, they're workin on my issue still
  • The rep definitely said the incorrect model, an RX 540 is definitely not correct, and that an RX 5400 is also definitely not (so says the new rep), so my theory is out, and spec wise that was kinda given. Whether such exists or not idk but he seemed not too inclined to answer that.
  • Equivalent cards are definitely 5000 series, specifically in the 5600-5700 XT range for what they offer (though we knew this already)
  • The W5500X-W5700X would also be in the same ballpark according to the rep, though whether this has any compatibility is hard to say since those arent HBM2
  • He also mentioned but glossed over the 5600M which and I quote "Likely doesn't help your specific case" which is a little funny given the results above from @mczkrt
Hopefully this clarified some more stuff, when I get a formal reply to things I'll update this again, I just took the moment I had since the RX 540 part has been bugging me a ton
 
Joined
May 20, 2023
Messages
73 (0.14/day)
It seems to me that the tech support doesn't fully understand the topic you are discussing. Maybe that is the reason they even agreed to open a ticket for officially unreleased product. Still, huge thumbs up for trying this route.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
It seems to me that the tech support doesn't fully understand the topic you are discussing. Maybe that is the reason they even agreed to open a ticket for officially unreleased product. Still, huge thumbs up for trying this route.
You're correct and thats why they had to come back with info at a later date as it needs to be reviewed by those who can answer it. The reps definitely had no idea, and that's totally fine with me, as long as someone who does know something looks at it I'm calling it a win.

As I understand it, In the ticket they detail exactly what the specific request is and mine was quite broad; I specifically mentioned the card definitely works, both onboard GPUs, but under Linux only one GPU on the board works but the other will not so simultaenous GPU functionalty is out of the question, and under Windows neither will work as it returns a resource error for unknown reasons. The exact request on ticket is along the lines of "any information to both troubleshoot the issues further, or general information about what hardware or software this GPU is looking for to operate correctly under Windows and Linux". I extended this to drivers, vbios, operating sytems, some motherboard configurations, any information at all that could help identify the issue further not necessarily asking for solution or fix.

The reason I did this mostly because I know it's a waste of everyone's time with an engineering sample, and seeing as my specific card is NOT an engineering sample and rather a final production one, it stands to reason that any issues I have would actually be taken seriously even if it's a shot in the dark. Since it's unreleased but also a production card it was worth trying at all since it could go either way. On top of this since it's in the "Professional Graphics" lineup, it means I dont have to call the normal consumer line, so there may be more chance of an answer just because it's not a regular consumer card. And to be fair to everyone including the rep did admit this was quite the unusual call to begin with so I did at least clarify that theres a number of engineering samples out there, but the one I'm inquiring about isnt one of those and I think that's why he took the case at all.

And besides in case this avenue didn't work I still had a fairly reasonable inquiry about the Pro VII infinity link bridge, which is the issue I opened with as it's significantly less annoying of an issue and probably a lot simpler to resolve anyway. I may be slightly abusing a few things here, but every time I have ever called AMD it has been genuinely pleasant or helpful. I wish I could say the same about their e-mail support on one instance. Sometimes just talking to a person gets results.

Realistically I just want these cards to work and be useful, not e-waste. They definitely mine which okay sure that's something but surely we can do more.
 
Last edited:
Joined
Mar 31, 2023
Messages
131 (0.22/day)
Well got a reply though cant say I'm super surprised with what it is, though we now have an "official enough" reply.
Thank you for contacting AMD.

The Radeon Pro V540 was never released to the market, and thus we're not sure how you were able to get one.
We recommend returning it to the vendor, as we might not be able to help with an unreleased product, for a variety of reasons (driver support, exact specs, etc).
I have sent a reply thanking them for their time, and will continue my equally annoying ask about the Infinity Link Bridge in response. I guess we really only have to hope some amazing driver wizard shows up, or flash the 5600M to it and roll with that even if that's probably not the best.

Another owner of one suggested cold emailing Shadow PC since supposedly they use the v520 for cloud gaming. I mean the idea is definitely sound.
 

Wolf9466

New Member
Joined
May 18, 2024
Messages
4 (0.02/day)
I have a V540. Added support for it to one of my personal tools, and modified my VBIOS on it.

Also, both dies work under Linux for me...
 

Wolf9466

New Member
Joined
May 18, 2024
Messages
4 (0.02/day)
I edited it, sorry. I forgot to include both dies work on mine - I mined on it a couple times to test. As for modifications, I did some changes to max and min clocks/voltages, and modified the GFXCLK curve because I didn't like it, and upped the memory clock some. Also modified the memory timings.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
I edited it, sorry. I forgot to include both dies work on mine - I mined on it a couple times to test. As for modifications, I did some changes to max and min clocks/voltages, and modified the GFXCLK curve because I didn't like it, and upped the memory clock some. Also modified the memory timings.
Well we already know mining works on these, that much is for sure, but have you been able to do any non OpenGL graphical workloads on it without any weird workarounds?
 

Wolf9466

New Member
Joined
May 18, 2024
Messages
4 (0.02/day)
That's... um... that's not going to work how you think. It may *work*, but only with one die. The reason is, there are literally two seperate Navi12 dies on this GPU, which (IIRC) can only really communicate over xGMI. Because most software isn't built to take advantage of multiple distinct GPUs, it just won't. Remember the death of CrossFire?
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
That's... um... that's not going to work how you think. It may *work*, but only with one die. The reason is, there are literally two seperate Navi12 dies on this GPU, which (IIRC) can only really communicate over xGMI. Because most software isn't built to take advantage of multiple distinct GPUs, it just won't. Remember the death of CrossFire?
By chance did you skip over the whole thread before posting? We are aware of what the card is, 2 dies on board, Navi12, and some definite cooling issues. The goal is to get it usable (preferably in non-mining manners) if possible.
The use case of this is clearly some cloud based workload, potentially (assuming the assumptions were correct given they had the drivers) for Amazon's AWS EC2 Instances, which do an assortment of graphical and non-graphical workloads for Windows and Linux
It stands to reason there is a way to do it, but the method is unclear. Mining has just always seemed to work for some reason, but outside of whatever was going on with AWS and mining it seems to be having trouble.
The only current success has been Linux with a particular setup on kernel and drivers plus a secondary GPU for outputting the workload, with Windows there's been virtually 0 success, just error codes with unclear solutions to them if any at all. Driver wise theres 4 sets so far (excluding the R.ID drivers), one of which is from a Chinese company utilizing v520s, which while the same underlying GPU, does not play as nice on the v540.
There is also the effort to flash the 5600M vbios to it which did get interesting results but is not nearly as useful as running stock.
 

Wolf9466

New Member
Joined
May 18, 2024
Messages
4 (0.02/day)
Windows, you will not have success (with AMD's drivers). I have seen those drivers, they are quite bad and they tend to implement only what they need to. I... would not flash the 5600M VBIOS to it - that may cause some issues depending on board config.

EDIT: I read over some of the thread again - in regards to virtualization, I do get vGPUs showing up and the control stuff in sysfs, if that helps. Also, its VBIOS indicates it supports virtualization.
 
Joined
Apr 18, 2019
Messages
2,328 (1.15/day)
Location
Olympia, WA
System Name Sleepy Painter
Processor AMD Ryzen 5 3600
Motherboard Asus TuF Gaming X570-PLUS/WIFI
Cooling FSP Windale 6 - Passive
Memory 2x16GB F4-3600C16-16GVKC @ 16-19-21-36-58-1T
Video Card(s) MSI RX580 8GB
Storage 2x Samsung PM963 960GB nVME RAID0, Crucial BX500 1TB SATA, WD Blue 3D 2TB SATA
Display(s) Microboard 32" Curved 1080P 144hz VA w/ Freesync
Case NZXT Gamma Classic Black
Audio Device(s) Asus Xonar D1
Power Supply Rosewill 1KW on 240V@60hz
Mouse Logitech MX518 Legend
Keyboard Red Dragon K552
Software Windows 10 Enterprise 2019 LTSC 1809 17763.1757
i sent AMD a email and hope to get a it followed up with a story

You will likely get the same response I did but I wish you luck. Least you didn't make a phone call like I did.

When AMD was firm with PixelButts about it being "never released" (despite having a non-ES card w/ a production vBIOS in-hand), I started to wonder if there was more to all of this than just a neat GPU that never saw widespread adoption (even in-industry).

It'd be hilarious (and terrible) if we enthusiasts got AMD in trouble w/ the IRS. From what I've ran across, ES and devkits are often supposed to be destroyed and written-off on taxes.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
When AMD was firm with PixelButts about it being "never released" (despite having a non-ES card w/ a production vBIOS in-hand), I started to wonder if there was more to all of this than just a neat GPU that never saw widespread adoption (even in-industry).
For what it's worth, and this may not matter to most people, I do have some insight into how this works specifically with console development hardware. The specifics vary a bit but GENERALLY speaking you are intended to return the hardware to them (see: Sony, Microsoft, Nintendo, Apple, Etc) to which they either A) get resold to newer devs at a lower price as the hardware is no longer the latest (assume final hardware for this situation), or B) disposed of by them in e-waste, typically boated over to China, where most of the e-waste recycling takes place.

Because of how things play out in reality you sometimes see 1) Auctions of the hardware when not authorized (accidental or intentional), 2) Companies going under and avenues to properly dispose/return of it are not done at all due to finances or people holding onto stuff with minimal to no accountability or adequate tracking, or my favorite 3) it was sent back, was boated over for recycling, but survived before it was destroyed and someone salvaged the parts or system and sold it. This is usually why a receipt of destruction is desired but sometimes its so much bulk that its easier to not do that. The third one is usually why you se mountains of 360 devkits just pieced together for a couple bucks. It works enough and theres pleny. They might red ring in 2 months but at the price you paid and condition its kinda expected. Sometimes rarer stuff gets through but depends on what it is or hardware.

It'd be hilarious (and terrible) if we enthusiasts got AMD in trouble w/ the IRS. From what I've ran across, ES and devkits are often supposed to be destroyed and written-off on taxes.
In this case, it is unclear what would happen but I am leaning toward nothing happening at all because almost 100 were sold on ebay, at a pretty low price, very publicly despite the mislabeled title of it, numerous people with them now so tracking that down would be expensive and perhaps pointless, and realistically the company they came from probably gets a slap on the wrist since it was as AMD described custom to whoever made it so its more likely the loss of that company is minimal enough that they dont care. This wasnt some back alley deal either this was pretty clearly lots of stock just being listed without a care in the world. If anyone got in trouble it would be whoever listed it, if at all. And even then they're a pain to even get working beyond mining so I very much doubt driver wise anyone is worried here either unless its like core to your system or internal stuff but it's drivers in the end it's not that bad.

Everything I've said typically applies to final hardware rather than prototypes or engineering samples (ES/EVT/DVT/EVA, etc). The more prototype-y it is the more companies care and the more likely they are destroyed for good making them significantly rarer in the long run. These boards are virtually final so it doesn't even seem that bad. If anything the only unusual aspect of this is it's unreleased entirely. I would question why there's even a full production version in the wild before questioning 100 ES cards being out there given what's known about it. If we saw a board where it was definitely straight out of the engineering department's bench then I would say worry.

I collect development hardware for an assortment of personal reasons, I just know theres some things not worth the trouble (expenses, questionable sources, complexity or damage of the device, software even existing at all, legality) or are basically bricks. The v540 is much closer to usable and weird than it is a useless brick, so long as the software side can somehow get things to play nice. But as it stands it's closer to a brick that mines until this changes. You can make a somewhat good argument that this is like the v420 (nice), which never released either and we've seen exactly 1 of thanks to Fouquin, and afaik nobody cared.

Then you see the people dropping 20k on an item in not even that great condition when its usually going around 600-1000.
 
Joined
May 20, 2023
Messages
73 (0.14/day)
I'm back with report of some progress.

I managed to run 1 GPU successfully on Windows 10 under VM, with Proxmox 8.2 as hypervisor, with production rom and 22.10.01.12-220930a-384126e-WHQLCert driver (from LTT). Device id is faked by PCIe passthrough to 7362 (V520) to allow drivers to behave.

This is still work in progress, I will try to document steps and setup, maybe try to get both GPUs working under same VM. But this has to wait until next weekend.
 

Attachments

  • v540-virt.png
    v540-virt.png
    2 MB · Views: 60
  • gpuz.gif
    gpuz.gif
    23.7 KB · Views: 64
  • devman.png
    devman.png
    41 KB · Views: 56
Joined
Mar 31, 2023
Messages
131 (0.22/day)
I'm back with report of some progress.

I managed to run 1 GPU successfully on Windows 10 under VM, with Proxmox 8.2 as hypervisor, with production rom and 22.10.01.12-220930a-384126e-WHQLCert driver (from LTT). Device id is faked by PCIe passthrough to 7362 (V520) to allow drivers to behave.

This is still work in progress, I will try to document steps and setup, maybe try to get both GPUs working under same VM. But this has to wait until next weekend.
Interesting. If you did this on the R.ID drivers would there be any difference?
I once setup a Proxmox machine for work with 3 GPUs in it and it didn't even occur to me to try spoofing IDs.
 
Joined
May 20, 2023
Messages
73 (0.14/day)
I tried R.ID initially, but installing drivers ended with error. This might be matter of choosing correct options during setup. I will try to get them to work again, as now it's way faster to iterate different versions with VM snapshot/restore.

Device ID spoofing made no difference when tested last year, with AWS drivers and older Proxmox/kernel.

For now I can say that results are quite promising. I can attach two GPUs to a single machine, both are operational. Drivers survive reboots of VM and host. No crashes or weird behavior so far.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
I tried R.ID initially, but installing drivers ended with error. This might be matter of choosing correct options during setup. I will try to get them to work again, as now it's way faster to iterate different versions with VM snapshot/restore.

Device ID spoofing made no difference when tested last year, with AWS drivers and older Proxmox/kernel.

For now I can say that results are quite promising. I can attach two GPUs to a single machine, both are operational. Drivers survive reboots of VM and host. No crashes or weird behavior so far.
If this is the case its likely just that they haven't added the normal v520 to the driver packages. It shouldn't be hard to do this though. I might ask them to as being reliant on a single driver package going forward would suck.
 
Joined
May 20, 2023
Messages
73 (0.14/day)
I managed to install R.ID driver, for now 23.40 with 5.2 kernel seems to work. It's not optimal though. Only 1st GPU will work reliably. Any access to 2nd GPU with driver installed, for example reading with GPUz will crash the system.
r-id.png

No tuning options, no MPT support
adrenalin-tuning.png
mpt.png

Performance is halved compared to 22.10
1080p-h.png
1080p-h-rid.png


There are still some driver setup options that can be tested, maybe someone with more experience with this driver can advise.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
I managed to install R.ID driver, for now 23.40 with 5.2 kernel seems to work. It's not optimal though. Only 1st GPU will work reliably. Any access to 2nd GPU with driver installed, for example reading with GPUz will crash the system.
View attachment 349060
No tuning options, no MPT support
View attachment 349061View attachment 349062
Performance is halved compared to 22.10
View attachment 349063View attachment 349064

There are still some driver setup options that can be tested, maybe someone with more experience with this driver can advise.
Halved? Well damn thats certainly interesting to say the least.

Did the drivers from AWS have any success or just the LTT ones?
 
Joined
May 20, 2023
Messages
73 (0.14/day)
It seems I got confused.

22.10.01.12-220930a-384126e-WHQLCert_.zip is actually driver version 22.7.3, these are from AWS, and by far the most reliable.
Both chips can be used, even Multi-GPU works.
gm-mgpu-d12.png
gm-mgpu-vk.png

Single GPU for reference:
gm-d12.png
gm-vk.png

Drivers from LTT are version 22.10, can be installed to both cards, won't report error 43 after reboot. But only 1st GPU will be usable. 2nd tends to disappear from system when accessed by GPUz for example. Performance I'd say is on par.
1080ph-22-10.png

@PixelButts if you are in contact with R.ID team and they are willing to work on this, it looks like AWS drivers are a good base to start.
 
Joined
Mar 31, 2023
Messages
131 (0.22/day)
It seems I got confused.

22.10.01.12-220930a-384126e-WHQLCert_.zip is actually driver version 22.7.3, these are from AWS, and by far the most reliable.
Both chips can be used, even Multi-GPU works.
View attachment 349125View attachment 349126
Single GPU for reference:
View attachment 349127View attachment 349128
Drivers from LTT are version 22.10, can be installed to both cards, won't report error 43 after reboot. But only 1st GPU will be usable. 2nd tends to disappear from system when accessed by GPUz for example. Performance I'd say is on par.
View attachment 349135
@PixelButts if you are in contact with R.ID team and they are willing to work on this, it looks like AWS drivers are a good base to start.
I'm in their discord server but I do not know what they actually do to make the drivers work at all. I do know they can just add IDs to the package and all that but whether it works well is another story. I can forward them the thread and drivers though and see what they do, if anything.
 
Joined
Jul 15, 2019
Messages
558 (0.29/day)
Location
Hungary
System Name Detox sleeper
Processor Intel i9-7980XE@4,5Ghz
Motherboard Asrock x299 Taichi XE (custom bios with ecc reg support, old microcode)
Cooling Custom water: Alphacool XT45 1080 + 9xArctic P12, EK-D5 pump combo, EK Velocity D-RGB block
Memory 8x16Gb Hynix DJR ECC REG 3200@4000
Video Card(s) Nvidia RTX 3080 FE 10Gb undervolted
Storage Samsung PM9A1 1Tb + PM981 512Gb + Kingston HyperX 480Gb + Samsung Evo 860 500Gb
Display(s) HP ZR30W (30" 2560x1600 10 bit)
Case Chieftec 1E0-500A-CT04 + AMD Sempron sticker
Audio Device(s) Genius Cavimanus
Power Supply Super Flower Leadex 750w Platinum
Mouse Logitech G400
Keyboard IBM Model M122 (boltmod, micro pro usbc)
Software Windows 11 Pro x64
Has anyone tried the vega 64 driver forced install method?
It works on the BC-160. (V520)

 

levelud

New Member
Joined
Aug 1, 2023
Messages
6 (0.01/day)
Has anyone tried the vega 64 driver forced install method?
It works on the BC-160. (V520)

Tried to force it with newest AMD driver for Vega64 with Intel based System i3 13100, sadly it didn't work for me... Perhaps also some bios settings are wrong, or it has to be an AMD system.
 
Top