AMD Radeon Pro v540 Research Thread

[mczkrt]

I have been using radeon-profile to manipulate clocks. Both GPUs will accept locking to lower power state, but second one is more fiddly - make too many changes or do it too fast and it will crash. No overclocking - anything above default clock and power limit is ignored.

Available states:
Core: 300, 700, 1450 MHz
Memory: 167, 350, 800, 1000 MHz

If You are interested in mining ETH/ETC on this, I have bad news - memory clocks don't ramp up and you end with something like 2,5MH/s (even with locking to highest frequency), it will also crash in 5-15 mins (my experience, YMMV). Randomx seems to work stable and with max clocks. I observed similar behavior with V520.

 

[JDcryptBB]

About the 2,5 Mh/s - I can verify with this card in Hiveos - without OC ability i got 49-52 Mh/s - (EthashB3) very similar to eth clocks - but it works. I dont think ive mined any other Eth based algos. If i do it is with a asic for ETH/ETC. Also got 25 or 30 Mh/s on ravencoin

I have also gotten a modified bios (I apologize - but i have been asked not to share it) / Working with a fellow that has a good idea about AMD hardware if we put it that way - But not much change as your limited to a similar sclk / mclk states -- Also I only have 2 power states on core appear in hiveos shell using Sysf. 4 Power states for mclk. i believe mine is set a bit higher in mem. But the 52 mh/s and other Mh/s from Kawpow algos was on the original stock bios that i got the card with.

PS - I was hoping to use some resources to grab a BC-160 Vbios - Still searching - But considering you have had some sort of luck with other vbios im wondering if the BC-160 will work - Although i dont think it will change OC ability - I was told its probably built into the SMU FW thats why mining OS can talk to most of the card but cant use the CPU built into the core for clocking, because it doesnt recognize the SMU FW side.

 

[mczkrt]

I admit I'm only testing on Ubuntu, as mining performance is not a priority, but still this is a great news.

Can you at least share header from your modded vbios? From offset 0x180 to 0x300, typically.

 

[JDcryptBB]

I admit I'm only testing on Ubuntu, as mining performance is not a priority, but still this is a great news.

Can you at least share header from your modded vbios? From offset 0x180 to 0x300, typically.

Im pretty sure 0x300 is where the signature lives still? Im not "great" with this stuff, and still prying to learn, afaik i would need to "dump" it, viewing it in a plain text editor does nothing for more, nor does Visual studio code - Apologies been looking at a way to actually view it.

 

[mczkrt]

I assumed You have the rom file, if not, You have to make a dump.

This should work, as root, where x is card number:
cat /sys/kernel/debug/dri/x/amdgpu_vbios.rom > vbios.rom

Use HxD or any other hex editor, but text editor will also work. What I'm talking about look like this:

113-D3050301-X00 FGL NAVI12 PCI_EXPRESS HBM2
NAVI12 A0 GLXLB D30503 8GB BC160 1150e/334m HYN/SAM                         
 
 (C) 1988-2018, Advanced Micro Devices, Inc. ATOMBIOSBK-AMD VER017.003.000.008.017114 AB574625.bin 2332358  574625
AMD_NAVI12_D30503_GLXLB_A0_HBM2_8GB_BC160\config.h

 

[JDcryptBB]

113-D3020100-012 NAVI12 PCI_EXPRESS�HBM2� NAVI12 A0 GLXT D30201 8GB 1150e / 167m HYN / SAM


(C)1988 - 2018,
Advanced Micro Devices, Inc.�ATOMBIOSBK - AMD VER017 .003.000.006.013451 D3020100 .012 2047344 495040 AMD_NAVI12_D30201_GTXL_A0_HBM2_8GB_GIBRALTAR\config.h

This is part of the Stock one i had - is this what your asking for? Mine comes out with a bunch of NUL all over + symbols and stuff im assuming binary but way messier than your link (just checking to see if its even what your asking for looks slightly different than your post for sure)

 

[mczkrt]

Yes, that's the thing, but I'm interested in header of the modded vbios, if You are willing to share of course.

 

[PixelButts]

Hey, I just picked one of these cards up and have been working to figure out what makes it tick in terms of gaining controls on the clocks, I wanted to ask if this ever went on, what you learned, ended up doing with this? It works on HiveOS miner, but no clocks unless you go in through SSH/shell and do it through sysf (changing dpm states) which can adjust clock but not very much/or very well. Any response from you would be great, me and a couple others with the card are pretty intrigued by these cards and want to see what we can push through them. Thank you!

We've been trying a lot of things, but the card is definitely fighting us every step of the way unfortunately. Results have been incredibly varied sadly.

 

[JDcryptBB]

Yes, that's the thing, but I'm interested in header of the modded vbios, if You are willing to share of course.

Header of the modded looks to be all the same info - with the exception of the date posted above it
761295520 12/17/19 17:11 113-D3020100-012 NAVI12 PCI_EXPRESS HBM2
NAVI12 A0 GLXT D30201 8GB 1150e/167m HYN/SAM


(C) 1988-2018, Advanced Micro Devices, Inc. ATOMBIOSBK-AMD VER017.003.000.006.013451 D3020100.012 2047344 495040 AMD_NAVI12_D30201_GTXL_A0_HBM2_8GB_GIBRALTAR\config.h

modded -
761295520 07/04/23 23:28 113-D3020100-012 NAVI12 PCI_EXPRESS HBM2
NAVI12 A0 GLXT D30201 8GB 1150e/167m HYN/SAM


(C) 1988-2018, Advanced Micro Devices, Inc. ATOMBIOSBK-AMD VER017.003.000.006.013451 D3020100.012 2047344 495040 AMD_NAVI12_D30201_GTXL_A0_HBM2_8GB_GIBRALTAR\config.h

 

[mczkrt]

Thank you.

Did you try flashing it? If so, could you share output from dmesg after boot?

dmesg | grep amdgpu

 

[JDcryptBB]

[67284.345899] amdgpu 0000:1e:00.0: amdgpu: GFXCLK_LIMITS not supported!
[67284.345903] amdgpu 0000:1e:00.0: amdgpu: GFXCLK_LIMITS not supported!
[67284.345955] amdgpu 0000:1e:00.0: amdgpu: UCLK_MAX not supported!
[67284.345957] amdgpu 0000:1e:00.0: amdgpu: UCLK_MAX not supported!
[67284.353904] amdgpu 0000:21:00.0: amdgpu: GFXCLK_LIMITS not supported!
[67284.353907] amdgpu 0000:21:00.0: amdgpu: GFXCLK_LIMITS not supported!
[67284.353965] amdgpu 0000:21:00.0: amdgpu: UCLK_MAX not supported!
[67284.353967] amdgpu 0000:21:00.0: amdgpu: UCLK_MAX not supported!

- I'm working with another guy, who said we may be able to go in and enable overdrive.
Still waiting to hear from him, sent him over the pp_table. Also have a BC-160 Vbios (which im told this is the same as the BC-160 vbios) but the table on this other BC-160 vbios seems like a bit more is enabled on it, still need to give that one a go.

 

[mczkrt]

By the timestamps, that is already some uptime, I was curious about how your card is being initialized.
This is mine, just after reboot:

Spoiler: dmesg

[    8.854746] [drm] amdgpu kernel modesetting enabled.
[    8.859731] [drm] amdgpu version: W.5.4.0.4
[    8.863922] [drm] amdgpu internal version: W.5.4.0.4.1184451
[    8.913923] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 0: 0x7c00000000 -> 0x7dffffffff
[    8.923528] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 2: 0x7e00800000 -> 0x7e009fffff
[    8.933110] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 5: 0xfc800000 -> 0xfc87ffff
[    8.942415] amdgpu 0000:09:00.0: enabling device (0000 -> 0002)
[    8.956894] amdgpu 0000:09:00.0: Trusted Memory Zone (TMZ) feature disabled as experimental (default)
[    9.111173] amdgpu 0000:09:00.0: VRAM: 8176M 0x0000008000000000 - 0x00000081FEFFFFFF (8176M used)
[    9.120043] amdgpu 0000:09:00.0: GART: 512M 0x0000000000000000 - 0x000000001FFFFFFF
[    9.158896] [drm] amdgpu: 8176M of VRAM memory ready
[    9.163875] [drm] amdgpu: 15915M of GTT memory ready.
[   10.104909] amdgpu 0000:09:00.0: RAS: ras ta ucode is not available
[   10.270800] amdgpu 0000:09:00.0: fb0: amdgpudrmfb frame buffer device
[   10.304934] amdgpu 0000:09:00.0: ring gfx_0.0.0 uses VM inv eng 0 on hub 0
[   10.311907] amdgpu 0000:09:00.0: ring comp_1.0.0 uses VM inv eng 1 on hub 0
[   10.318923] amdgpu 0000:09:00.0: ring comp_1.1.0 uses VM inv eng 4 on hub 0
[   10.325938] amdgpu 0000:09:00.0: ring comp_1.2.0 uses VM inv eng 5 on hub 0
[   10.332974] amdgpu 0000:09:00.0: ring comp_1.3.0 uses VM inv eng 6 on hub 0
[   10.339992] amdgpu 0000:09:00.0: ring comp_1.0.1 uses VM inv eng 7 on hub 0
[   10.347018] amdgpu 0000:09:00.0: ring comp_1.1.1 uses VM inv eng 8 on hub 0
[   10.354036] amdgpu 0000:09:00.0: ring comp_1.2.1 uses VM inv eng 9 on hub 0
[   10.361054] amdgpu 0000:09:00.0: ring comp_1.3.1 uses VM inv eng 10 on hub 0
[   10.368177] amdgpu 0000:09:00.0: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
[   10.375200] amdgpu 0000:09:00.0: ring sdma0 uses VM inv eng 12 on hub 0
[   10.381871] amdgpu 0000:09:00.0: ring sdma1 uses VM inv eng 13 on hub 0
[   10.388532] amdgpu 0000:09:00.0: ring vcn_dec uses VM inv eng 0 on hub 1
[   10.395273] amdgpu 0000:09:00.0: ring vcn_enc0 uses VM inv eng 1 on hub 1
[   10.402142] amdgpu 0000:09:00.0: ring vcn_enc1 uses VM inv eng 4 on hub 1
[   10.408979] amdgpu 0000:09:00.0: ring jpeg_dec uses VM inv eng 5 on hub 1
[   10.416005] [drm] Initialized amdgpu 3.36.0 20150101 for 0000:09:00.0 on minor 0
[   10.423522] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 0: 0x7000000000 -> 0x71ffffffff
[   10.433177] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 2: 0x7200800000 -> 0x72009fffff
[   10.442843] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 5: 0xfc400000 -> 0xfc47ffff
[   10.452198] amdgpu 0000:0c:00.0: enabling device (0000 -> 0002)
[   10.466737] amdgpu 0000:0c:00.0: Trusted Memory Zone (TMZ) feature disabled as experimental (default)
[   10.621040] amdgpu 0000:0c:00.0: VRAM: 8176M 0x0000008000000000 - 0x00000081FEFFFFFF (8176M used)
[   10.630015] amdgpu 0000:0c:00.0: GART: 512M 0x0000000000000000 - 0x000000001FFFFFFF
[   10.646789] [drm] amdgpu: 8176M of VRAM memory ready
[   10.651801] [drm] amdgpu: 15915M of GTT memory ready.
[   11.605318] amdgpu 0000:0c:00.0: RAS: ras ta ucode is not available
[   11.754055] amdgpu 0000:0c:00.0: ring gfx_0.0.0 uses VM inv eng 0 on hub 0
[   11.763902] amdgpu 0000:0c:00.0: ring comp_1.0.0 uses VM inv eng 1 on hub 0
[   11.773782] amdgpu 0000:0c:00.0: ring comp_1.1.0 uses VM inv eng 4 on hub 0
[   11.783647] amdgpu 0000:0c:00.0: ring comp_1.2.0 uses VM inv eng 5 on hub 0
[   11.793454] amdgpu 0000:0c:00.0: ring comp_1.3.0 uses VM inv eng 6 on hub 0
[   11.803262] amdgpu 0000:0c:00.0: ring comp_1.0.1 uses VM inv eng 7 on hub 0
[   11.813027] amdgpu 0000:0c:00.0: ring comp_1.1.1 uses VM inv eng 8 on hub 0
[   11.822782] amdgpu 0000:0c:00.0: ring comp_1.2.1 uses VM inv eng 9 on hub 0
[   11.832467] amdgpu 0000:0c:00.0: ring comp_1.3.1 uses VM inv eng 10 on hub 0
[   11.842207] amdgpu 0000:0c:00.0: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
[   11.851862] amdgpu 0000:0c:00.0: ring sdma0 uses VM inv eng 12 on hub 0
[   11.861156] amdgpu 0000:0c:00.0: ring sdma1 uses VM inv eng 13 on hub 0
[   11.870465] amdgpu 0000:0c:00.0: ring vcn_dec uses VM inv eng 0 on hub 1
[   11.879853] amdgpu 0000:0c:00.0: ring vcn_enc0 uses VM inv eng 1 on hub 1
[   11.889310] amdgpu 0000:0c:00.0: ring vcn_enc1 uses VM inv eng 4 on hub 1
[   11.898774] amdgpu 0000:0c:00.0: ring jpeg_dec uses VM inv eng 5 on hub 1
[   11.908408] [drm] Initialized amdgpu 3.36.0 20150101 for 0000:0c:00.0 on minor 1

I also think about getting BC-160, but I'm not sure if it will help in anyway and I won't have much use for it otherwise. Maybe if some good deal pops up.

 

[Beatyoubeach]

By the timestamps, that is already some uptime, I was curious about how your card is being initialized.
This is mine, just after reboot:

Spoiler: dmesg

[    8.854746] [drm] amdgpu kernel modesetting enabled.
[    8.859731] [drm] amdgpu version: W.5.4.0.4
[    8.863922] [drm] amdgpu internal version: W.5.4.0.4.1184451
[    8.913923] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 0: 0x7c00000000 -> 0x7dffffffff
[    8.923528] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 2: 0x7e00800000 -> 0x7e009fffff
[    8.933110] amdgpu 0000:09:00.0: remove_conflicting_pci_framebuffers: bar 5: 0xfc800000 -> 0xfc87ffff
[    8.942415] amdgpu 0000:09:00.0: enabling device (0000 -> 0002)
[    8.956894] amdgpu 0000:09:00.0: Trusted Memory Zone (TMZ) feature disabled as experimental (default)
[    9.111173] amdgpu 0000:09:00.0: VRAM: 8176M 0x0000008000000000 - 0x00000081FEFFFFFF (8176M used)
[    9.120043] amdgpu 0000:09:00.0: GART: 512M 0x0000000000000000 - 0x000000001FFFFFFF
[    9.158896] [drm] amdgpu: 8176M of VRAM memory ready
[    9.163875] [drm] amdgpu: 15915M of GTT memory ready.
[   10.104909] amdgpu 0000:09:00.0: RAS: ras ta ucode is not available
[   10.270800] amdgpu 0000:09:00.0: fb0: amdgpudrmfb frame buffer device
[   10.304934] amdgpu 0000:09:00.0: ring gfx_0.0.0 uses VM inv eng 0 on hub 0
[   10.311907] amdgpu 0000:09:00.0: ring comp_1.0.0 uses VM inv eng 1 on hub 0
[   10.318923] amdgpu 0000:09:00.0: ring comp_1.1.0 uses VM inv eng 4 on hub 0
[   10.325938] amdgpu 0000:09:00.0: ring comp_1.2.0 uses VM inv eng 5 on hub 0
[   10.332974] amdgpu 0000:09:00.0: ring comp_1.3.0 uses VM inv eng 6 on hub 0
[   10.339992] amdgpu 0000:09:00.0: ring comp_1.0.1 uses VM inv eng 7 on hub 0
[   10.347018] amdgpu 0000:09:00.0: ring comp_1.1.1 uses VM inv eng 8 on hub 0
[   10.354036] amdgpu 0000:09:00.0: ring comp_1.2.1 uses VM inv eng 9 on hub 0
[   10.361054] amdgpu 0000:09:00.0: ring comp_1.3.1 uses VM inv eng 10 on hub 0
[   10.368177] amdgpu 0000:09:00.0: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
[   10.375200] amdgpu 0000:09:00.0: ring sdma0 uses VM inv eng 12 on hub 0
[   10.381871] amdgpu 0000:09:00.0: ring sdma1 uses VM inv eng 13 on hub 0
[   10.388532] amdgpu 0000:09:00.0: ring vcn_dec uses VM inv eng 0 on hub 1
[   10.395273] amdgpu 0000:09:00.0: ring vcn_enc0 uses VM inv eng 1 on hub 1
[   10.402142] amdgpu 0000:09:00.0: ring vcn_enc1 uses VM inv eng 4 on hub 1
[   10.408979] amdgpu 0000:09:00.0: ring jpeg_dec uses VM inv eng 5 on hub 1
[   10.416005] [drm] Initialized amdgpu 3.36.0 20150101 for 0000:09:00.0 on minor 0
[   10.423522] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 0: 0x7000000000 -> 0x71ffffffff
[   10.433177] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 2: 0x7200800000 -> 0x72009fffff
[   10.442843] amdgpu 0000:0c:00.0: remove_conflicting_pci_framebuffers: bar 5: 0xfc400000 -> 0xfc47ffff
[   10.452198] amdgpu 0000:0c:00.0: enabling device (0000 -> 0002)
[   10.466737] amdgpu 0000:0c:00.0: Trusted Memory Zone (TMZ) feature disabled as experimental (default)
[   10.621040] amdgpu 0000:0c:00.0: VRAM: 8176M 0x0000008000000000 - 0x00000081FEFFFFFF (8176M used)
[   10.630015] amdgpu 0000:0c:00.0: GART: 512M 0x0000000000000000 - 0x000000001FFFFFFF
[   10.646789] [drm] amdgpu: 8176M of VRAM memory ready
[   10.651801] [drm] amdgpu: 15915M of GTT memory ready.
[   11.605318] amdgpu 0000:0c:00.0: RAS: ras ta ucode is not available
[   11.754055] amdgpu 0000:0c:00.0: ring gfx_0.0.0 uses VM inv eng 0 on hub 0
[   11.763902] amdgpu 0000:0c:00.0: ring comp_1.0.0 uses VM inv eng 1 on hub 0
[   11.773782] amdgpu 0000:0c:00.0: ring comp_1.1.0 uses VM inv eng 4 on hub 0
[   11.783647] amdgpu 0000:0c:00.0: ring comp_1.2.0 uses VM inv eng 5 on hub 0
[   11.793454] amdgpu 0000:0c:00.0: ring comp_1.3.0 uses VM inv eng 6 on hub 0
[   11.803262] amdgpu 0000:0c:00.0: ring comp_1.0.1 uses VM inv eng 7 on hub 0
[   11.813027] amdgpu 0000:0c:00.0: ring comp_1.1.1 uses VM inv eng 8 on hub 0
[   11.822782] amdgpu 0000:0c:00.0: ring comp_1.2.1 uses VM inv eng 9 on hub 0
[   11.832467] amdgpu 0000:0c:00.0: ring comp_1.3.1 uses VM inv eng 10 on hub 0
[   11.842207] amdgpu 0000:0c:00.0: ring kiq_2.1.0 uses VM inv eng 11 on hub 0
[   11.851862] amdgpu 0000:0c:00.0: ring sdma0 uses VM inv eng 12 on hub 0
[   11.861156] amdgpu 0000:0c:00.0: ring sdma1 uses VM inv eng 13 on hub 0
[   11.870465] amdgpu 0000:0c:00.0: ring vcn_dec uses VM inv eng 0 on hub 1
[   11.879853] amdgpu 0000:0c:00.0: ring vcn_enc0 uses VM inv eng 1 on hub 1
[   11.889310] amdgpu 0000:0c:00.0: ring vcn_enc1 uses VM inv eng 4 on hub 1
[   11.898774] amdgpu 0000:0c:00.0: ring jpeg_dec uses VM inv eng 5 on hub 1
[   11.908408] [drm] Initialized amdgpu 3.36.0 20150101 for 0000:0c:00.0 on minor 1

I also think about getting BC-160, but I'm not sure if it will help in anyway and I won't have much use for it otherwise. Maybe if some good deal pops up.

If all fails and you can't get it to run anything hardware accelerated crypto mining doesn't seem too bad considering how powerful and efficient it is.

If all fails and you can't get it to run anything hardware accelerated crypto mining doesn't seem too bad considering how powerful and efficient it is.

i got some more photos

 

[JDcryptBB]

LMAO the scenic pics though - Apologies @mczkrt im working on a fan adapter for this thing - Also i know it isn't much of your interest (crypto mining). But!!! to show movement here the guy who does MmpOS has helped me to achieve clocking capabilities - i had to give him the pp_table, he did some stuff on his OS and boom, now i can control clocks, he apparently enabled Overdrive. If you have more interest in this guy and what he did maybe you can add me in Discord and i can point you in his direction, or join MmpOS discord and look for him, he made his own OS so not sure maybe you can chat with him a little to see how you can do that in windows. His OS is linux based im almost certain of that.

PS - Me and others with this card im going to try to run all of our stock bios' through GPU-Z the other guy i know has 2 both show in hiveOS but the driver is just labeled xxx-xxx-xxxx --- This is a copy of that one. (The bios i have clock control on - IS my stock bios). If your interested in giving it a shot - I obtained a BC-160 vbios - It didnt work through HiveOS flash utility, but maybe just using the vbflash or other utility maybe you will have some luck. As I said if i find more with this card I will do what i can to start getting some of them uploaded to GPU-Z database.

 

[mczkrt]

Thanks for an update. This rom is identical to stock V540 no-GOP variant. There is BC-160 rom in TPU database, one I used to get Windows driver semi-working, but if you can get different build, all the better.

I have a programmer, so flashing and recovery is not a problem.

Can you check if you have amdgpu.ppfeaturemask set in your boot parameters, and to what value?

dmesg | grep amdgpu.ppfeaturemask

 

[Beatyoubeach]

LMAO the scenic pics though - Apologies @mczkrt im working on a fan adapter for this thing - Also i know it isn't much of your interest (crypto mining). But!!! to show movement here the guy who does MmpOS has helped me to achieve clocking capabilities - i had to give him the pp_table, he did some stuff on his OS and boom, now i can control clocks, he apparently enabled Overdrive. If you have more interest in this guy and what he did maybe you can add me in Discord and i can point you in his direction, or join MmpOS discord and look for him, he made his own OS so not sure maybe you can chat with him a little to see how you can do that in windows. His OS is linux based im almost certain of that.

PS - Me and others with this card im going to try to run all of our stock bios' through GPU-Z the other guy i know has 2 both show in hiveOS but the driver is just labeled xxx-xxx-xxxx --- This is a copy of that one. (The bios i have clock control on - IS my stock bios). If your interested in giving it a shot - I obtained a BC-160 vbios - It didnt work through HiveOS flash utility, but maybe just using the vbflash or other utility maybe you will have some luck. As I said if i find more with this card I will do what i can to start getting some of them uploaded to GPU-Z database.

I'll check it out thanks

 

[laceflower]

Available states:
Core: 300, 700, 1450 MHz

I might be stating the obvious, but those are also the same clock states as the reference RX5700. It has the same SU/TMU/ROP numbers as well. Is it safe to assume Navi 12 is basically a Navi 10 die stuck on top of an interposer?

 

[mczkrt]

I would say that is unlikely, as you need die space for 2048 bit memory bus, plus I imagine HBM memory controller is also a distinct piece of silicon.

 

[laceflower]

Does anyone have a Hawaii bracket/backplate to test out on the mounting holes? V540 mounting is ostensibly 54x54 (not 53.5x53.5) but I still don't have the card to test. If anyones in aus I'm happy to just mail you my spare

Also, it's a 1024 bit memory bus. There's 2 4-hi hbm stacks on the die with 4 channels to each stack - they're visible on either side of the core.

It makes you wonder why they'd opt for older, cheaper 4hi stock when they're already shelling for HBM on a limited run DC component and could've opted for a simpler 8hi design... unless you consider that Navi 10 already has a split IMC. If you look at a die topography of Navi 10, there are two IMCs, one at the top and bottom of the die. It's unlikely that Navi 12 retains the top IMC but I speculate it was instead moved next to the bottom IMC, which is part of the long thin bar on the bottom of the V520 die.

 

[mczkrt]

It's 2048 bit per AMD.

https://www.amd.com/en/products/server-accelerators/amd-radeon-pro-v520#product-specs

 

[laceflower]

That's my bad lol
Has anyone attempted enabling HBCC or even seeing if its available? I know it's mostly driver-level but it was advertised as a feature on the MI25/50. I think it's likely that there's a significant amount of stuff on the bios level, but I'm not a Vega bios dev/modder so I should ask around

Its possible that HBCC needs an additional control hardware that may be missing on the V520/540 but yeah, idk

 

[laceflower]

V540s now appear to be out of stock. Grats to everyone who got a hold of one!

 

[LabRat 891]

V540s now appear to be out of stock. Grats to everyone who got a hold of one!

Sadness.

Thanks for the update.

 

[PixelButts]

Theres still a few for sale on the good ol ebay, one from CISCO of all things (comes with power connectors too) but it does look to be that 92 were sold or so from that single seller.

Grats everyone

 

[Beatyoubeach]

Theres still a few for sale on the good ol ebay, one from CISCO of all things (comes with power connectors too) but it does look to be that 92 were sold or so from that single seller.

Grats everyone

fun fact that's where I got mine