[Merutsu]

Edited: Merutsu, I have added a second inmage of your table with the pointers and the pointed voltages at the vddc table, as you can see you are right about that some voltages of the vddc table are used for UVD an VCE clocks, but I dont see a 1.150 voltage, the highest one is 1.100v.

Thanks =)
It's because I've changed almost all values in these tables. I am using this modded bios with my nitro rx580. By default there was max 1150mv (7E 04) in the 2nd table and I got random voltage spikes up to 1150mv during video playback on twitch 1080p60 with HW acceleration enabled. Now it's 1100mv and I think I can lower it to 1000mv, but haven't tested yet.

 

[mynm]

Thanks =)
It's because I've changed almost all values in these tables. I am using this modded bios with my nitro rx580. By default there was max 1150mv (7E 04) in the 2nd table and I got random voltage spikes up to 1150mv during video playback on twitch 1080p60 with HW acceleration enabled. Now it's 1100mv and I think I can lower it to 1000mv, but haven't tested yet.

Ok so it seems that these voltages into VDDC voltage table are used for UVD D-clock, UVD V-clock, VCE clocks, etc, and for per dpm memory vddc voltage floor. So maybe you could help korakios better than me because I don't have a polaris gpu to test.

There are also other voltage offsets for vddgfx voltages into MM dependency table, but I don't know for what are the vddgfx voltages used or if they are used, so I think is better to not to change the vddgfx voltages table or the MM dependency table offsets.

 

[Merutsu]

There are also other voltage offsets for vddgfx voltages into MM dependency table, but I don't know for what are the vddgfx voltages used or if they are used, so I think is better to not to change the vddgfx voltages table or the MM dependency table offsets.

I don't see any offsets for VDDGFX (not sure if this table is used at all). I see pointers (inside MM dependency table) that point to voltages from the 2nd table (marked by a green rectangle on my screenshot) and offsets after them: 00 00, B4 FF, 9B FF, 82 FF etc. I am making a very similar pic to what you have posted, but it is a bit different and still unfinished...

btw, some offset values are pretty weird in the MM dependency table. 1st pointer is pointing to DPM0 that has no offset and use 800mV (20 03), but 2nd pointer is pointing to DPM1 that has -75mV offset for 850mV (52 03) which means 775mV - this is lower than DPM0 voltage. Even 7th pointer is pointing to DPM6 that has -250mV offset for 1100mv (4C 04), so actual voltage is 850mV, but 8th pointer is pointing to DPM7 1150mV (7E 04) without offset - that's why I am getting voltage spikes up to 1150mv with default values.

I think it's okay to remove all offsets and modify some pointers since I am using static voltages for all DPM states and have tested everything for stability. So i will check what clocks and offsets are used for pointed voltages and do some changes if needed, but i am too lazy so no idea when I do that

 

[mynm]

I don't see any offsets for VDDGFX (not sure if this table is used at all). I see pointers (inside MM dependency table) that point to voltages from the 2nd table (marked by a green rectangle on my screenshot) and offsets after them: 00 00, B4 FF, 9B FF, 82 FF etc. I am making a very similar pic to what you have posted, but it is a bit different and still unfinished...

btw, some offset values are pretty weird in the MM dependency table. 1st pointer is pointing to DPM0 that has no offset and use 800mV (20 03), but 2nd pointer is pointing to DPM1 that has -75mV offset for 850mV (52 03) which means 775mV - this is lower than DPM0 voltage. Even 7th pointer is pointing to DPM6 that has -250mV offset for 1100mv (4C 04), so actual voltage is 850mV, but 8th pointer is pointing to DPM7 1150mV (7E 04) without offset - that's why I am getting voltage spikes up to 1150mv with default values.

I think it's okay to remove all offsets and modify some pointers since I am using static voltages for all DPM states and have tested everything for stability. So i will check what clocks and offsets are used for pointed voltages and do some changes if needed, but i am too lazy so no idea when I do that

MM dependency table is pointing to voltages 8 to 15 in VDDC voltage table like you said are the voltage for the UVD D-clock, UVD V-clock, VCE clocks, etc.

00 00, B4 FF, 9B FF, 82 FF, 69 FF, 37 FF, 05 FF, 00 00 are the Vddgfx offsets are words so they are: 00 mv ,-76 mv, -100 mv, -126 mv, -151 mv, -251 mv, 00 mv are votage offsets. (I suppose them are subtracting vddgfx voltage, not vddc voltages) this is wrong, sorry), but vddgfx as far as i know can't be monitored with any software or maybe isn't used.

Edited: I just remember that Vddgfx offsets are sustrating the vddc values with tonga gpus, so Vddgfx offsets are subtracting those mv to voltages 8 to 15 in VDDC voltage table to be used for vddgfx voltages.

Edited: Yes in tonga Vddgfx offsets are sustrating the vddc like you can see in the new attached image, but in tonga gpus vddgfx voltages are the voltages for the core and vddc voltages seems to be vddci, but I am not sure.

You can see with Vento041's BIOS Editor as I said the names for the values, like you can see in the image:

 

[Merutsu]

(I suppose them are subtracting vddgfx voltage, not vddc voltages) this is wrong, sorry)

Yes. Have changed offsets in MMdep table and flashed the modded bios - now voltage spikes when watching online video 1080p60 is only up to 950mV (was 1150mV with stock bios). Everything works as was expected =)

 

[mynm]

Yes. Have changed offsets in MMdep table and flashed the modded bios - now voltage spikes when watching online video 1080p60 is only up to 950mV (was 1150mV with stock bios). Everything works as was expected =)

So do you have changed the Vddgfxoffsets and now the voltages are lowered? so for example do you have added a -201 mv offset to the last Vddgfxoffsets where ucVddcIndex is pointing to the 15 voltage into vddc table that is 1.150v, and now is 950mv?.

 

[Merutsu]

So do you have changed the Vddgfxoffsets and now the voltages are lowered? so for example do you have added a -201 mv offset to the last Vddgfxoffsets where ucVddcIndex is pointing to the 15 voltage into vddc table that is 1.150v, and now is 950mv?.

I don't understand why you call offsets in a MMdep table as vddgfxoffsets because you have marked as vddgfx table on my screenshot table that actually is not used at all =)
But yes - that's what i did.

Here is a simple table that shows what I expected from changing offsets and how offsets should work. Confirm, it works that way. Now the GPU voltage during online video 1080p60fps playback isn't going above 950mV (with stock bios there was spikes up to 1150mV). The last two voltages in a table are 1075mv and 1100mv because I've changed them before. It's 1100mv and 1150mv in stock bios.

 

[mynm]

I don't understand why you call offsets in a MMdep table as vddgfxoffsets because you have marked as vddgfx table on my screenshot table that actually is not used at all =)
But yes - that's what i did.

Here is a simple table that shows what I expected from changing offsets and how offsets should work. Confirm, it works that way. Now the GPU voltage during online video 1080p60fps playback isn't going above 950mV (with stock bios there was spikes up to 1150mV). The last two voltages in a table are 1075mv and 1100mv because I've changed them before. It's 1100mv and 1150mv in stock bios.

Ok so it is using the vddgfxoffsets.

I don't call it vddgfxoffsets is named like this in Vento041's bios editor and I suppose he get the info from the linux kernel info: https://github.com/libos-nuse/net-n...s/gpu/drm/amd/powerplay/hwmgr/tonga_pptable.h .

There marked as vddgfx table is also a table named vddgfxlookup, I am not sure what is it doing, maybe it is not used, but you have changed the values on it, so I have marked those values apart of the Vddc ones.

I am not sure if these vddgfx table values are used or not, or if is used for vddci values. I see some polaris gpus with a 81022 voltage controller that have a second vddci voltage monitored with Hwinfo64: https://www.overclock.net/forum/25870177-post244.html , but I have no idea if this vddc voltage is related to the vddgfx table. In tonga these monitored vddc and vddci values can be changed with vddc and vddgfx tables. But in polaris gpus as far I know nobody with a 81022 have tested this.

 

[Merutsu]

This is a picture what I made for myself when I started editing the bios and didn't found how to change some voltages in PBE. Maybe it will be helpful to someone.
The nitro rx580's stock bios has UVD DPM6 voltage 850mV (because of offset -250mV) and UVD DPM7 voltage 1150mV (no offset) what causes voltage spikes up to 1150mV during video playback. Changing voltage and(or) offset fixes it.

upd: have changed the picture a bit

 

[mynm]

I don't understand why you call offsets in a MMdep table as vddgfxoffsets because you have marked as vddgfx table on my screenshot table that actually is not used at all =)
But yes - that's what i did.

Here is a simple table that shows what I expected from changing offsets and how offsets should work. Confirm, it works that way. Now the GPU voltage during online video 1080p60fps playback isn't going above 950mV (with stock bios there was spikes up to 1150mV). The last two voltages in a table are 1075mv and 1100mv because I've changed them before. It's 1100mv and 1150mv in stock bios.

OK so you have edited this post with this image https://www.overclock.net/forum/attachment.php?attachmentid=230582&stc=1&d=1541893373

So the vddgfxoffset are substracting the pointed vddc voltages to get the voltages for the UVD D-clock, UVD V-clock, VCE clocks, etc.

This is a picture what I made for myself when I started editing the bios and didn't found how to change some voltages in PBE. Maybe it will be helpful to someone.
The nitro rx580's stock bios has UVD DPM6 voltage 850mV (because of offset -250mV) and UVD DPM7 voltage 1150mV (no offset) what causes voltage spikes up to 1150mV during video playback. Changing voltage and(or) offset fixes it.

I have edited your image with what I think the values are according to Vento041's bios editor names and my tests with my 380. Please, see your bios with Vento041's bios editor to se the names for the values.

 

[generaleramon]

is it possible to use other values for the "GPU Vcore offset for each DPM"? let's say that i want to use a -50mv offset on the DPM 5.
Great info in the screenshot, i managed to use 0.7v for both idle and light videos (720-1080p). The card (reference RX480) runs with the fan OFF at around 32-35 C*

 

[dsanke]

rx 570 to rx 580 2048sp : search df 67 and change them to df 6f via hex editor ( 2 places ) , change 0x7d to 0xfd at offset 0xd5 , fix checksum. gop will not work , and driver need patch . ( rx 580 might become rx 590 by this way but i dont see rx 590 section in inf file of driver 18.11.1 )
after flash to rx580 2048sp , atiflash 2.84 will not work too ( it will say no adapter found ), search df 67 in atiflash.exe and change them to df 6f via hex editor ( 2 places ) , then atiflash will resume work .
test with my rx470 4g ( using msi rx 580 armor oc bios ).

 

[ku4eto]

rx 570 to rx 580 2048sp : search df 67 and change them to df 6f via hex editor ( 3 places ) , fix checksum. gop will not work , and driver need patch . ( rx 580 might become rx 590 by this way but i dont see rx 590 section in inf file of driver 18.11.1 )
after flash to rx580 2048sp , atiflash 2.84 will not work too ( it will say no adapter found ), search df 67 in atiflash.exe and change them to df 6f via hex editor ( 2 places ) , then atiflash will resume work .
test with my rx470 4g ( using msi rx 580 armor oc bios ).

You are just changing the Device ID. You are not doing anything else, but renaming the GPU. The Total Shader count is still a *70 amount.

 

[dsanke]

rx 570 to rx 580 2048sp : search df 67 and change them to df 6f via hex editor ( 3 places ) , fix checksum. gop will not work , and driver need patch . ( rx 580 might become rx 590 by this way but i dont see rx 590 section in inf file of driver 18.11.1 )
after flash to rx580 2048sp , atiflash 2.84 will not work too ( it will say no adapter found ), search df 67 in atiflash.exe and change them to df 6f via hex editor ( 2 places ) , then atiflash will resume work .
test with my rx470 4g ( using msi rx 580 armor oc bios ).

You are just changing the Device ID. You are not doing anything else, but renaming the GPU. The Total Shader count is still a *70 amount.

you're right. i think amd just doing the same thing.

 

[ku4eto]

you're right. i think amd just doing the same thing.

No, they are not... The RX 580 cards have 2304 Shaders, while the RX 570 have 2048. They are locking shaders on hardware level, not via BIOS.

 

[dsanke]

you're right. i think amd just doing the same thing.

No, they are not... The RX 580 cards have 2304 Shaders, while the RX 570 have 2048. They are locking shaders on hardware level, not via BIOS.

https://videocardz.com/78588/amd-launches-radeon-rx-580-with-2048-stream-processors-in-china

 

[ku4eto]

https://videocardz.com/78588/amd-launches-radeon-rx-580-with-2048-stream-processors-in-china

Oh... those... I am not looking at anything thats targeted for the Asian market. Sorry China :/

But anyway, its the AIB makers that are making thte BIOS and labeling it RX580, while in fact its a RX570 core. Go figure. Literally the only difference between the 570 and 580 is the amount of shaders (core configuration). Even if its named 580, it is still a 570.

 

[mynm]

is it possible to use other values for the "GPU Vcore offset for each DPM"? let's say that i want to use a -50mv offset on the DPM 5.
Great info in the screenshot, i managed to use 0.7v for both idle and light videos (720-1080p). The card (reference RX480) runs with the fan OFF at around 32-35 C*

Acording to my test with my 380 it can be changed, but they are subtracting vddc voltages to get vddci voltages. But The 380 is using the vddgfxlookup table for the vddc voltages, and Polaris gpus are using the vddclookup table for the vddc voltage, so maybe is a -25 voltage offset. I can't test so wait for a @Merutsu response.

I remember to see here only some polaris users like @Ansau or @chris89 testing these vddcoffsets years ago, but they didn't say that vddcoffsets were Vcore offset for each DPM.

 

[Merutsu]

is it possible to use other values for the "GPU Vcore offset for each DPM"? let's say that i want to use a -50mv offset on the DPM 5.

For -50mV replace "E6 FF" with "CD FF".
I think it's possible, but I didn't test how they work. Since I use static voltages for all DPM states I just removed these offsets.

 

[Merutsu]

so maybe is a -25 voltage offset. I can't test so wait for a @Merutsu response.

MMdep table has the same structure. For each DPM state it has:
[voltage pointer][offset][gpu clock].

Here the same, pointers are used (00, 01, 02, 03 etc, not marked on my screenshot since I never planned to change them for this table). 99.9% it's a voltage offset for gpu and it works the same way as offset in mmdep table.