New to Overclock.net
Join Date: Aug 2016
Posts: 30
Rep: 1
(Unique: 1)
|
|
Software for R9 285 bios edit
|
|
Thread Tools |
I don't think it is a driver issue, this card accepts the drivers quite well, and I usually do a clean install using ddu and ccleaner. Performance numbers are similar of what I use to see on the internet in benchmarks.
My case is a thermaltake v3, quite old and not so cool anymore, but not critical. While my idle temps are a bit on the hotter side, between 45 ~ 55ºC in hot days, load temps are always under 80ºC. Maybe I just lost in the silicon lottery this time...
My case is a thermaltake v3, quite old and not so cool anymore, but not critical. While my idle temps are a bit on the hotter side, between 45 ~ 55ºC in hot days, load temps are always under 80ºC. Maybe I just lost in the silicon lottery this time...
| Sponsored Links | |||
|
Advertisement |
|
||
New to Overclock.net
Join Date: Jan 2016
Posts: 623
Rep: 54
(Unique: 20)
Warning: Spoiler! (Click to show)
I have done some more test with the ASIC_ProfilingInfo table, and with a vddci reduction to 837mv: https://www.overclock.net/t/1630601/mod-sapphire-rx-480-nitro-oc-better-custom-bios/30#post_26120982 , and some memory timings to: 77713320000000006BBD572F40550F0E2892F7060048C50033BB9D084C0D14205A8900A000000120100C20246F1E2912 (H5GC4H24AJR), plus more things.
And now I get with 1125 core and 1475 memory and Load Line slope to 1 (12,9%) I get a 10160 graphics score in firestrike with 17.8.2 drivers: Warning: Spoiler! (Click to show)
With the 17.10.1 drivers I get less score 10117 Warning: Spoiler! (Click to show)
I'm using this clocks amd voltages for the bios: Warning: Spoiler! (Click to show)
And I'm using the wattman % OC and OverdriveNTool to set the clocks to 1125 1.137v like this: Warning: Spoiler! (Click to show)
I'm using the ASIC_ProfilingInfo table to change the voltages, and seems to be better to set the voltage to 1.193v to the state 7 and then block the state 6 and 7 to 1125 and 1.137v than to have 1.137v at state 7.
This is the ASIC_ProfilingInfo table I'm using: Asicprof375b.zip 0k .zip file
These are my stock and modded bioses: 308A0500.zip 41k .zip file
mod.zip 41k .zip file
You can compare the bioses to see the changes, and I could explain the things I have done if somebody want.
Quote:
Originally Posted by mynm 
@ZealotKi11er Do you still have the 285X? I have found the table where all the EVV voltages are calculatred. Is the ASIC_ProfilingInfo table.
I see that in the 285X is like this:
Warning: Spoiler! (Click to show)
And I have tested it with your clocks with my 380 and I see at wattman an 1.225v fot states 4 to 7. Do you see that?
If you want you could test the table of the 380(X) and 285 I see lower voltages with them:
285:
Warning: Spoiler! (Click to show)
380:
Warning: Spoiler! (Click to show)
I have tested as well an 390X table and it's working:
Warning: Spoiler! (Click to show)
Fiji one isn't working and It's a double table, I have tested the first half that have the same 3.3 version but it isn't working.
This is the structure for the table:
Warning: Spoiler! (Click to show)
You could see it here: https://www.google.es/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&cad=rja&uact=8&ved=0ahUKEwjexbut557UAhVEY1AKHV6uAakQFggvMAE&url=https%3A%2F%2Fgithub.com%2Fipts-linux-org%2Fipts-linux%2Fblob%2Fmaster%2Fdrivers%2Fgpu%2Fdrm%2Famd%2Finclude%2Fatombios.h&usg=AFQjCNE5ym-YkQeMOWpSdFTF4mpLKJTIuQ
This is a comparison between 285X 380(X) 285 tables:
Warning: Spoiler! (Click to show)
The gray highlighted values are changing the voltages lower values are lower voltages, higher values are higher voltages. I only have tested to lower them. Edit: I don't have tested the gray highlighted 33 37 and 43. Edited: about the 33, 37 and 43, I have tested the value to 43 from 37 and dpm7 voltage is increased 12mv, with 0 is reduced 18mv.
About ULONG ulMaxVddc; ULONG ulMinVddc (Edited: the bolded ones at the tables) if some one want to test them are the max and the min EVV voltages, they are the voltage x4 in hesadecimal. And it have to be rounded but I don't know now hexadecimal numbers are working, for 1237mv is working with 4950 not with 4948 and I don't have tested with 4949, all are the same /4 value of 1237 in hexadecimal. So maybe it have to be rounded to the nearest 5 or 0 ending value to be working.
Edited: ulMinVddc can be used to undervolt, dpm0 will be the same an dpm1 to 7 or 4 to 7 will be reduced. If dpm6 voltage is the same as dpm7 after that you could reduce dpm 6 clock to a lower clock to reduce the voltage.
If you change the min VDDC you have to change DMP0 VDDC to the same value and one offset to change the wattman memory voltage, if not wattman isn't working and AB is only working with or without powerplay support.
@ZealotKi11er Do you still have the 285X? I have found the table where all the EVV voltages are calculatred. Is the ASIC_ProfilingInfo table.
I see that in the 285X is like this:
Warning: Spoiler! (Click to show)
E0 00 03 03 55 00 00 00 10 04 00 00 24 13 00 00 10 0E 00 00 09 00 00 00 E8 03 00 00 48 0D 00 00 48 02 08 08 B8 0B 00 00 00 00 00 00 48 02 10 08 10 27 00 00 00 00 00 00 4C 02 00 08 A0 0F 00 00 B8 0B 00 00 4C 02 10 08 AC 0D 00 80 D6 06 00 00 4C 02 18 08 40 9C 00 00 50 C3 00 00 34 00 06 0A D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00 01 00 A8 02 2A 03 AC 03 2E 04 B0 04 32 05 AA 05 4A 09 00 00 4A 09 00 00 4A 09 00 00 4A 09 00 00 4A 09 00 00 4A 09 00 00 4A 09 00 00 4A 09 00 00 48 02 18 08 B6 03 00 00 46 05 00 00 61 93 00 00 AE 23 00 00 38 0D 01 00 D9 F4 00 00 00 00 00 00 00 00 00 00 33 00 00 00 00 00 00 00 48 01 00 00 00 00 00 00 3C 00 00 00 00 00 00 00 9B 02 00 00
And I have tested it with your clocks with my 380 and I see at wattman an 1.225v fot states 4 to 7. Do you see that?
If you want you could test the table of the 380(X) and 285 I see lower voltages with them:
285:
Warning: Spoiler! (Click to show)
E0 00 03 03 55 00 00 00 10 04 00 00 24 13 00 00 10 0E 00 00 09 00 00 00 E8 03 00 00 48 0D 00 00 48 02 08 08 B8 0B 00 00 00 00 00 00 48 02 10 08 10 27 00 00 00 00 00 00 4C 02 00 08 A0 0F 00 00 B8 0B 00 00 4C 02 10 08 AC 0D 00 80 D6 06 00 00 4C 02 18 08 40 9C 00 00 50 C3 00 00 34 00 06 0A D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00 00 00 E4 02 48 03 AC 03 1A 04 7E 04 E2 04 46 05 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 48 02 18 08 B6 03 00 00 46 05 00 00 64 36 00 00 D8 36 00 00 40 9D 00 00 79 8B 00 00 00 00 00 00 00 00 00 00 43 00 00 00 00 00 00 00 84 00 00 00 00 00 00 00 3C 00 00 00 00 00 00 00 9B 02 00 00
380:
Warning: Spoiler! (Click to show)
E0 00 03 03 55 00 00 00 10 04 00 00 24 13 00 00 10 0E 00 00 09 00 00 00 E8 03 00 00 48 0D 00 00 48 02 08 08 B8 0B 00 00 00 00 00 00 48 02 10 08 10 27 00 00 00 00 00 00 4C 02 00 08 A0 0F 00 00 B8 0B 00 00 4C 02 10 08 AC 0D 00 80 D6 06 00 00 4C 02 18 08 40 9C 00 00 50 C3 00 00 34 00 06 0A D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00 00 00 76 02 EE 02 5C 03 D4 03 42 04 BA 04 28 05 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00 48 02 18 08 B6 03 00 00 46 05 00 00 BE 2C 00 00 7A 36 00 00 1E 8F 00 00 AF 81 00 00 00 00 00 00 00 00 00 00 37 00 00 00 00 00 00 00 84 00 00 00 00 00 00 00 3C 00 00 00 00 00 00 00 9B 02 00 00
I have tested as well an 390X table and it's working:
Warning: Spoiler! (Click to show)
E0 00 03 03 64 00 00 00 1A 04 00 00 EC 13 00 00 10 0E 00 00 08 00 00 00 B6 03 00 00 48 0D 00 00 94 00 1B 08 B8 0B 00 00 00 00 00 00 98 00 03 08 10 27 00 00 00 00 00 00 98 00 13 08 A0 0F 00 00 B8 0B 00 00 9C 00 03 08 AC 0D 00 80 D6 06 00 00 9C 00 0B 08 40 9C 00 00 50 C3 00 00 34 00 1A 0A D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00 00 00 10 04 EC 04 BE 05 9A 06 6C 07 48 08 1A 09 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 A1 0A 00 00 98 00 0B 08 EE 02 00 00 E2 04 00 00 14 75 00 00 ED 02 00 00 D9 F9 00 00 1C D8 00 00 00 00 00 00 00 00 00 00 32 00 00 00 00 00 00 00 9B 00 00 00 00 00 00 00 4B 00 00 00 00 00 00 00 9B 02 00 00
Fiji one isn't working and It's a double table, I have tested the first half that have the same 3.3 version but it isn't working.
This is the structure for the table:
Warning: Spoiler! (Click to show)
Code:
// for Tonga/Fiji speed EVV algorithm
typedef struct _ATOM_ASIC_PROFILING_INFO_V3_3
{
ATOM_COMMON_TABLE_HEADER asHeader;
ULONG ulEvvLkgFactor;
ULONG ulBoardCoreTemp;
ULONG ulMaxVddc;
ULONG ulMinVddc;
ULONG ulLoadLineSlop;
ULONG ulLeakageTemp;
ULONG ulLeakageVoltage;
EFUSE_LINEAR_FUNC_PARAM sCACm;
EFUSE_LINEAR_FUNC_PARAM sCACb;
EFUSE_LOGISTIC_FUNC_PARAM sKt_b;
EFUSE_LOGISTIC_FUNC_PARAM sKv_m;
EFUSE_LOGISTIC_FUNC_PARAM sKv_b;
USHORT usLkgEuseIndex;
UCHAR ucLkgEfuseBitLSB;
UCHAR ucLkgEfuseLength;
FA0 ULONG ulLkgEncodeLn_MaxDivMin;
ULONG ulLkgEncodeMax;
ULONG ulLkgEncodeMin;
ULONG ulEfuseLogisticAlpha;
union{
USHORT usPowerDpm0;
USHORT usParamNegFlag; //bit0 =1 :indicate ulRoBeta is Negative, bit1=1 indicate Kv_m max is postive
};
USHORT usPowerDpm1;
USHORT usPowerDpm2;
USHORT usPowerDpm3;
USHORT usPowerDpm4;
USHORT usPowerDpm5;
USHORT usPowerDpm6;
USHORT usPowerDpm7;
ULONG ulTdpDerateDPM0;
ULONG ulTdpDerateDPM1;
ULONG ulTdpDerateDPM2;
ULONG ulTdpDerateDPM3;
ULONG ulTdpDerateDPM4;
ULONG ulTdpDerateDPM5;
ULONG ulTdpDerateDPM6;
ULONG ulTdpDerateDPM7;
EFUSE_LINEAR_FUNC_PARAM sRoFuse;
ULONG ulRoAlpha;
ULONG ulRoBeta;
ULONG ulRoGamma;
ULONG ulRoEpsilon;
ULONG ulATermRo;
ULONG ulBTermRo;
ULONG ulCTermRo;
ULONG ulSclkMargin;
ULONG ulFmaxPercent;
ULONG ulCRPercent;
ULONG ulSFmaxPercent;
ULONG ulSCRPercent;
ULONG ulSDCMargine;
}ATOM_ASIC_PROFILING_INFO_V3_3;
You could see it here: https://www.google.es/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&cad=rja&uact=8&ved=0ahUKEwjexbut557UAhVEY1AKHV6uAakQFggvMAE&url=https%3A%2F%2Fgithub.com%2Fipts-linux-org%2Fipts-linux%2Fblob%2Fmaster%2Fdrivers%2Fgpu%2Fdrm%2Famd%2Finclude%2Fatombios.h&usg=AFQjCNE5ym-YkQeMOWpSdFTF4mpLKJTIuQ
This is a comparison between 285X 380(X) 285 tables:
Warning: Spoiler! (Click to show)
The gray highlighted values are changing the voltages lower values are lower voltages, higher values are higher voltages. I only have tested to lower them. Edit: I don't have tested the gray highlighted 33 37 and 43. Edited: about the 33, 37 and 43, I have tested the value to 43 from 37 and dpm7 voltage is increased 12mv, with 0 is reduced 18mv.
About ULONG ulMaxVddc; ULONG ulMinVddc (Edited: the bolded ones at the tables) if some one want to test them are the max and the min EVV voltages, they are the voltage x4 in hesadecimal. And it have to be rounded but I don't know now hexadecimal numbers are working, for 1237mv is working with 4950 not with 4948 and I don't have tested with 4949, all are the same /4 value of 1237 in hexadecimal. So maybe it have to be rounded to the nearest 5 or 0 ending value to be working.
Edited: ulMinVddc can be used to undervolt, dpm0 will be the same an dpm1 to 7 or 4 to 7 will be reduced. If dpm6 voltage is the same as dpm7 after that you could reduce dpm 6 clock to a lower clock to reduce the voltage.
If you change the min VDDC you have to change DMP0 VDDC to the same value and one offset to change the wattman memory voltage, if not wattman isn't working and AB is only working with or without powerplay support.
I have done some more test with the ASIC_ProfilingInfo table, and with a vddci reduction to 837mv: https://www.overclock.net/t/1630601/mod-sapphire-rx-480-nitro-oc-better-custom-bios/30#post_26120982 , and some memory timings to: 77713320000000006BBD572F40550F0E2892F7060048C50033BB9D084C0D14205A8900A000000120100C20246F1E2912 (H5GC4H24AJR), plus more things.
And now I get with 1125 core and 1475 memory and Load Line slope to 1 (12,9%) I get a 10160 graphics score in firestrike with 17.8.2 drivers: Warning: Spoiler! (Click to show)
With the 17.10.1 drivers I get less score 10117 Warning: Spoiler! (Click to show)
I'm using this clocks amd voltages for the bios: Warning: Spoiler! (Click to show)
And I'm using the wattman % OC and OverdriveNTool to set the clocks to 1125 1.137v like this: Warning: Spoiler! (Click to show)
I'm using the ASIC_ProfilingInfo table to change the voltages, and seems to be better to set the voltage to 1.193v to the state 7 and then block the state 6 and 7 to 1125 and 1.137v than to have 1.137v at state 7.
This is the ASIC_ProfilingInfo table I'm using: Asicprof375b.zip 0k .zip file
These are my stock and modded bioses: 308A0500.zip 41k .zip file
mod.zip 41k .zip file
You can compare the bioses to see the changes, and I could explain the things I have done if somebody want.
Chief R.B.R.T. (Red BIOS Rebellion Team)
ASIC profiling voltages mod for 380 + CUs lock ; ASIC profiling voltages mod for polaris gpus; WattMan memory voltage explanations; Adding VOI control for ncp 81022 for polaris gpus; Load Line + offset voltage VOI mod via svid2 for 380; Load Line + offset voltage VOI mod via svid2 for polaris; Load Line + offset voltage VOI mod via svid2 for hawaii ; Voltage offset for 380 gpus with an ncp 81022
@integralfx
Join Date: Feb 2014
Posts: 197
Rep: 8
(Unique: 7)
| Sponsored Links | |||
|
Advertisement |
|
||
New to Overclock.net
Join Date: Jan 2016
Posts: 623
Rep: 54
(Unique: 20)
Quote:
Hi, yes reducing TDP values can reduce VRM temperatures. I see Chris89 have lowerd tdp and tdc values to 128W and 128A and the max tjmax temp to 75º. Maybe theses values are low, you will see throttling, I need 176W to 185W and ~ 180A for 1125mhz core 1475mhz memory.
The 0MHz core and memory clock issue, maybe is related to same values that some Asus bioses have at the end of the bios, and that must be changed to do to work the mhz changes. I don't have tested to change these values to see if it is working. But according to my tests, changing the mhz values to more than 1040 in the bios is not worth it, is better to OC with software.
As you have a R3567B voltage controller there is other way to reduce the VRM temp by setteing a VRM max temp. Is explained in one gupsterg's pdf I gave you:
You have to change registers:
0x31 = OTP, bits [4:0] of byte is threshold in °C, range is +0°C to 32°C, (5 bits so max 31°C ??, also PDF show max VRHot+OTP 134°C??).
0x32 = VR_Hot, bits [7:2] of byte is temp_max (64°C+data value), range is 64°C to 127°C, (6 bits so max 63 integer), to read register take
hexadecimal byte, convert to binary, bits [7:2] convert to hexadecimal.
You can test to change that registers with Afterburner and see if them are working and are changed in the i2c dump.
But I don't know well how is it working. Maybe somebody in hawaii bios mod or fiji bios mod thread could help you better.
Chief R.B.R.T. (Red BIOS Rebellion Team)
ASIC profiling voltages mod for 380 + CUs lock ; ASIC profiling voltages mod for polaris gpus; WattMan memory voltage explanations; Adding VOI control for ncp 81022 for polaris gpus; Load Line + offset voltage VOI mod via svid2 for 380; Load Line + offset voltage VOI mod via svid2 for polaris; Load Line + offset voltage VOI mod via svid2 for hawaii ; Voltage offset for 380 gpus with an ncp 81022
@integralfx
Join Date: Feb 2014
Posts: 197
Rep: 8
(Unique: 7)
So I decided to swap out the stock fan with CM JetFlo 120mm fans.
The core temperatures are 5°C lower, but that might be because I swapped out the thermal paste. VRM temperatures are still the same (80-90°C at 1.15v VDDC), so most likely the stock thermal pads on the VRMs aren't that great. I'm considering getting some better thermal pads, particularly these FujiPoly ones. Maybe that will reduce the VRM temperatures a fair bit.
The core temperatures are 5°C lower, but that might be because I swapped out the thermal paste. VRM temperatures are still the same (80-90°C at 1.15v VDDC), so most likely the stock thermal pads on the VRMs aren't that great. I'm considering getting some better thermal pads, particularly these FujiPoly ones. Maybe that will reduce the VRM temperatures a fair bit.
New to Overclock.net
Join Date: Jun 2017
Posts: 14
Rep: 0
I did that and it's okay but i want to overclock memory clock @1650 and how to monitor VRM Temps ?
My PC Shutdowns Sometime in Furemark Benchmark Test so Whats the problem Here ? and it would hurt the Card In Gaming as It Never Shuts down in Games what ever title it's ??
My PC Shutdowns Sometime in Furemark Benchmark Test so Whats the problem Here ? and it would hurt the Card In Gaming as It Never Shuts down in Games what ever title it's ??
@integralfx
Join Date: Feb 2014
Posts: 197
Rep: 8
(Unique: 7)
Quote:
Originally Posted by Kanoti1
I did that and it's okay but i want to overclock memory clock @1650 and how to monitor VRM Temps ?
My PC Shutdowns Sometime in Furemark Benchmark Test so Whats the problem Here ? and it would hurt the Card In Gaming as It Never Shuts down in Games what ever title it's ??
I did that and it's okay but i want to overclock memory clock @1650 and how to monitor VRM Temps ?
My PC Shutdowns Sometime in Furemark Benchmark Test so Whats the problem Here ? and it would hurt the Card In Gaming as It Never Shuts down in Games what ever title it's ??
I use HWINFO64 but some people use GPU-Z. GPU-Z doesn't show VRM temperatures for me, though.
You shouldn't be using FurMark to stress. It really heats up your VRMs to temperatures that you won't experience in real world scenarios, which is probably why your GPU is shutting down.
Use Unigine Heaven or SuperPosition to stress your overclock.
New to Overclock.net
Join Date: May 2008
Posts: 2,523
Rep: 84
(Unique: 62)
Quote:
Originally Posted by BLUuuE
So I decided to swap out the stock fan with CM JetFlo 120mm fans.
The core temperatures are 5°C lower, but that might be because I swapped out the thermal paste. VRM temperatures are still the same (80-90°C at 1.15v VDDC), so most likely the stock thermal pads on the VRMs aren't that great. I'm considering getting some better thermal pads, particularly these FujiPoly ones. Maybe that will reduce the VRM temperatures a fair bit.
So I decided to swap out the stock fan with CM JetFlo 120mm fans.
The core temperatures are 5°C lower, but that might be because I swapped out the thermal paste. VRM temperatures are still the same (80-90°C at 1.15v VDDC), so most likely the stock thermal pads on the VRMs aren't that great. I'm considering getting some better thermal pads, particularly these FujiPoly ones. Maybe that will reduce the VRM temperatures a fair bit.
https://www.ebay.com/itm/NEW-Thermagon-thermal-gap-filler-pad-T-PLI-2200-A1-12mm-x-12mm-x-5mm-49-per-pack/182814014721?ssPageName=STRK%3AMEBIDX%3AIT&_trksid=p2060353.m2749.l2649
Dell Precision T7500 : Dual Xeon X5675 @ 3.536Ghz X 24 Threads : 48 Gigabytes : Cinebench R15 : 1505 : AMD R9 390X 8GB @ 1173Mhz 1350mv 57599 Power Limit @ 1250Mhz memory error free on tight timings / since 1500mhz+ had errors galore
New to Overclock.net
Join Date: Jan 2016
Posts: 623
Rep: 54
(Unique: 20)
Edited: with more interesting things
Hi, I found one funny thing, how to lock cores to the r9 380
.
You only have to replace the bolded 6F to 70, in the TV1OutputControl or havesting table:
D1 00 01 01 08 00 37 00 00 66 FF 2D 0D 42 BD 00
56 00 41 03 21 41 00 15 0D 41 02 2D 0A 42 41 03
04 01 00 00 66 FF 2D 0D 42 C9 00 03 05 00 00 00
00 40 56 00 40 54 00 00 00 03 4C 00 00 00 01 02
00 C2 00 01 0A 00 00 01 03 98 41 01 00 56 08 41
03 22 43 41 4C 8A 41 44 49 5F 00 2D 65 41 01 2D
25 41 01 3E 39 41 00 45 50 00 3E 71 41 00 47 A1
00 33 71 41 00 03 39 41 00 33 25 41 01 03 22 43
41 4C 8A 41 44 49 79 00 0F 8A 41 44 33 65 41 01
3E 65 41 00 49 79 00 01 1A 00 00 01 01 9A 01 00
41 2D 0D 42 02 00 2D 25 40 01 3E 25 40 04 49 39
00 01 05 00 C2 00 00 00 E0 5B 7A 14 00 6F 22 70
22 3D 26 DF 26 40 22 41 22 00 00 00 01 00 02 00
03
This is what gpuz shows:
This is what cuinfo17 shows with the stock value:
This is what cuinfo17 shows with the modded value:
You can see that a 4 CUs software lock is added .
The firestrike score is lowered ~ 700 points, power consumption is 5w less, and temps are 1º less.
I suppose CUs can't be un Unlocked.
Other thing that I know now is the structure of the ASIC_ProfilingInfo info table thanks to this:
[/URL]source: https://github.com/ipts-linux-org/ip...ude/atombios.h
//Here is parameter to convert Efuse value to Measure value
//Measured = LN((2^Bitsize-1)/EFUSE-1)*(Range)/(-alpha)+(Max+Min)/2
typedef struct _EFUSE_LOGISTIC_FUNC_PARAM
{
USHORT usEfuseIndex; // Efuse Index in DWORD address, for example Index 911, usEuseIndex=112
UCHAR ucEfuseBitLSB; // Efuse bit LSB in DWORD address, for example Index 911, usEfuseBitLSB= 911-112*8=15
UCHAR ucEfuseLength; // Efuse bits length,
ULONG ulEfuseEncodeRange; // Range = Max - Min, bit31 indicate the efuse is negative number
ULONG ulEfuseEncodeAverage; // Average = ( Max + Min )/2
}EFUSE_LOGISTIC_FUNC_PARAM;
//Linear Function: Measured = Round ( Efuse * ( Max-Min )/(2^BitSize -1 ) + Min )
typedef struct _EFUSE_LINEAR_FUNC_PARAM
{
USHORT usEfuseIndex; // Efuse Index in DWORD address, for example Index 911, usEuseIndex=112
UCHAR ucEfuseBitLSB; // Efuse bit LSB in DWORD address, for example Index 911, usEfuseBitLSB= 911-112*8=15
UCHAR ucEfuseLength; // Efuse bits length,
ULONG ulEfuseEncodeRange; // Range = Max - Min, bit31 indicate the efuse is negative number
ULONG ulEfuseMin; // Min
So for this stock table:
E0 00 03 03 55 00 00 00 10 04 00 00 24 13 00 00
10 0E 00 00 09 00 00 00 E8 03 00 00 48 0D 00 00
48 02 08 08 B8 0B 00 00 00 00 00 00 48 02 10 08
10 27 00 00 00 00 00 00 4C 02 00 08 A0 0F 00 00
B8 0B 00 00 4C 02 10 08 AC 0D 00 80 D6 06 00 00
4C 02 18 08 40 9C 00 00 50 C3 00 00 34 00 06 0A
D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00
00 00 76 02 EE 02 5C 03 D4 03 42 04 BA 04 28 05
E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00
E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00
48 02 18 08 B6 03 00 00 46 05 00 00 BE 2C 00 00
7A 36 00 00 1E 8F 00 00 AF 81 00 00 00 00 00 00
00 00 00 00 37 00 00 00 00 00 00 00 84 00 00 00
00 00 00 00 3C 00 00 00 00 00 00 00 9B 02 00 00
Is like this:
ATOM_COMMON_TABLE_HEADER asHeader;
55 ULONG ulEvvLkgFactor; 85
10 04 ULONG ulBoardCoreTemp; 1040 104º
24 13 ULONG ulMaxVddc; 4900 /4= 1225v
10 E0 ULONG ulMinVddc; 3600 /4= 900v
9 ULONG ulLoadLineSlop;
e8 03 ULONG ulLeakageTemp; 1000
48 D0 ULONG ulLeakageVoltage; 3400 /4=850v
EFUSE_LINEAR_FUNC_PARAM sCACm; (584*3000/255)= 6870 /4 1717.5
48 02 USHORT usEfuseIndex; 584
08 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
B8 0B ULONG ulEfuseEncodeRange;
00 ULONG ulEfuseMin;
EFUSE_LINEAR_FUNC_PARAM sCACb; (584*10000/255) =22901 /4= 5725.25
48 02 USHORT usEfuseIndex;
10 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
10 27 ULONG ulEfuseEncodeRange;
00 ULONG ulEfuseMin;
EFUSE_LOGISTIC_FUNC_PARAM sKt_b; LN((2^8-1)/588-1)*(4000)/(-13)+ 3000 = -0,83154950342 = 46240 /4 11560
4C02 USHORT usEfuseIndex; 588
00 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
A0 0F ULONG ulEfuseEncodeRange; 4000
B8 B0 ULONG ulEfuseEncodeAverage; 3000
EFUSE_LOGISTIC_FUNC_PARAM sKv_m; LN((2^8-1)/588-1)*(2147537068)/(-13)+1645=-0,83154950342* = 137369597 /4 = 34342399.5
4C02 USHORT usEfuseIndex;
10 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
AC 0D 00 80 ULONG ulEfuseEncodeRange; 2147537068
D6 06 ULONG ulEfuseEncodeAverage; 1645
EFUSE_LOGISTIC_FUNC_PARAM sKv_b; LN((2^8-1)/588-1)*(40000)/(-13)+50000 = 50255 /4=12563.75
4C 02 USHORT usEfuseIndex;
18 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
40 9C ULONG ulEfuseEncodeRange; 40000
50 C3 ULONG ulEfuseEncodeAverage; 50000
34 00 USHORT usLkgEuseIndex;
06 UCHAR ucLkgEfuseBitLSB;
0A UCHAR ucLkgEfuseLength;
D5 0D 01 ULONG ulLkgEncodeLn_MaxDivMin;
40 42 0F ULONG ulLkgEncodeMax; 1000000
E8 03 ULONG ulLkgEncodeMin; 1000
D ULONG ulEfuseLogisticAlpha
0000 USHORT usPowerDpm0;
0000 USHORT usParamNegFlag; //bit0 =1 :indicate ulRoBeta is Negative, bit1=1 indicate Kv_m max is postive
76 02USHORT usPowerDpm1; 630
EE 02 USHORT usPowerDpm2; 750
5C 03 USHORT usPowerDpm3; 860
D4 03 USHORT usPowerDpm4; 980
42 04 USHORT usPowerDpm5; 1090
BA 04 USHORT usPowerDpm6; 1204
28 05 USHORT usPowerDpm7; 1320
E3 08 ULONG ulTdpDerateDPM0; 2272
E3 08 ULONG ulTdpDerateDPM1;
E3 08 ULONG ulTdpDerateDPM2;
E3 08 ULONG ulTdpDerateDPM3;
E3 08 ULONG ulTdpDerateDPM4;
E3 08 ULONG ulTdpDerateDPM5;
E3 08 ULONG ulTdpDerateDPM6;
E3 08 ULONG ulTdpDerateDPM7;
EFUSE_LINEAR_FUNC_PARAM sRoFuse; (584*950)/255) + 1350= 3525 /4 = 881.25
48 02 USHORT usEfuseIndex; 584
18 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
B6 03 ULONG ulEfuseEncodeRange; max 2186
46 05 ULONG ulEfuseMin; 1350
BE 2C ULONG ulRoAlpha; 11454
7A 36 ULONG ulRoBeta; 13946
1E 8F ULONG ulRoGamma; 36638
AF 81 ULONG ulRoEpsilon; 33199
00 ULONG ulATermRo;
00 ULONG ulBTermRo;
37 ULONG ulCTermRo; 91
00 ULONG ulSclkMargin;
84 ULONG ulFmaxPercent;
00 ULONG ulCRPercent;
3c ULONG ulSFmaxPercent; 60
00 ULONG ulSCRPercent;
9b 02 ULONG ulSDCMargine; 667
I have tried to calculate what are named "Measured value" for the Efuse values, but I don't see any sense on it or I have done it wrong.
I'm using this values to change voltages:
The value 46 05 = 1350 is ulEfuseMin and is lowering in littel increases the voltages from DPM 1 to 6 in big increase is lowering DPM 1 to 7 voltages. And increasing the voltage if is lowered
EDITED: I MADE TWO MISTAKES ulCTermRo is 37 and is ulFmaxPercent not ulSFmaxPercent, sorry.
The value 37 = 55 is ulCTermRo and 84 =132 that is ulFmaxPercent ulSFmaxPercent are lowering DPM 1 to 7 voltages if are lowerd and incrasing them if are increased.
The values:
BE 2C = 11454 ulRoAlpha;
7A 30 = 13945 ulRoBeta;
1E 8F = 36638 ulRoGamma;
AF 81 =33199 ulRoEpsilon;
Are lowering DPM 1 to 7 voltages if are lowerd and incrasing them if are increased. But I have lower scores changing them.
EDITED:
B6 03 ULONG ulEfuseEncodeRange is like a voltage offset higher values are less volage for DPM 1 to 7, and lower values more moltage. + or -15 or 20 devimal values for the value is ~ + or - 6.25v, but I have to test more. I don't see performance lost.
Thanks for the person who gave me a rep+
Hi, I found one funny thing, how to lock cores to the r9 380
.You only have to replace the bolded 6F to 70, in the TV1OutputControl or havesting table:
D1 00 01 01 08 00 37 00 00 66 FF 2D 0D 42 BD 00
56 00 41 03 21 41 00 15 0D 41 02 2D 0A 42 41 03
04 01 00 00 66 FF 2D 0D 42 C9 00 03 05 00 00 00
00 40 56 00 40 54 00 00 00 03 4C 00 00 00 01 02
00 C2 00 01 0A 00 00 01 03 98 41 01 00 56 08 41
03 22 43 41 4C 8A 41 44 49 5F 00 2D 65 41 01 2D
25 41 01 3E 39 41 00 45 50 00 3E 71 41 00 47 A1
00 33 71 41 00 03 39 41 00 33 25 41 01 03 22 43
41 4C 8A 41 44 49 79 00 0F 8A 41 44 33 65 41 01
3E 65 41 00 49 79 00 01 1A 00 00 01 01 9A 01 00
41 2D 0D 42 02 00 2D 25 40 01 3E 25 40 04 49 39
00 01 05 00 C2 00 00 00 E0 5B 7A 14 00 6F 22 70
22 3D 26 DF 26 40 22 41 22 00 00 00 01 00 02 00
03
This is what gpuz shows:
This is what cuinfo17 shows with the stock value:
This is what cuinfo17 shows with the modded value:
You can see that a 4 CUs software lock is added
.The firestrike score is lowered ~ 700 points, power consumption is 5w less, and temps are 1º less.
I suppose CUs can't be un Unlocked.
Other thing that I know now is the structure of the ASIC_ProfilingInfo info table thanks to this:
[/URL]source: https://github.com/ipts-linux-org/ip...ude/atombios.h
//Here is parameter to convert Efuse value to Measure value
//Measured = LN((2^Bitsize-1)/EFUSE-1)*(Range)/(-alpha)+(Max+Min)/2
typedef struct _EFUSE_LOGISTIC_FUNC_PARAM
{
USHORT usEfuseIndex; // Efuse Index in DWORD address, for example Index 911, usEuseIndex=112
UCHAR ucEfuseBitLSB; // Efuse bit LSB in DWORD address, for example Index 911, usEfuseBitLSB= 911-112*8=15
UCHAR ucEfuseLength; // Efuse bits length,
ULONG ulEfuseEncodeRange; // Range = Max - Min, bit31 indicate the efuse is negative number
ULONG ulEfuseEncodeAverage; // Average = ( Max + Min )/2
}EFUSE_LOGISTIC_FUNC_PARAM;
//Linear Function: Measured = Round ( Efuse * ( Max-Min )/(2^BitSize -1 ) + Min )
typedef struct _EFUSE_LINEAR_FUNC_PARAM
{
USHORT usEfuseIndex; // Efuse Index in DWORD address, for example Index 911, usEuseIndex=112
UCHAR ucEfuseBitLSB; // Efuse bit LSB in DWORD address, for example Index 911, usEfuseBitLSB= 911-112*8=15
UCHAR ucEfuseLength; // Efuse bits length,
ULONG ulEfuseEncodeRange; // Range = Max - Min, bit31 indicate the efuse is negative number
ULONG ulEfuseMin; // Min
So for this stock table:
E0 00 03 03 55 00 00 00 10 04 00 00 24 13 00 00
10 0E 00 00 09 00 00 00 E8 03 00 00 48 0D 00 00
48 02 08 08 B8 0B 00 00 00 00 00 00 48 02 10 08
10 27 00 00 00 00 00 00 4C 02 00 08 A0 0F 00 00
B8 0B 00 00 4C 02 10 08 AC 0D 00 80 D6 06 00 00
4C 02 18 08 40 9C 00 00 50 C3 00 00 34 00 06 0A
D5 0D 01 00 40 42 0F 00 E8 03 00 00 0D 00 00 00
00 00 76 02 EE 02 5C 03 D4 03 42 04 BA 04 28 05
E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00
E3 08 00 00 E3 08 00 00 E3 08 00 00 E3 08 00 00
48 02 18 08 B6 03 00 00 46 05 00 00 BE 2C 00 00
7A 36 00 00 1E 8F 00 00 AF 81 00 00 00 00 00 00
00 00 00 00 37 00 00 00 00 00 00 00 84 00 00 00
00 00 00 00 3C 00 00 00 00 00 00 00 9B 02 00 00
Is like this:
ATOM_COMMON_TABLE_HEADER asHeader;
55 ULONG ulEvvLkgFactor; 85
10 04 ULONG ulBoardCoreTemp; 1040 104º
24 13 ULONG ulMaxVddc; 4900 /4= 1225v
10 E0 ULONG ulMinVddc; 3600 /4= 900v
9 ULONG ulLoadLineSlop;
e8 03 ULONG ulLeakageTemp; 1000
48 D0 ULONG ulLeakageVoltage; 3400 /4=850v
EFUSE_LINEAR_FUNC_PARAM sCACm; (584*3000/255)= 6870 /4 1717.5
48 02 USHORT usEfuseIndex; 584
08 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
B8 0B ULONG ulEfuseEncodeRange;
00 ULONG ulEfuseMin;
EFUSE_LINEAR_FUNC_PARAM sCACb; (584*10000/255) =22901 /4= 5725.25
48 02 USHORT usEfuseIndex;
10 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
10 27 ULONG ulEfuseEncodeRange;
00 ULONG ulEfuseMin;
EFUSE_LOGISTIC_FUNC_PARAM sKt_b; LN((2^8-1)/588-1)*(4000)/(-13)+ 3000 = -0,83154950342 = 46240 /4 11560
4C02 USHORT usEfuseIndex; 588
00 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
A0 0F ULONG ulEfuseEncodeRange; 4000
B8 B0 ULONG ulEfuseEncodeAverage; 3000
EFUSE_LOGISTIC_FUNC_PARAM sKv_m; LN((2^8-1)/588-1)*(2147537068)/(-13)+1645=-0,83154950342* = 137369597 /4 = 34342399.5
4C02 USHORT usEfuseIndex;
10 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
AC 0D 00 80 ULONG ulEfuseEncodeRange; 2147537068
D6 06 ULONG ulEfuseEncodeAverage; 1645
EFUSE_LOGISTIC_FUNC_PARAM sKv_b; LN((2^8-1)/588-1)*(40000)/(-13)+50000 = 50255 /4=12563.75
4C 02 USHORT usEfuseIndex;
18 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
40 9C ULONG ulEfuseEncodeRange; 40000
50 C3 ULONG ulEfuseEncodeAverage; 50000
34 00 USHORT usLkgEuseIndex;
06 UCHAR ucLkgEfuseBitLSB;
0A UCHAR ucLkgEfuseLength;
D5 0D 01 ULONG ulLkgEncodeLn_MaxDivMin;
40 42 0F ULONG ulLkgEncodeMax; 1000000
E8 03 ULONG ulLkgEncodeMin; 1000
D ULONG ulEfuseLogisticAlpha
0000 USHORT usPowerDpm0;
0000 USHORT usParamNegFlag; //bit0 =1 :indicate ulRoBeta is Negative, bit1=1 indicate Kv_m max is postive
76 02USHORT usPowerDpm1; 630
EE 02 USHORT usPowerDpm2; 750
5C 03 USHORT usPowerDpm3; 860
D4 03 USHORT usPowerDpm4; 980
42 04 USHORT usPowerDpm5; 1090
BA 04 USHORT usPowerDpm6; 1204
28 05 USHORT usPowerDpm7; 1320
E3 08 ULONG ulTdpDerateDPM0; 2272
E3 08 ULONG ulTdpDerateDPM1;
E3 08 ULONG ulTdpDerateDPM2;
E3 08 ULONG ulTdpDerateDPM3;
E3 08 ULONG ulTdpDerateDPM4;
E3 08 ULONG ulTdpDerateDPM5;
E3 08 ULONG ulTdpDerateDPM6;
E3 08 ULONG ulTdpDerateDPM7;
EFUSE_LINEAR_FUNC_PARAM sRoFuse; (584*950)/255) + 1350= 3525 /4 = 881.25
48 02 USHORT usEfuseIndex; 584
18 UCHAR ucEfuseBitLSB;
08 UCHAR ucEfuseLength;
B6 03 ULONG ulEfuseEncodeRange; max 2186
46 05 ULONG ulEfuseMin; 1350
BE 2C ULONG ulRoAlpha; 11454
7A 36 ULONG ulRoBeta; 13946
1E 8F ULONG ulRoGamma; 36638
AF 81 ULONG ulRoEpsilon; 33199
00 ULONG ulATermRo;
00 ULONG ulBTermRo;
37 ULONG ulCTermRo; 91
00 ULONG ulSclkMargin;
84 ULONG ulFmaxPercent;
00 ULONG ulCRPercent;
3c ULONG ulSFmaxPercent; 60
00 ULONG ulSCRPercent;
9b 02 ULONG ulSDCMargine; 667
I have tried to calculate what are named "Measured value" for the Efuse values, but I don't see any sense on it or I have done it wrong.
I'm using this values to change voltages:
The value 46 05 = 1350 is ulEfuseMin and is lowering in littel increases the voltages from DPM 1 to 6 in big increase is lowering DPM 1 to 7 voltages. And increasing the voltage if is lowered
EDITED: I MADE TWO MISTAKES ulCTermRo is 37 and is ulFmaxPercent not ulSFmaxPercent, sorry.
The value 37 = 55 is ulCTermRo and 84 =132 that is ulFmaxPercent ulSFmaxPercent are lowering DPM 1 to 7 voltages if are lowerd and incrasing them if are increased.
The values:
BE 2C = 11454 ulRoAlpha;
7A 30 = 13945 ulRoBeta;
1E 8F = 36638 ulRoGamma;
AF 81 =33199 ulRoEpsilon;
Are lowering DPM 1 to 7 voltages if are lowerd and incrasing them if are increased. But I have lower scores changing them.
EDITED:
B6 03 ULONG ulEfuseEncodeRange is like a voltage offset higher values are less volage for DPM 1 to 7, and lower values more moltage. + or -15 or 20 devimal values for the value is ~ + or - 6.25v, but I have to test more. I don't see performance lost.
Thanks for the person who gave me a rep+
Last edited by mynm; 10-27-2018 at 03:37 AM.
New to Overclock.net
Join Date: Jul 2015
Posts: 751
Rep: 36
(Unique: 29)
Good stuff as always mynm 
I just sold my 285 for $100 *tear*
It was only $200 new so $100 for 3 years of gaming ain't too shabby. I enjoyed this thread a lot. Thanks everyone.
I just sold my 285 for $100 *tear*
It was only $200 new so $100 for 3 years of gaming ain't too shabby. I enjoyed this thread a lot. Thanks everyone.
|
|
| Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
| Thread Tools | |
|
Show Printable Version Show Printable Version
Email this Page Email this Page
|
|
|
| Posting Rules | |