[MOD] Sapphire RX 480 Nitro+ OC - Better Custom BIOS

Vento041 · 05-18-2017, 04:31 PM

LAST MAJOR FIX: 0.05.3 (480 ID version had the wrong ID -.-" I'm dumb - 580 ID version untouched)
LAST THREAD CHANGES: Added ZeroFan offset to the offsets list
SHOULD YOU UPDATE?

Use ANY ID version < 0.05.1? YES - MAJOR FIX RELEASED
Use 480 ID version < 0.05.3? YES - MAJOR FIX RELEASED
Use 580 ID version = 0.05.1? Not needed, but suggested (minor mem OC fix)
Use 580 ID version = 0.05.2? Not needed

FEATURES + CHANGE-LOGS:

Spoiler!

(From 0.01) Based on the last available OC bios (The one clocked at 1342 MHz at with stock +25 mV voltage offset) for Sapphire RX 480 Nitro+ OC 8GB.
(From 0.01) 80 °C target temperature (from stock 75 °C, delay temp throttling)
(From 0.01) TDP = Max Power Limit = 300 W and TDC 200 A (0.05 update improved this). Basically you obtain the same result of 50% PowerLimit without touching that slider
(0.05 update) UEFI Patch added (untested since I do/can not use UEFI)!
- (0.05.2 update) Updated to Version: 1.67.0.15.50 (Build date: Oct 18 2018.14:11:59)
(0.02 update) Raised the bios image size from 0xE600 to ~~0xEC00~~ 0xE800 (0.05 update).
(0.02 update) 1000 MHz memory step (like an RX 580 actually implemented not copy-paste from a RX 580 bios)
(0.04 update) RX 580 Fake ID - Skip driver bios check, basically the card is now recognized as RX 580 and not anymore as RX 480. This is done because the RX 5xx series bios is not checked by the driver, so you do not need atikmdag-patcher to make the driver load properly. (0.05 update) "Minimal patch", now only 1 byte is changed (OFFSET = 0xD6, value changed from 0xC6 to 0xE6).
- (0.05.2 update) Also RX 480 ID version of this bios is available. May help with mem OC instability with 18.11.1+ drivers. You will need atikmdag-patcher, since drivers checks for modded bios on 4xx series.
Some VRM related settings (the most important ones) are present inside the bios. You can find the best values with software like VRMTool and after that you can make them permanent by change the value directly inside the bios file (with hex editors like HxD).
- Voltage Offset Loop 1 (BIOS Stock)
- VRM Switching Frequency Loop 1 (From 0.01)
- VRM Switching Frequency Loop 2 (0.05 Update) It seems to help with memory stability.
- Loadline Calibration (From 0.01)
- Vmax Limit booth loops (0.03 update but broken, 0.05 fixed it)
- (0.05 update) Removed "Load Balancing" and "Output Current Scale" (ask if you want them back)
(0.05.1 update) Fixed a wrong offset (thanks @hellm to point that out), inside the PowerPlay subtables! This could have been really bad! Luckily if your card never hit 109 °C you should be fine.
(0.05.2 update) Fixed vrm switching frequency for Loop 2, Was 0xA0 (300 kHz, I do not know... why I wrote that ?????) when it should be 0x78 (400 kHz stock value). May help with mem OC instability with 18.11.1+ drivers.
(0.05.2 update) Changed/Updated the "Vento's" version of this bios (the settings that I use every day).
(0.05.2 update) The 1.3 Volt WattMan Version now has a Vmax limit set to 0x53. See the dedicated paragraphs to know more.
(0.05.3 fix) 480 ID version was using the 470 ID... I'm ultra dumb... sorry everyone for the mistake

INTERESTING OFFSETS LIST:

Spoiler!

Target temperature: OFFSET = 0x9F18 (Value 0x50 = 80 -> 80 °C)
Last memory step VDDCI voltage: OFFSET = 0x9E2D (Value: 0x0B = 11 -> index of 1000 mV)
- This is not the value that you see inside PBE! This is the "real" VDDCI
- The value is an index of a list of possible values. Here I'll wrote only the value that makes "sense". DO NOT TRY OTHER VALUES! TELL ME IF YOU NEED DIFFERENT ONES:
  - 0x08 -> 850 mV
  - 0x09 -> 900 mV
  - 0x0A -> 950 mV
  - 0x0B -> 1000 mV
  - 0x0C -> 1050 mV
  - 0x0D -> 1100 mV
  - 0x0E -> 1150 mV
ZeroFan: OFFSET = 0x9F28 (Default value: 0x01 -> ON)
- 0x00 -> OFF
- 0x01 -> ON
Voltage Offset Loop 1: OFFSET = 0xA9EC (Value: 0x04 = 4 -> 4 * 6.25 mV = 25 mV)
- Value * 6.25 mV = Actual offset
- Yes, you can use negative values but I forgot how this register want them represented (signed magnitude, One's complement or Two's complement)
VRM Switching Frequency
- Loop 1: OFFSET = 0xA9F0 (Value: 0x60 -> 500 kHz)
- Loop 2: OFFSET = 0xA9F4 (Value: 0x78 -> 400 kHz).
  - "Wasn't Loop 2 unused?..." It seems to help with memory stability.
- See this guide (done by @gupsterg) to properly know how to change those values.
- This graphics card has IR3553 power stages.
  - There is no real point to exceed 800 kHz (0x3C). Sweet-spot got from datasheet.
  - Can handle at max 1 MHz (0x30). Tons of heat!
Loadline Calibration: OFFSET = A9F8 (Default value: 0x81 -> OFF)
- 0x81 -> OFF
- 0x01 -> ON
Vmax Limit (both loops): OFFSET = 0xAA00 (Value: 0xFF -> unlimited)
- See this guide (done by @gupsterg) to properly know how to change this value
- Useful for spikes reductions!
- 1 byte for both loops.
  - High half-byte -> Loop 1
  - Low half-byte -> Loop 2
- I strongly suggest to set Vmax to the lowest value above you max voltage
  - For stock voltage (1150 mv + 25 mv offset = 1175 mV) use 0x44 to get ~1.250 Volt Limit.
  - The next "step" is 0x55 and gives you a ~1.365 Volt limit. Btw, if you do not plan to go aboive 1.250 0x44 is perfectly fine.
    - Anyway 0x54 would make more sense (VDDCI on Loop 2 sould never go too high)
  - "Why not 0x33?" Too low even for stock... ~1.140 Volt limit.
  - If you do not plan to rise VDDCI above 1100 mV (the next available step is 1150 mV) you write 3 in the low half byte.
    - Example: I use 0x43, ~1.25 Volt loop 1 limit and ~1.14 Volt loop2 limit.

BUG??? WattMan settings are weird / Overclock fail to apply / Mem Overclock fails / How to properly change OC inside WattMan

Spoiler!

F.A.Q.:

Spoiler!

HOW TO UPDATE (paranoid way):

Spoiler!

"WHICH VALUES/MODS/SETTINGS DO YOU USE VENTO"?:

Spoiler!

Right now I'm more interested on finding a good mem OC (while keeping 0 memory errors), so my current settings is:

Stock switching frequency both loops
- In future I will try 800 kHz on loop 2 to test/improve mem OC
Loadline Calibration disabled (stock)
- Will use it later to test more aggressive core OC
2100 MHz Mem OC
Custom strap on the 2000 memory step
- based on 1750 and changed one value at time (with R_Timings, a program by me)
Vmax set to 0x43. Nice spikes reductions.
- ~1.25 Volt limit loop 1
- ~1.14 Volt limit loop 2.
VDDCI voltage set to 1100 mV
0.5% Core OC on all states -> Max OC is 1350 MHz.
- No fixed voltage values trick so base voltages raise a bit (+6.25 mV, not everytime).
0 voltage offset -> -25 mV from stock
- GPU is not "starving" but it's the bare minimum to sustain my small core OC.
- Lower temp than stock
Custom "Silent" fan curve (set by TRIXX, I'm working to get this properly inside BIOS)
ULPS Disabled (Regedit tweak)
PCIe Link State Power Management Disabled (Windows power settings)
480 ID + atikmdag-patcher. Mem OC seems easier and more stable if the card reports the real ID
0% power limit. Since it's already have a +50% applied to base limit I set the power limit slider to reach max 0% so I do not tuch it by error.

Remember that my card use hybrid cooling (water and air) and your card may not be able to sustain the core OC (even if really small, +0.5%) at ~-25 mV from stock settings.

In the zip attached to this thread there is a "Vento's" version, with the settings previously described, but there are some difference:

Custom timings strap on the 2000 step. You'll find the base 1750 strap since the custom one may not work on every card.
UEFI disabled, I do not use it so I remove it. All the bios I post have the UEFI patch available and enabled.
I use the 480 ID version (card is recognized properly as 480, as stock) but there is also a 580 ID version (mem oc seems to suffer from it). Of course I have to use the atikmdag-patcher or the drivers do not properly work.

1.300 Volt WATTMAN VERSION:

Spoiler!

TODO:

Spoiler!

DO YOU HAVE ANOTHER Sapphire RX 480 MODEL? (4 GB or not-"OC" or not-"NITRO+")

Spoiler!

Special Thanks to...

@Jestercore we were trying RX 580/480 bios and exchanging results for months
@generaleramon for general VRM & memory mod help, and because he's Italian like me
@deeper-blue for VRMTool and general VRM help
@BromoL for WattTool
@gupsterg I lurked for years thanks to yours excellent guides
buildzoid for PCB breakdowns videos and knowledge about VRM and memory timings
@mynm "ASIC_ProfilingInfo" table tweaks. Team member.
@hellm UEFI Patch :D. Team member.

Mhazz · 05-19-2017, 10:55 PM

That post on Reddit is 5 month old...

It is THAT BIOS, isn't it?

Vento041 · 05-20-2017, 12:51 AM

Quote:

Originally Posted by Mhazz

That post on Reddit is 5 month old...

It is THAT BIOS, isn't it?

It's based on those (you can find them in the internet since there is the sha)

Those two bios are just an updated version of the silent and oc bios for a nitro+ oc, they act like usual. My mod is based on these two bios and all the features implemented were done manually by me xD (except for the 1.175 vmax, that's a peculiarity of the silent bios).

Do you have a nitro+ oc?

Mhazz · 05-20-2017, 01:16 AM

Quote:

Originally Posted by Vento041

It's based on those (you can find them in the internet since there is the sha)

Those two bios are just an updated version of the silent and oc bios for a nitro+ oc, they act like usual. My mod is based on these two bios and all the features implemented were done manually by me xD (except for the 1.175 vmax, that's a peculiarity of the silent bios).

Do you have a nitro+ oc?

Yes I have, 8Gb samsung memory, that don't OC very well by the way...

In afterburner I have -6mv core voltage +0PL% 1360Mhz clock and 2050memory. That the best I can get considering heat, noise, memory error (i like to have 0 all the time).
Oh, and I have set Core voltage to 1135 in WattTool. So it peaks at 1.156V (GPU-z and HWinfo sensor reading) Should'nt it be 1.154 (1135 + 25 - 6)?

All that with ubermix timing 3.1 from generaleramon. That's all I have modded in the bios.

Have test some RX580 mod, but gave up as now I just want to enjoy and play games lol.
So I'm stick with RX480bios and some minor tweeks.
Or maybe try BIOS from someone with same card that shares his work with others, like you do

Vento041 · 05-20-2017, 02:10 AM

Quote:

Originally Posted by Mhazz

In afterburner I have -6mv core voltage +0PL% 1360Mhz clock and 2050memory. That the best I can get considering heat, noise, memory error (i like to have 0 all the time).
Oh, and I have set Core voltage to 1135 in WattTool. So it peaks at 1.156V (GPU-z and HWinfo sensor reading) Should'nt it be 1.154 (1135 + 25 - 6)?

Or maybe try BIOS from someone with same card that shares his work with others, like you do

[/quote]

Are you using my bios?
Does it works well? The 2mV spike could be just loadline effect, if you use my bios remember that the native offset is 0 so if you set 1000 mV core voltage you'll get 1000 mV.

Mhazz · 05-20-2017, 02:21 AM

Quote:

Originally Posted by Vento041

Are you using my bios?
Does it works well? The 2mV spike could be just loadline effect, if you use my bios remember that the native offset is 0 so if you set 1000 mV core voltage you'll get 1000 mV.

No it's the original BIOS (not MY original, but the one I linked in my previous post. I will give a try to yours this evening and tell you what happens.

Thanks for help and rep anyway

hellm · 05-20-2017, 08:36 AM

how did you implemented the 1000MHz mem step? Does this also include any GPU stuff? The complete 580 PowerPlayTable?

Vento041 · 05-20-2017, 10:53 AM

Quote:

Originally Posted by hellm

how did you implemented the 1000MHz mem step? Does this also include any GPU stuff? The complete 580 PowerPlayTable?

Nope, I studied how the PowerPlayTable is structured in both an RX 4xx bios and a RX 5xx bios, but evrything was don by patching the original powerplaytable

I used as reference readon linux drivers (atombios.h, etc, ... ) and OhGodATool source (there is an error in the offsets, "uint32_t MaxODMemClk;" start from 0x1B not 0x1A, and all the other offsets after that! So the sub-table starts from 0x23 not 0x22!).

Since I inserted a frequency step between 300 and 2000, the "index" of the 2000 step has changed so I had to replace most of the reference to this step (I remember that in IT we start to count from 0!!!! so 300 have index 0 and 2000 had index 1).

I modded these sub-tables:

PolarisMemClkDepTable, added new record, updated records count and updated lenght, each record is D bytes long (D hex = 14 dec)
PolarisVCEStateTable, indexes of some records
PolarisState, "MemClockIdxHigh" from 1 (the old index of 2000) to 2 (the new index of 2000)
PPTableHeader, updated lenght and changed sub-tables offsets since PolarisMemClkDepTable is bigger

Fun fact: if you forget to update the indexes your memory will (logically) cap at 1000 MHz.

hellm · 05-20-2017, 07:14 PM

thx for the info

i need just a little more.. VRM switching frequency:

Quote:

Change VRM Frequency: search these Hex Values "00 62 00 7E 00 69 00 20 00 8D 00 00 00 22 00 60" and change the "60" at the end.

i looked over the 480 nitro+ bios (that bios^^), and the voltage object info table:

..and it stops right after byte 2B. No fSW, no register 22. Where did u find 500kHz switching frequency? Not set is default?

Vento041 · 05-21-2017, 03:59 AM

Quote:

Originally Posted by hellm

thx for the info

i need just a little more.. VRM switching frequency:
i looked over the 480 nitro+ bios (that bios^^), and the voltage object info table:

..and it stops right after byte 2B. No fSW, no register 22. Where did u find 500kHz switching frequency? Not set is default?

The silent bios has nor voltage offset or fSW, the oc bios has only the voltage offset, since I base my work on the silent bios (higher Vmax) i added them both manually. The default values comes from what VRMTool report (or directly from any kind of way to dump I2C and reading registers directly) with stock OC bios installed.

I'm not completely sure about the origin of the default values, since different bios, all without the 22 register explicitly set, can have different default value. The reference, as example, should use ~300 MHz if I'm not wrong. Probably all these different values are "auto-set" by IR3567B, calculated from other register values and/or from live on-boot PCB measurements (I do not really know I'm just assuming how it could have been done).

[MOD] Sapphire RX 480 Nitro+ OC - Better Custom BIOS

Register Now

Log-in

Premium Vendor Showcase

Recent Discussions

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
You may post new threads You may post replies You may not post attachments You may not edit your posts BB code is On Smilies are On [IMG] code is On HTML code is Off Trackbacks are Off Pingbacks are Off Refbacks are Off Forum Rules

vBulletin Message

Cancel Changes