[Problem] Enabling AVX-512 on ASUS Prime Z790-P WiFi

Hi, I’m trying to enable AVX-512 on my system (ASUS Prime Z790-P WiFi + Intel Core i7-12700KF) and I’ve done quite a bit of research and analysis already. Here’s what I’ve found so far.

Current BIOS situation:

The stock BIOS does not contain microcode v15 for CPU ID 0672, I have already replaced it manually using MMTool with microcode v15 (date 2021/10/21, took it from here). The modified BIOS with the new microcode is ready. The AVX512 option exists in the IFR but is hidden by a hardcoded SuppressIf:

0x6C11F: SuppressIf
Uint64 Value: 0x0
Uint64 Value: 0x0
Equal ← always true, always hidden
0x6C33: OneOf Prompt: "AVX512", VarStoreId: 0x2D, VarOffset: 0xCB
Auto=0 (default), Disabled=1, Enabled=2

The same hardcoded (0==0) SuppressIf is also on AVX512 Ratio Offset (0x6C1B7) and AVX512 Voltage Guardband (0x6C263). The first BIOS release of this board had the same hardcoded suppression from day one, there is no older BIOS with the option exposed.

What I’ve tried:

Using UEFI Editor (boringboredom), I modified the access level of all 5 AVX512-related options in AMITSESetupData (09→05 and 29→05) at these offsets:

  • 0x12234 → QuestionId 0xA8 (AVX512 main option) — was 09
  • 0x1230C → QuestionId 0x2D72 (AVX512 Ratio Offset to per-core Ratio Limit) — was 29
  • 0x12354 → QuestionId 0x2D73 (AVX512 Ratio Offset numeric) — was 29
  • 0x1242C → QuestionId 0xA9 (AVX512 Voltage Guardband Scale Factor) — was 09
  • 0x12474 → QuestionId 0xAC (AVX512 Voltage Guardband Scale Factor numeric) — was 09

This only modifies SetupData.bin cleanly (single byte changes, no structural modifications to the Setup module).

Additional note on SiInitFsp:

I also attempted to locate the SiInitFsp module in the BIOS to apply the 24 01 → 24 00 patch described in the main AVX-512 thread (dsanke/RyanCawdor method). However, the module could not be found either by text search or by GUID in UEFITool NE. It seems ASUS Z790 may compress or package FSP modules differently. If anyone knows how to locate and extract SiInitFsp from an ASUS Z790 BIOS, that would also be very helpful.

Questions:

  1. Will changing the access level to 05 in AMITSESetupData be enough to bypass the hardcoded (0==0) SuppressIf and make the AVX512 menu visible? Or does the SuppressIf in the Setup module take priority regardless?

  2. Would changing 09→0D (user-set flag) instead of 09→05 (access level) be more effective in this case?

  3. Some sub-options (like AVX512 Ratio Offset to per-core Ratio Limit) had access level 29 instead of 09. I also changed these with UEFI Editor to 05 as well. Is 29 a different type of access level, and is changing it to 05 the correct approach for these sub-options, or should they be handled differently?

  4. Is there a better approach altogether, for example modifying the SuppressIf condition directly in the Setup module (changing 45 8A 00 → 45 8A 01 at offset 0x6C121)?

  5. Is the microcode v15 added to the BIOS sufficient to enable AVX-512 once the option is exposed? Or is a specific older microcode build required?

  6. This board has no BIOS Flashback and a single BIOS chip. EZ Flash will likely reject a modified CAP file. Is FPT the only viable software flash method, and what’s the correct procedure to unlock BIOS lock + FPRR before flashing?

  7. As an alternative to software flashing, would it be safer or more practical to flash the modified BIOS directly via a CH341A programmer? The board has a single BIOS chip and no recovery mechanism, so having a hardware fallback is appealing. Any advice on this approach for ASUS Z790?

I’ll share IFR.txt, original SetupData.bin, modified SetupData.bin, and the stock BIOS file in case they are needed.
BIOS FILES
ZIP Password: winraid

Tagging some users who have worked on similar issues: @dsanke @Sweet_Kitten @553349149 @rofikkernel , any input would be greatly appreciated!

Thanks in advance.


Edit by Fernando: Thread moved into the „BIOS Problems“ Category and title shortened

Yes it does.

Setting access level to user (05) or extended user is the recommended approach. I can send info on what these values mean later, if you like.

Disabling the locks require the nvram store to be unprotected, so you could use software like setup_var.efi to set the locks off. Otherwise, you need an unlocked bios to be able to switch the options from asus native UI.

I believe I found how to make built-in EZ-flash utility accept modified image.
However, the user I was doing tests with had quit after a successful bios downgrade. EZ-flash stopped throwing “Error: Selected file is not a proper BIOS!”, and they went all the way back to v303, but were too afraid to try any modified image.

That confirms I need to modify the Setup module directly as well. Could you point me in the right direction on how to correctly modify the SuppressIf condition?
I tried removing the SuppressIf block entirely via UEFI Editor but it caused a byte shift which I think can be problematic.

Sure, I would appreciate it.

I see. Infact I tested the NVRAM protection status using setup_var.efi.
Reading works fine (AVX512 offset 0xCB returns 0x00 as expected), but writing is blocked.

Could you share the method you found? I’d like to test it.
One note though: confirming that EZ Flash accepts a downgraded BIOS doesn’t necessarily prove the checksum bypass works, since ASUS already allows downgrades natively, unless I’m missing something here.

I don’t blame him, flashing a modified BIOS without a safety net is a leap of faith. Before doing anything myself, I plan to use the CH341A to read and dump the current BIOS chip first, so I have a guaranteed recovery image ready before touching anything.

On that note, wouldn’t it just make more sense to use the CH341A for the entire process?

See the following post

By default, bios access level is assigned as either User or Admin. Setting control access flags in setup data changes the level required to see the corresponding option.

Control Access Definitions
CONTROL_ACCESS_DEFAULT	 0 //Do not override VFR
CONTROL_ACCESS_EXT_USER	 1 //Admin - Editable; User - Editable;
CONTROL_ACCESS_USER		 2 //Admin - Editable; User - Visible but not editable;
CONTROL_ACCESS_ADMIN	 3 //Admin - Editable; User - Invisible;
Flags definition, values on the right are amounts of bits
struct _CONTROL_FLAGS  //32 bit structure containing the attributes for a specific control
{
  UINT32 ControlVisible : 1;//Whether or not a control is shown
  UINT32 ControlAccess : 2;//Access to control
  UINT32 ControlReset : 1;//Whether or not changing a value requires a system reset
  UINT32 ControlReadOnly : 1;//Whether or not the control is read only
  UINT32 ControlInteractive : 1;//Whether the Control is interactive or not
  UINT32 ControlEvaluateDefault : 1;//If set Default is obtained by expression evaluation
  UINT32 Reserved1 : 1;
  UINT32 ControlRefresh : 8;//Refresh timer (in 0.1s intervals)
  UINT32 RefreshID : 1;
  UINT32 ControlRWEvaluate : 1;//set 0 if read/write expression is not found otherwise set
  UINT32 ControlReconnect : 1;//Whether or not changing a value requires a system reconnect
  UINT32 BitFieldVarstore : 1;
  UINT32 Reserved2 : 12;
};

So 0x29 (0010_1001) means ControlVisible = TRUE, ControlAccess=DEFAULT, ControlReset=TRUE, ControlInteractive=TRUE.
0x9 is the same, but ControlInteractive=FALSE.
0x5 means ControlVisible = TRUE, ControlAccess=ACCESS_USER.

It required patching PEFirmwareUpdateDxe RAM ImageHandle in multiple places. And it’s not as simple to share the method as you might think, because patterns are different for your board.
After all, it’s not confirmed whether the checksum bypass works. Would you really like to test it?

You right, if you get the ch314a it wouldn’t require any more software fiddling to flashe the chip.

Thanks for the link. I’ll try to see if I can apply the same workaround.

Thanks. I’ll use 0x05 as stated before as you confirmed being the best approach.

Do you think it’s not even worth to try?

I already have the programmer. I need to check where bios chip is located on the board and what type of chip is used.
I might send some pics of the board in case I’m not able to find it or I’m not sure if it’s the correct one.

I mean I can’t explain how things work so that you could try the workaround on your own. The process is not as simple as “click here and there and bang you entered debug mode”. Since uefi images for the two boards are different, it’ll require me to come up with a new set of memory patches, and then guide you through the process.
I asked a question to confirm your intentions. If not strong enough, why would I even bother to do all the work only for it to turn out being needless.