• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

EVGA XC GTX 1660 Ti 8GB ROM

bielfrbfr

New Member
Joined
Jul 4, 2025
Messages
6 (3.00/day)
I have made a quick edit and believe it should work. Perhaps someone with more expertise can respond here.

So far, I’ve identified in the BIOS:
  • section1_tables starts at offset 0x28600
  • MemInfoTblPtr = 0x4B23 → points to offset 0x2D123, containing mem_variants_by_strap data with memory type and config
  • MemPartInfoTblPtr = 0x5010 → points to offset 0x2D610, which seems to be the memory partition table where the active VRAM chip count (6 or 8 modules) is likely defined
I’ve extracted the first 128 bytes from 0x2D610 to analyze this table, but I need help interpreting and correctly modifying the value that indicates the number of active VRAM chips so the BIOS can recognize all 8GB.
Has anyone worked with this table or knows how to identify and change the “chip count” or active module mask in this section?
 

Attachments

The 1660 Ti is specified to have 6GB of VRAM.

Have you modified the VRAM on the card?

You cannot make a 6GB card a 8GB with a vBIOS modification.
 
The 1660 Ti is specified to have 6GB of VRAM.

Have you modified the VRAM on the card?

You cannot make a 6GB card a 8GB with a vBIOS modification.
Yes, that’s common sense. The card has been upgraded to 8 GB using two 1 GB GDDR6 BGA modules. It originally had two unused spots.

I also own a CH341A programmer in case something goes wrong, but I would prefer to make it work with NVFlash on the first attempt.
 
I hate to say it but it's not gonna work. Why? Because TU116 physically is a 192-bit GPU die. Meaning it can only "see" six VRAM chips. You can swap modules for 2 GB ones and maybe even make it work as a 12 GB GPU (probably with some caveats, this Brazilian video for the reference) but don't quote me on that. It might end up impossible.
 
This is a fun little project let us know if it works
 
I hate to say it but it's not gonna work. Why? Because TU116 physically is a 192-bit GPU die. Meaning it can only "see" six VRAM chips. You can swap modules for 2 GB ones and maybe even make it work as a 12 GB GPU (probably with some caveats, this Brazilian video for the reference) but don't quote me on that. It might end up impossible.
^Adding to this : Places for extra VRAM were there, because this PCB is reused on some of EVGA's RTX 2060 lineup.
Note : TU106 and TU116 are pin compatible.
 
Last edited:
It might be defined through straps, not 100% sure
0x08 at Byte Index 3 is very likely the VRAM module count. (Which I already edited to 0x08, before was 0x06)
unsigned char ucDataBlock[16] = {
// Offset 0x2D610 to 0x2D61F
0x10, 0x04, 0x02, 0x08, // ← Byte 3 = 0x08
0x00, 0x01, 0x00, 0x07,
0x00, 0x07, 0x00, 0x07,
0x00, 0x07, 0x00, 0x07
};

Byte IndexValueMeaning (assumed based on common NVIDIA BIOS structure)
00x10Possibly structure version or size identifier
10x04Possibly revision or a sub-identifier
20x02Could indicate memory vendor or type
30x08Likely VRAM chip count (modules) → this is the one you care about
4–15...Configuration values for each memory channel or controller

section1_tables and mem_variants_by_strap
1751736576258.png

1751736621399.png
 
^Adding to this : Places for extra VRAM were there, because this PCB is reused on some of EVGA's RTX 2060 lineup.
Note : TU106 and TU116 are pin compatible.
However, it's worth noting that the EVGA RTX 2060 Ti (if referring to a non-standard model) still comes with 6GB of VRAM. Only the RTX 2060 Super variant features 8GB.
 
I believe I have identified the issue. The card contains 8 physical memory pads, which correspond to 8 straps in the BIOS. The first 6 straps are configured to utilize a total of 6 GB, allocating 1 GB per strap. The remaining 2 pads, numbers 7 and 8, currently have no VRAM modules installed.

By changing the density setting on the first 6 straps from 0x5 to 0x4, the total recognized memory now reflects 4 GB. This raises a question: since straps 7 and 8 are pre-configured with a value of 0x6, which corresponds to 16 GB, would it be sufficient to solder the two additional VRAM modules for the system to recognize the full capacity?

1751742650243.png
 
Byte IndexValueMeaning (assumed based on common NVIDIA BIOS structure)
00x10Possibly structure version or size identifier
10x04Possibly revision or a sub-identifier
20x02Could indicate memory vendor or type
30x08Likely VRAM chip count (modules) → this is the one you care about
4–15...Configuration values for each memory channel or controller
Memory Partition Information Table
Roughly
0 version
1 header size
2 bytes per entry
3 number of entries

My GTX 1080 which has 8GB has 6 entries ;)
00 01
00 07
00 07
00 07
00 07
00 07

I believe I have identified the issue. The card contains 8 physical memory pads, which correspond to 8 straps in the BIOS.
The straps are for different memory make and density. ie from your table if I were using Hynix 4Gb chips then I would be using strap 2.


There was some talk of TU116 shortage and TU106 being used for 1660's but never seen one in the wild although did run across a schematic for a 1660 which identified a TU106 but still 6GB. Falcon has VBIOS pretty much locked down. Originally nvidia were selling hulk certs to bypass that but due to whining from people who probably wouldn't use them anyway they stopped and seem to have disappeared in the latest cards.

The thing is though even if you don't get your end result you'll probably learn lots attempting it. :)
 
Memory Partition Information Table
Roughly
0 version
1 header size
2 bytes per entry
3 number of entries

My GTX 1080 which has 8GB has 6 entries ;)
00 01
00 07
00 07
00 07
00 07
00 07


The straps are for different memory make and density. ie from your table if I were using Hynix 4Gb chips then I would be using strap 2.


There was some talk of TU116 shortage and TU106 being used for 1660's but never seen one in the wild although did run across a schematic for a 1660 which identified a TU106 but still 6GB. Falcon has VBIOS pretty much locked down. Originally nvidia were selling hulk certs to bypass that but due to whining from people who probably wouldn't use them anyway they stopped and seem to have disappeared in the latest cards.

The thing is though even if you don't get your end result you'll probably learn lots attempting it. :)
So, changing the density value wouldn’t have any effect? I’ve compared it with other ROMs from graphics cards with more GB, and what I’ve been thinking seems to match.

I would love that. The GTX 1660 Ti is an excellent graphics card, and even more so coming from EVGA. Graphics cards of this quality are no longer being produced. I can only hope that Intel Arc partners with EVGA so they can return to the market.
 
I can only hope that Intel Arc partners with EVGA so they can return to the market.
The owner of EVGA would either need live forever, or would need to change his stance on 'never selling or passing the company on to anyone else'.
He said that he is retiring, will not pass the company on, and is letting the company wind down.
 
So, changing the density value wouldn’t have any effect?
I've not seen anyone changing the VBIOS. Normally (used loosely here) some skilled people have changed the chips to higher density to increase VRAM. For instance 6x 8Gb chips would give 6GiB of VRAM, changing the chips to 16Gb would double that to 12GiB. Again 'normally' the straps would be changed by arranging configuration strap resistors on the board between high and low rails to give high, mid and low configuration values.
 
Back
Top