• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA RTX 2070 Modded to Support 16GB Memory

btarunr

btarunr

Editor & Senior Moderator
Staff member
Joined
Oct 9, 2007
Messages
47,670 (7.43/day)
Location
Dublin, Ireland
System Specs
System Name RBMK-1000
Processor AMD Ryzen 7 5700G
Motherboard Gigabyte B550 AORUS Elite V2
Cooling DeepCool Gammax L240 V2
Memory 2x 16GB DDR4-3200
Video Card(s) Galax RTX 4070 Ti EX
Storage Samsung 990 1TB
Display(s) BenQ 1440p 60 Hz 27-inch
Case Corsair Carbide 100R
Audio Device(s) ASUS SupremeFX S1220A
Power Supply Cooler Master MWE Gold 650W
Mouse ASUS ROG Strix Impact
Keyboard Gamdias Hermes E2
Software Windows 11 Pro
PC enthusiast VIK-on pulled off a sophisticated memory mod for the GeForce RTX 2070, doubling its memory amount to 16 GB. In a detailed video presentation (linked below), VIK-on demonstrated how he carefully removed the 8 Gb Micron-made GDDR6 memory chips of his card, with 16 Gb Samsung-made chips he bought off AliExpress for $200. Memory replacement mods are extremely difficult to pull off, as you first de-solder the existing chips using a hot air gun while keeping the contacts on the PCB intact (ensuring no pins short); and solder the replacement BGA memory chips in place.

In addition, a set of "jumpers" on the PCB need to be modified to make it recognize the Samsung memory. The resulting card booted to desktop successfully, with GPU-Z reading its full 16 GB memory size. The card successfully made it through 3DMark TimeSpy, albeit with 30% lower performance than a normal RTX 2070 (6176 points vs. ~9107 points). The card would also crash Furmark. Still, it's mighty impressive that the "TU106" recognizes 16 GB of addressable memory (which means all its memory channels are intact), without the need for any BIOS mods, which is impossible to pull off.



Watch the VIK-on video presentation here.

View at TechPowerUp Main Site
 
Even if not 100% successful, that's impressive as hell!
 
Now that's interesting. He configured the memory controller register by "jumpers" where it's normally done in the firmware. ..My question is, how did he gain that information that those jumpers control the register.

One way is to look at other 16GB cards & compare all resisters. Doing such a mod, I would have looked at the Titan cards & gone the full 24GB+, if the register supports anything higher than 24GB.

The other way to know that those jumpers control the register is to have full documentation of the GFX Chip itself, so i'm guessing he got that information from one of Nvidia support manufactures that make their cards.

It's remarkable he done this with a hot air gun, I use full infrared with vision base display so I can see the solder balls & PCB trace layout at the same time for perfect alignment. (Aoyue BGA9000A). The risk of just one poor solder contact will make the card not function as it should.


EDIT: You can't do this modding with just a air gun, you will overheat the memory chip & damage it internally. This kind of modding you also need a preheater, "mandatory".
 
Last edited:
Oooo, it's things like this that make my inner nerd go "Oh yeah!" It must have been a fun experiment.
 
come onvidia we need more vram
hope nvidia can recognize that
 
I can see on the pic that some of the PCIe series caps got knocked out. The protocol is resilient to these issues, but probably fell back to running in PCIe x1 or x2. Certainly not getting full PCIe bandwidth.
 
delshay said:
Now that's interesting. He configured the memory controller register by "jumpers" where it's normally done in the firmware. ..My question is, how did he gain that information that those jumpers control the register.
The other way to know that those jumpers control the register is to have full documentation of the GFX Chip itself, so i'm guessing he got that information from one of Nvidia support manufactures that make their cards.
Click to expand...
That is on the 3rd screenshot (maybe was added later). He has documentation :)
 
blobster21 said:
Explanation as to why the card ended underperforming after this surgery ?
Click to expand...

You know there's old saying that if you wanna go faster it is best to tune your engine not make your tank bigger.
 
I think the reason the card loses performance and crashes is probably just the memory straps being wrong for the memory chips.
 
1d10t said:
You know there's old saying that if you wanna go faster it is best to tune your engine not make your tank bigger.
Click to expand...
Adding extra lanes to the highway doesn't make traffic faster.
 
ityrant said:
come onvidia we need more vram
hope nvidia can recognize that
Click to expand...

Don't worry, they have truckloads of 2060's with 6GB ready to rock! Just get two for 12GB, the more you buy...
 
This forum soon will be flooded with..., please help to install more ram on my gpu, hahaha. Can't wait for these.
 
Sir Alex Ice said:
Proof that 16Gb is possible, just nvidia are being dicks.
Click to expand...
Why this obsession with VRAM lately? Until three weeks ago I had a 11GB 1080ti, now I have the 10GB 3080. I almost never saw allocation above 7GB (and remember, allocation is not actual usage), with the worst game being Horizon: Zero Dawn, but I'm fairly certain it's just a bug or a weird design choice. The game allocates 8GB of VRAM and 14GB of RAM at launch, even before any assets are loaded. Even cyberpunk 2077 at 3840x2160 (no DLSS) with everything maxed, barely touches 8GB - and the game is unplayable at these settings anyway. With the prices of GDDR as high as they currently are, I wouldn't want to pay for memory I'm not going to use anyway.
 
You must log in or register to reply here.
Back
Top