Friday, October 30th 2026

New UEFI Tool Lifts NVIDIA's Tensor Cap on RTX 20 to 40 Series Cards in Windows

Last week, modders found a way to double the tensor throughput of GeForce RTX 20, RTX 30, and RTX 40 cards, though Zagitta's original tool only ran on Linux. Now, Chinese modder physicsdolphin has released nv-uefi-unlocker, a UEFI tool that lifts the same limit on Windows. NVIDIA runs FP16, BF16, and TF32 tensor math with FP32 accumulation at half rate on GeForce cards, while the same silicon in professional cards runs at full speed. The tool removes that cap before the driver even loads, without any VBIOS flashing. An earlier Windows port from impacifist targeted a single RTX 3090 system, while this one covers all three generations.

The gains look substantial. One user's RTX 4070 Ti Super went from 57.00 to 150.93 TFLOPS in FP16 (2.65x), though board power also climbed from 227 W to nearly 292 W. An RTX 2070 reportedly jumped from 36 to around 60-63 TFLOPS. Not every card has been tested; RTX 20 owners have to enable GSP firmware mode first, and RTX 50 isn't supported at all. The unlock resets on every power-off, so it runs from a USB stick on each cold boot, with Secure Boot disabled or a self-signed certificate enrolled. That could be a problem for games whose anti-cheat requires Secure Boot.
Gamers shouldn't expect anything here, since games and regular FP32 CUDA work see no change. This is mostly for local AI work like image generation and fine-tuning. The developer also warns that pairing it with a power limit unlock can push power and heat well past stock limits, so readers are advised to keep an eye on temperatures.
Sources: nv-uefi-unlocker (GitHub), Reddit, via Notebookcheck
Add your own comment

28 Comments on New UEFI Tool Lifts NVIDIA's Tensor Cap on RTX 20 to 40 Series Cards in Windows

#1
jsfitz54
Great business plan! Blowup your old cards by bypassing the power limits and force you to buy new ones at inflated cartel prices.
Posted on Reply
#2
chrcoluk
Does this boost DLSS, or was that never affected by this gimping?
Posted on Reply
#3
Lew Zealand
chrcolukDoes this boost DLSS, or was that never affected by this gimping?
This is my first question. I'll tentatively assume 'no' as DLSS doesn't seem to work in a lower-power or lower-effectiveness state compared to Ampere or Ada.
Posted on Reply
#4
ymdhis
jsfitz54Great business plan! Blowup your old cards by bypassing the power limits and force you to buy new ones at inflated cartel prices.
As opposed to burning down your cards by using the factory 12VHPWR connector?

It actually IS a good business plan mind you: more dead cards means users have to buy more new cards, which is more profit for Nvidia. Which explains why the 12VHPWR is so garbage.
Posted on Reply
#5
N/A
The connector is not the end of the card itself. If anything you can always ghetto solder a bypass modular cable directly to the pins already sticking out on the back of the PCB. And under volt works just fine, 20 and 40 series work well with 66% power limit. And also have a very strict power limit imposed out of the box and are technically overbuilt. So unless you don't get repasted on time like mine is already 8 years old and i can hear it crack under load, there is nothing bad that could happen that wasn't already on the cards.
Posted on Reply
#7
bug
Vya DomusGimpvidia.
How quick are we to forget this limitations were put in place to make the cards less appealing for miners...
Posted on Reply
#8
ymdhis
N/AThe connector is not the end of the card itself.
That's exactly what it is if it ends up melting, lol.
Posted on Reply
#9
R-T-B
bugHow quick are we to forget this limitations were put in place to make the cards less appealing for miners...
Thats actually completely separate from the tensor gimping.
Posted on Reply
#10
bug
R-T-BThats actually completely separate from the tensor gimping.
It's still a cost control measure.
I mean, what's the alternative? Create new silicon for each and every SKU? That could easily double the prices, if they need to manage more variants while while making fewer for each.
Posted on Reply
#11
R-T-B
bugIt's still a cost control measure.
I mean, what's the alternative? Create new silicon for each and every SKU? That could easily double the prices, if they need to manage more variants while while making fewer for each.
Of course, I was only pointing out its not the same gimp.
Posted on Reply
#12
kondamin
jsfitz54Great business plan! Blowup your old cards by bypassing the power limits and force you to buy new ones at inflated cartel prices.
cartel of one
Posted on Reply
#13
evilpaul
That the cards haven't upped the VRAM in like 15 years, outside a handful of halo products, is Nvidia nerfing them to bleed more money out of people.
Posted on Reply
#14
demirael
jsfitz54Great business plan! Blowup your old cards by bypassing the power limits and force you to buy new ones at inflated cartel prices.
Where do you see a power limit bypass?
Posted on Reply
#16
ZeDestructor
chrcolukDoes this boost DLSS, or was that never affected by this gimping?
The performance needs of DLSS and other upscalers are so far below the artificial limits that it never mattered for that use.
Posted on Reply
#17
chrcoluk
ZeDestructorThe performance needs of DLSS and other upscalers are so far below the artificial limits that it never mattered for that use.
Thats interesting given Nvidia state various features cant be used because of tensor core limits, and that DLSS performance overhead is smaller on newer cards indicating tensor resource saturation.
Posted on Reply
#18
Onasi
evilpaulThat the cards haven't upped the VRAM in like 15 years, outside a handful of halo products, is Nvidia nerfing them to bleed more money out of people.
Huh? Mainstream cards today have more memory than flaships of 15 tears ago, often more than double, in fact. I can see saying 5-7 years, but 15 is pushing it.
chrcolukThats interesting given Nvidia state various features cant be used because of tensor core limits, and that DLSS performance overhead is smaller on newer cards indicating tensor resource saturation.
NVidia say a lot of things. Theoretically, they might be right and new Tensor cores are just more performant by themselves. In practice - who knows, it’s NV. Realistically, yes, newer models are slightly heavier in terms of performance cost, it’s been tested. It’s not clear if this “unlock” would even apply here, considering it’s very much hacked together.
Posted on Reply
#19
ZeDestructor
chrcolukThats interesting given Nvidia state various features cant be used because of tensor core limits, and that DLSS performance overhead is smaller on newer cards indicating tensor resource saturation.
For sure, but that's more about having the newer datatypes, not a lack of performance.

That does indeed mean a higher performance penalty, but still well below the total tensor performance - remember, you still gotta do the basic real rendering before upscaling and that will still eat most of the computational power of the GPU.
Posted on Reply
#20
bug
chrcolukThats interesting given Nvidia state various features cant be used because of tensor core limits, and that DLSS performance overhead is smaller on newer cards indicating tensor resource saturation.
You're thinking RT. The question was about DLSS.
Posted on Reply
#21
chrcoluk
bugYou're thinking RT. The question was about DLSS.
No DLSS, there has been many articles which benchmark the loss of frames when switching to transformer e.g. Its a bigger hit on older cards, if it wasnt stressing the hardware, then there would be no extra hit.
Posted on Reply
#22
evilpaul
OnasiHuh? Mainstream cards today have more memory than flaships of 15 tears ago, often more than double, in fact. I can see saying 5-7 years, but 15 is pushing it.


NVidia say a lot of things. Theoretically, they might be right and new Tensor cores are just more performant by themselves. In practice - who knows, it’s NV. Realistically, yes, newer models are slightly heavier in terms of performance cost, it’s been tested. It’s not clear if this “unlock” would even apply here, considering it’s very much hacked together.
The 1080 and 1080Ti are ten years old. Anyone still using a quadcore with 6MB L3?
Posted on Reply
#23
rethcirE
So who is ACTUALLY doing this?

I took my 4070 from 60-65 TFLOPS to 99-101 TFLOPS just by booting from the USB first. Simple!

Posted on Reply
#24
bug
chrcolukNo DLSS, there has been many articles which benchmark the loss of frames when switching to transformer e.g. Its a bigger hit on older cards, if it wasnt stressing the hardware, then there would be no extra hit.
I wasn't saying DLSS won't stress the hardware, just that it doesn't run on tensor cores.
Posted on Reply
#25
igormp
chrcolukDoes this boost DLSS, or was that never affected by this gimping?
No, it was never affected. The kernels for it use FP16 accumulation, which was not gimped and already ran at full speed.
ZeDestructorThe performance needs of DLSS and other upscalers are so far below the artificial limits that it never mattered for that use.
With the exception being DLSS 5. That one does need quite a bit of grunt to run properly.
bugI wasn't saying DLSS won't stress the hardware, just that it doesn't run on tensor cores.
What do you mean? DLSS does run on tensor cores.
Posted on Reply
Add your own comment
Nov 19th, 2026 13:23 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts