Friday, January 17th 2025

NVIDIA Reveals Secret Weapon Behind DLSS Evolution: Dedicated Supercomputer Running for Six Years

At the RTX "Blackwell" Editor's Day during CES 2025, NVIDIA pulled back the curtain on one of its most powerful tools: a dedicated supercomputer that has been continuously improving DLSS (Deep Learning Super Sampling) for the past six years. Brian Catanzaro, NVIDIA's VP of applied deep learning research, disclosed that thousands of the company's latest GPUs have been working round-the-clock, analyzing and perfecting the technology that has revolutionized gaming graphics. "We have a big supercomputer at NVIDIA that is running 24/7, 365 days a year improving DLSS," Catanzaro explained during his presentation on DLSS 4. The supercomputer's primary task involves analyzing failures in DLSS performance, such as ghosting, flickering, or blurriness across hundreds of games. When issues are identified, the system augments its training data sets with new examples of optimal graphics and challenging scenarios that DLSS needs to address.

DLSS 4 is the first move from convolutional neural networks to a transformer model that runs locally on client PCs. The continuous learning process has been crucial in refining the technology, with the dedicated supercomputer serving as the backbone of this evolution. The scale of resources allocated to DLSS development is massive, as the entire pipeline for a self-improving DLSS model must consist of not only thousands but tens of thousands of GPUs. Of course, a company making 100,000 GPU data centers (xAI's Colossus) must save some for itself and is proactively using it to improve its software stack. NVIDIA's CEO Jensen Huang famously said that DLSS can predict the future. Of course, these statements are to be tested when the Blackwell series launches. However, the approach of using massive data centers to improve DLSS is quite interesting, and with each new GPU generation NVIDIA release, the process is getting significantly sped up.
Source: via PC Gamer
Add your own comment

62 Comments on NVIDIA Reveals Secret Weapon Behind DLSS Evolution: Dedicated Supercomputer Running for Six Years

#51
Dr. Dro
Prima.VeraI don't know man. The DLAA still looks like a shitty anti-aliasing solution. Still a complete garbage blur mess, just a little better than the worst AA ever invented, TAA.
You want a good AA technique? Just have a look on how older games were looking with 4xSSAA or even 8xSSAA.
Those were the best times.
How? It's DLSS applied to the native resolution, ergo only scene reconstruction algorithm is running. It can't be blurry unless the frame rate is faltering, the perception of blurriness comes from poor resolution or frame rate. DLAA tends to enhance scene detail in some engines that only enable certain aspects of the image when the resolution is high enough.

SSAA was simply increasing the internal rendering resolution (NV DSR and AMD VSR were a way to universally apply this without support in the game), MSAA has always been the clearest in my opinion but it's very resource-inefficient and has many compatibility issues with modern rendering techniques (iirc it doesn't play well with deferred rendering).
Posted on Reply
#52
tpuuser256
The whole point of AI is to surpass the limited human intelligence as well as workforce automation. If all goes well AI will be better than humans at pretty much anything including creating stories or generating worlds (the model would use algorithm for more realisitc and efficient compute)

With nuclear fusion, energy could become much cheaper in the near future and allow exponential growth of civilization (technology and number of people)
Posted on Reply
#53
igormp
Dr. DroPSSR, as everything Sony, is fully proprietary, poorly documented to the public and apparently has been relatively poorly received so far. I don't believe it has any particular need for ML hardware since the PS5 Pro's graphics are still based on RDNA 2, which does not have this capability. Unless there is a semicustom solution, but I don't believe this to be the case.
It's a ML-based upscaling as well, but it doesn't make use of any extra specific hardware. RDNA3.5 (which the PS5 Pro kinda uses) has some extra instructions meant to process some stuff relevant for matmul in lower precision, you can read more about it here:
chipsandcheese.com/p/amd-rdna-3-5s-llvm-changes

With the extra hardware bump, it should be able to run an upscaling CNN without much issues and no need for extra hardware (apart from what's in the GPU itself).
Rightness_1Got you... But what I'm wondering is how can AMD/Sony (allegedly) do this in FSR4 without some "supercomputer" doing the work for them to upscale the image with minimal artifacts?
Sony's PSSR did do something similar to what Nvidia has done with DLSS, by training a model with tons of compute during a long period of time, which can then be used by the actual consoles, they just did not announce it like Nvidia did now. And if Nvidia never gave up this detail away, you wouldn't be making this complaint.
FSR4, if truly based on ML, will also require lots of compute time beforehand in order to create a model that can perform this task in your local GPU.

Let me try to give you a better example: do you know that feature in phone's gallery that are able to recognize people or places?
That's a machine learning model that's running in your phone and tagging those images behind the scenes.

That "model" (think of a "model" as a binary or dll that contains the "runtime" of the AI stuff) has been trained by google/samsung/apple in their servers for long hours with tons of examples saying "this picture is a dog", "this is a car", "this is a beach", "this person X is different from person Y", etc etc. This part is the "training" part, which is really compute intensive and takes really long time. As an example, the GPT model behind ChatGPT took around 5~6 months to train.
The outcome of this model is then shipped into your phone, where it's able to use what it has learnt and apply it to your cat pictures, and say that it is a cat. This part is called the "inference" part, and is often really fast. Think how DLSS, even in its first version, was able to upscale a frame from a smaller res into a higher one with really fast FPS (so for each frame, it upscaled it in less than 10ms!). In a similar manner, think how your phone is able to tag a pic as a "dog" really quick, or how ChatGPT is able to give you answers reasonably fast, even though the training part for all of those tasks took weeks, months, or even years.
Posted on Reply
#54
LittleBro
Dr. DroHow? It's DLSS applied to the native resolution, ergo only scene reconstruction algorithm is running. It can't be blurry unless the frame rate is faltering, the perception of blurriness comes from poor resolution or frame rate. DLAA tends to enhance scene detail in some engines that only enable certain aspects of the image when the resolution is high enough.

SSAA was simply increasing the internal rendering resolution (NV DSR and AMD VSR were a way to universally apply this without support in the game), MSAA has always been the clearest in my opinion but it's very resource-inefficient and has many compatibility issues with modern rendering techniques (iirc it doesn't play well with deferred rendering).
DLSS itself causes distortions (blurs image). You won't notice it so much because there's a series of various filters and sharpening tools after upscaling. You actually answered your question.
tpuuser256The whole point of AI is to surpass the limited human intelligence as well as workforce automation. If all goes well AI will be better than humans at pretty much anything including creating stories or generating worlds (the model would use algorithm for more realisitc and efficient compute)

With nuclear fusion, energy could become much cheaper in the near future and allow exponential growth of civilization (technology and number of people)
It's fair to mention that this is not a real AI. This term is being misused too much nowadays for naming any neural network (model) involved system which gets accelerated by GPUs. Neural networks, machine learning, deep learning is part of AI, but not AI itself.
Posted on Reply
#55
igormp
LittleBroDLSS itself causes distortions (blurs image). You won't notice it so much because there's a series of various filters and sharpening tools after upscaling. You actually answered your question.


It's fair to mention that this is not a real AI. This term is being misused too much nowadays for naming any neural network (model) involved system which gets accelerated by GPUs. Neural networks, machine learning, deep learning is part of AI, but not AI itself.
I mean, academically AI is a more generic term that could range its meaning from a basic rule-based system (bunch of if/else) up to ML/deep learning.
Aren't you thinking about AGI?
Posted on Reply
#56
Zach_01
igormpI mean, academically AI is a more generic term that could range its meaning from a basic rule-based system (bunch of if/else) up to ML/deep learning.
Aren't you thinking about AGI?
I remember since for ever the in game NPC opponents behavior to be called the "AI of the game" which exactly falls under the example of if/else I think. Couldn't learn anything... lol
I think of AI as a general term. There are terms after AI that can distinguish it into multiple levels, from sub-human, on par, up to beyond.
All some form of AI though.

To me the idea of intelligence is not equal to something necessarily high.
Is not correct to say... this has low/very low intelligence? I think it is.
And when it is made it is artificial.
Posted on Reply
#57
Kapone33
I am confident that if the title said AMD this thread would be 20 pages. Here we go with the FSR comments from people that have never used it in a Nvidia thread.
Posted on Reply
#58
JustBenching
kapone32I am confident that if the title said AMD this thread would be 20 pages. Here we go with the FSR comments from people that have never used it in a Nvidia thread.
It would just be comical, imagine "AMD reveals secret weapon behind FSR". I mean why would it even need to be secret, nobody would try copying it, it's bad.
Posted on Reply
#59
LittleBro
kapone32I am confident that if the title said AMD this thread would be 20 pages. Here we go with the FSR comments from people that have never used it in a Nvidia thread.
Funny thing is that they don't even know that Nvidia cards can do FSR as well as XeSS.
JustBenchingIt would just be comical, imagine "AMD reveals secret weapon behind FSR". I mean why would it even need to be secret, nobody would try copying it, it's bad.
If AMD is not keeping it's own secret in form of hidden supercomputer somewhere to continuously smooth out FSR model as Nvidia does with DLSS, then if you compare effort invested in the technology by AMD and Nvidia, AMD's results are not so bad.
Posted on Reply
#60
JustBenching
LittleBroIf AMD is not keeping it's own secret in form of hidden supercomputer somewhere to continuously smooth out FSR model as Nvidia does with DLSS, then if you compare effort invested in the technology by AMD and Nvidia, AMD's results are not so bad.
Sure, does it matter? Im not paying for the effort, im paying for the result.
Posted on Reply
#61
LittleBro
JustBenchingSure, does it matter? Im not paying for the effort, im paying for the result.
Whether it matters, that's personal. It does certainly shed other light.
Posted on Reply
#62
Dr. Dro
kapone32I am confident that if the title said AMD this thread would be 20 pages. Here we go with the FSR comments from people that have never used it in a Nvidia thread.
You do realize Nvidia cards can run all vendors' upscaling solutions, right? As for the "secret weapon behind FSR" argument... it's been there since day one? It is open source, after all...
igormpIt's a ML-based upscaling as well, but it doesn't make use of any extra specific hardware. RDNA3.5 (which the PS5 Pro kinda uses) has some extra instructions meant to process some stuff relevant for matmul in lower precision, you can read more about it here:
chipsandcheese.com/p/amd-rdna-3-5s-llvm-changes

With the extra hardware bump, it should be able to run an upscaling CNN without much issues and no need for extra hardware (apart from what's in the GPU itself).
I was aware that this was a RDNA 2 solution with some improvements ported from RDNA 3-3.5, I suppose this is one of them. Weird custom chip.
Posted on Reply
Add your own comment
Nov 19th, 2026 17:18 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts