Nvidia DLSS Source Code Leaked
techpowerup.com
techpowerup.com
To get an idea of what image upscaling can do, check out this subreddit dedicated to artificially increasing the resolution of old video game assets to give them a new life: https://www.reddit.com/r/GameUpscale/
Here's eg some personal experiments with the game Baldur's Gate: https://www.reddit.com/r/baldursgate/comments/aemu8u/upscali...
- Image in: https://i.imgur.com/xMbqIeo.jpeg
- Image out, without human intervention: https://i.imgur.com/nC2o1vM.jpeg
The community is using tools such as ESRGAN for that (https://github.com/xinntao/ESRGAN), which leverages GPU acceleration as it uses PyTorch.
So far, this upsampling ability was only used in commercial games: https://en.wikipedia.org/wiki/Deep_learning_super_sampling
That's already a thing (especially with AMDs FSR already being open source), but regardless keep in mind that DLSS 2.0+ requires engine support because it also uses things like motion vectors. Retrofitting this into old games is possible, but likely non-trivial, and would require the game to be adjusted it's not something the driver can just do.
> However, recent NVIDIA GPUs contain dedicated silicon to perform this type of upscaling, which I'm really hoping will make things even faster!
There's no dedicated silicon for DLSS. It's "just" the tensor cores, which are programmable and is something PyTorch already has access to. CUDA support for those have been around for a while after all: https://developer.nvidia.com/blog/programming-tensor-cores-c...
You're right, corrected.
It is also a legal minefield. Nvidia is certainly going to look closely at whatever may contain the leaked code. They are unlikely to do much if you just copy their techniques (unless they are patented, you need to watch that out too), but on community projects, you have to make sure that none of your contributors copy the code itself.
ReactOS (essentially a Windows reimplementation) explicitly tells people who had access to the Microsoft source code not to work on the relevant parts of the project. Probably for that reason.
Be aware that anyone who works in projects conceivably related to the source code in question needs to be very careful to never see it. Otherwise it becomes a legal minefield where future code can be accused to have been stolen (even if it very clearly never was).
But I will admit that incidents like these or the Windows XP source leak are fascinating. I'd hope others share the same curiosity I have for getting to "peek behind the scenes" of some impressive software and engineering accomplishments.
But... an initialism is an acronym. So it's both.
> an initialism is an acronym that is pronounced as individual letters
EDIT: Guess it depends on the source; even my link above says an acronym is "a pronounceable word", but another link I had clicked on had the quote above. shrug
Edit: even if the Nvidia blob were used, it would (i) be a big departure from Mesa architecture and infrastructure and (ii) it would be illegal to distribute it. Nvidia could also introduce an arbitrary amount of protocol madness to make accessing the firmware blob as difficult as possible.
Fortunately Nvidia provides all their drivers free of charge on their website, so it shouldn't be hard to download and extract it at install time?
Not really. Having a closed-source on-card firmware blob but linking it with an open source kernel-side driver would have so many benefits, first of all not needing to depend on NVIDIA to keep up with upstream changes and stuff breaking at every kernel, xorg or whatever upgrade.
This is a good thing, as this code should be studied and documented and free alternatives written while referencing this documentation.
By country count, in most of the world (>50%), there is not a notion of "stealing source code." There are notions of stealing cattle and pottery. But legal frameworks have not really caught up.
Also, their Clean Room Guidelines: https://wiki.winehq.org/Clean_Room_Guidelines
Valve and everyone has been stealing pc games for 23+ years since 1997. In case you weren't aware, in the 90's any hacker worth his salt, if you suggested taking up any client-server app you were looked at as a pariah, it's the ultimate security risk.
Trusted computing is literally them stealing ownership of your PC, the TPM and "security chips" are all about ending you controlling your pc, you can't have any privacy when microsoft can force update your bios with windows update. That's what the big plan was since the mid 90's.
Go have a read, if you still think personal computer ownership isn't under attack you're clueless:
https://web2.qatar.cmu.edu/cs/15349/dl/DRM-TC.pdf
You don't grasp that two or more computers in a network become and hehave as single device. So you never want a client-server exe, that's the same thing as getting your computer hacked and owned, they can spy on you and do anything because you can't audit the source. That's why anyone with a clue wants to avoid windows 10+ with TPM chips because they can remote update your system and disable cracked or other software they don't approve of, when there is plenty of legit reasons for cracking old software.
It's also why level editors disappeared in AAA Games, basic multiplayer was a feature that came in every AAA Game in the 90's until Richard garriot had the bright idea of stealing the networking code, coding it in a weird way and selling it back to the public minus ownership.
You don't seem to grasp after the 2008 bailouts and the war in IRAQ, you live in a lawless oligarchy who's laws have been bribed into being for 200 years.
Copyright law was originally invented to preserve human culture, in the modern era, it's just been endlessly extended. The average citizen is politically illiterate and ignorant that he doensn't live in a democracy but a lawless oligarchy.
Like - I even agree with most of what was typed, but I don't see it's relevance to anything here.
Nintendo's HDL and source code also leaked from a hack a couple of years ago but given how litigious Nintendo is, nothing came out of it despite everyone expecting some emulators to pop op.
Obviously anyone copying the code verbatim or close to it deserves to get the lawyers coming after them, but plausible deniability should be strongly kept in mind for people doing RE+reimplement work anyway.
Maybe they can borrow some of the DLSS techniques
Hence why dlss 2.0 doesn't need per game training anymore. The "machine learning" part of it seems to be mostly killed off unless they somehow managed to train a fully generic model that can handle all art designs and game engine settings (eg, motion blur on vs. off). But it seems more likely it's a fancy generic take on temporal upscaling, possible influenced by ML but not dominated by it.
Which I guess is something we'll find out with this leak. How much AI is actually in dlss 2.0 and how much is "just" offloaded temporal super sampling (such as epic's TSR)