nathanhoobler.bsky.social
Pixel Pusher and Digital Illusionist
@ Nvidia
All opinions my own
62 posts
192 followers
127 following
Regular Contributor
Active Commenter
comment in response to
post
Lithuania has offered citizenship for descendants of political refugees, but unfortunately when I looked into it I found out my ancestor fled from Imperial Russians instead of Soviet Russians so my family wasn't eligible.
comment in response to
post
boy I feel the wisdom of all that but I also feel like it's going to be tricky to thread the needle between conserving ones energy and avoiding the atomization and aversion to politics that characterizes many modern autocracies.
comment in response to
post
I think there's also sometimes a problem with the modern marketplace relying so much on monetization of user data and ecosystem, which turns every product into a loss leader that small, non-vertically integrated companies simply can't compete with on value.
comment in response to
post
Nest smart thermostats (before being acquired by Google)?
comment in response to
post
The UK has a similar Territorial Army dating from the 1920s, and the German Territorialheer was a similar structure that existed from 1957-2001. Is this specifically different than those reserve armies in some specific way?
comment in response to
post
Why does NASA outsource things to SpaceX rather than doing them in house? Well a large part is that SpaceX can hire like a normal fucking company and not underpay people with options elsewhere or take a literal year to go from application to offer.
comment in response to
post
Naively, no: random sampling is used in training and random offsets are used to soften quantization.
However, the compression can be given a seed value, letting it behave deterministically between runs provided identical settings. This should allow for bit comparisons in CI systems (for example).
comment in response to
post
One big difference is that BCn+lz4 decompresses on the CPU and streams the deflated BCn over PCIE whereas NTC streams the (smaller) data from disk and doesn't decompress until vram.
(GPU deflation would obviously help the CPU load in the lz4 case)
comment in response to
post
We did some light testing. As I recall, lz4 gives you something like 2x compression of BCn. We see very little compression of NTC bundles (though we did not try optimizing the format). Net compression of NTC is closer to 2-4x over BCn+lz4 for similar IQ.
(Cont).
comment in response to
post
This is all available as part of the newly released RTX Kit which includes a suite of new rendering techniques leveraging ray tracing and neural inference technology to push the boundaries of real time rendering.
More: developer.nvidia.com/blog/get-sta...
comment in response to
post
The "compression" process produces a set of latent parameters covering the texture space and a lightweight multi-layer perceptron (2x64 nodes) which is small enough to inference at runtime - either on load or within the sampling pixel shader itself using new Cooperative Vector instructions.
comment in response to
post
Physically-based textures include several image layers representing physical properties which, while potentially independent, often contain strong correlations. The actual information content is far less and NTC finds a latent space representing it with fewer and lower bitrate channels:
comment in response to
post
This library implements and expands upon the work presented in "Random Access Neural Compression of Material Textures", and uses neural training techniques to exploit correlation between the different parallel texture channels commonly found in PBR materials.
research.nvidia.com/labs/rtr/neu...
comment in response to
post
The Virgin "kinetic kill vehicle" vs the Chad "enhanced-radiation W66 warhead"
comment in response to
post
Absolutely avoidable tragedy, but also worth noting that the area around DCA is a total mess (due to a combination of military flights, restricted airspace, and increased traffic).
bsky.app/profile/osin...
comment in response to
post
Counterpoint: Daniel Inouye
comment in response to
post
If only someone would retire and leave an open NY Senate seat.
comment in response to
post
It's the "is physics prescriptive or merely descriptive" debate taken to absurdity -- every time the models built to match observed phenomenon predict something weird people are like "well clearly that can't be right, guess we've reached the limit of the math" but nope! Reality is just broken.
comment in response to
post
This is Lee Pace erasure
comment in response to
post
I forget, do you do much DCS or more automotive sim? The big thing I've read about the Pimax displays is that unlike Quest 3 they have the resolution to make the cockpit instruments easily readable without messing with zoom, which seems like a big usability feature. No firsthand experience though.
comment in response to
post
Agreed. It was amazing when I first used it, and then became infuriating as it felt like it would lose tracking, the mappings felt awkward, etc.
I've been carefully eying the Pimax offerings this CES, wondering if it was time to make the leap.
comment in response to
post
That's a surprisingly high number! I would have guessed 15-25%. I'm curious how many of the non-VR users are using TrackIR (or some camera-based variation)...
comment in response to
post
First of all, it was *Autobahn*
comment in response to
post
This is going to look really bad on their EU application
comment in response to
post
It really nailed the challenge of a stealth game where mis-steps don't become immediate failures but a complication you need to juggle, and making success require multiple schools of approach operating in concert.
comment in response to
post
Mail-order compounding pharmacies that offer stuff like GLP-1 and TRT based on a virtual consult with an MD and remote lab work. There are a few I can think of.
comment in response to
post
Based on what we know about human visual perception (at the very least) I'm not sure that's entirely true. I think there *is* a distinction like the one you are making, but I'm not sure it's quite as simple as that.
We aren't always aware of the abstractions and approximations our brain is making.
comment in response to
post
I believe part of what we consider "intelligence" in humanity is creating frameworks of concepts about our universe that lets us make sense of and expand our understanding.
A necessary (if not sufficient) part of AGI must be accurate extrapolation of expertise to domains beyond its training set.
comment in response to
post
Honestly I don't know if I am more surprised by the fact that there is a Desmos Demo scene or by the realization that I didn't just assume that one existed before.
comment in response to
post
To be fair, I thought the telegraph repair bit did make it into a scene in the film? So adapted in spirit at least.
comment in response to
post
The mixbox talk is actually a really good intro to some basic colour theory, very clear explanations, highly recommend watching www.youtube.com/watch?v=_qa5...
e.g. I'd never thought about "pigment colour is just subsurface scattering" but of course it is!
comment in response to
post
The oldest dinosaurs are farther in time from the youngest dinosaurs than the youngest are to us. We are closer to the T. Rex than the T. Rex is to Diplodocus.
comment in response to
post
For someone with no real knowledge of EU governance, how consequential has Hungary's presidency been? I had the impression the office was more administrative (not that it couldn't still be bent to influence policy).
comment in response to
post
I don't think Anderson is surreal so much as dreamlike (more Impressionist than Surrealist). His visuals and stories *make sense* but in an exaggerated way. Fincher I think is similar, but with a darker tone.
Lynch isn't quite fully Dada but closer to the "signifiers w/o explicit meaning" approach.
comment in response to
post
Also "Wes Anderson Spy Thriller" is just _The Grand Budapest Hotel_
comment in response to
post
I wish I still knew where that book was