opliko
@opliko.dev
If you decide to do it, would this post count as a BGP announcement announcement announcement?
November 26, 2025 at 7:15 AM
If you decide to do it, would this post count as a BGP announcement announcement announcement?
Ah, there was also Gogs but I only used it once for something that turned out to not need a forge at all. It's neat to run (I used it because it was a single binary :) but seems to have a lot less features vs gitea/gitlab.
November 24, 2025 at 10:40 AM
Ah, there was also Gogs but I only used it once for something that turned out to not need a forge at all. It's neat to run (I used it because it was a single binary :) but seems to have a lot less features vs gitea/gitlab.
Sourcehut is also OSS and self-hostable, but I haven't used it enough to say if it's worth checking out tbh.
Tangled can also be self-hosted, but they're focused on self-hosting the knots and I don't know how usable a full self-hosted appview is (especially if you want a private forge)
Tangled can also be self-hosted, but they're focused on self-hosting the knots and I don't know how usable a full self-hosted appview is (especially if you want a private forge)
November 24, 2025 at 10:33 AM
Sourcehut is also OSS and self-hostable, but I haven't used it enough to say if it's worth checking out tbh.
Tangled can also be self-hosted, but they're focused on self-hosting the knots and I don't know how usable a full self-hosted appview is (especially if you want a private forge)
Tangled can also be self-hosted, but they're focused on self-hosting the knots and I don't know how usable a full self-hosted appview is (especially if you want a private forge)
(And using non-XR SoC also explains why they're only using 4 external cameras and only doing monochrome passthrough. I'm honestly really interested to see some good perf comparisons between it and XR2 Gen 2 devices to see how significantly the dedicated hardware acceleration will affect it)
November 12, 2025 at 7:24 PM
(And using non-XR SoC also explains why they're only using 4 external cameras and only doing monochrome passthrough. I'm honestly really interested to see some good perf comparisons between it and XR2 Gen 2 devices to see how significantly the dedicated hardware acceleration will affect it)
Unless they pull a software platform miracle, I imagine gaming on SteamOS on a mobile Snapdragon SoC is going to be pretty limited at the start.
This is roughly comparable in specs with Quest 3 and Galaxy XR, phone not desktop-VR level.
This is roughly comparable in specs with Quest 3 and Galaxy XR, phone not desktop-VR level.
November 12, 2025 at 7:24 PM
Unless they pull a software platform miracle, I imagine gaming on SteamOS on a mobile Snapdragon SoC is going to be pretty limited at the start.
This is roughly comparable in specs with Quest 3 and Galaxy XR, phone not desktop-VR level.
This is roughly comparable in specs with Quest 3 and Galaxy XR, phone not desktop-VR level.
(I've looked at it deeper around a year ago and it was to find out how one particular part worked, but my general vibe was MS somehow picking the most cursed solution that still seems somewhat reasonable for their problem at most steps)
November 11, 2025 at 11:45 PM
(I've looked at it deeper around a year ago and it was to find out how one particular part worked, but my general vibe was MS somehow picking the most cursed solution that still seems somewhat reasonable for their problem at most steps)
It's not, but it's just bundled and minified JS, fairly easy to read except for psychic damage caused by understanding how it works.
(Fun fact: if you have too many "always installed" extensions it could run into 8191 character limit of cmd.exe which it uses only to call... powershell)
(Fun fact: if you have too many "always installed" extensions it could run into 8191 character limit of cmd.exe which it uses only to call... powershell)
November 11, 2025 at 11:42 PM
It's not, but it's just bundled and minified JS, fairly easy to read except for psychic damage caused by understanding how it works.
(Fun fact: if you have too many "always installed" extensions it could run into 8191 character limit of cmd.exe which it uses only to call... powershell)
(Fun fact: if you have too many "always installed" extensions it could run into 8191 character limit of cmd.exe which it uses only to call... powershell)
From what I've seen sd needs a lot bigger gap between models to be worth it. And here you're only going from 5.1B to 3.6B active parameters. NVIDIA has some small models trained for speculative decoding with gpt-oss (e.g. huggingface.co/nvidia/gpt-o...), but llama.cpp doesn't support them yet
October 19, 2025 at 10:07 AM
From what I've seen sd needs a lot bigger gap between models to be worth it. And here you're only going from 5.1B to 3.6B active parameters. NVIDIA has some small models trained for speculative decoding with gpt-oss (e.g. huggingface.co/nvidia/gpt-o...), but llama.cpp doesn't support them yet
Though I think in this case I think a lot of people would be surprised that TLS is actually end to end - since the term is associated with interpersonal communication with servers just in between, but here the bank is actually the other end :)
October 7, 2025 at 9:38 AM
Though I think in this case I think a lot of people would be surprised that TLS is actually end to end - since the term is associated with interpersonal communication with servers just in between, but here the bank is actually the other end :)
But someone at some point had to design each and every one of these parts for the laptop to boot, and there is no single person who understands how exactly even the finished product works.
September 25, 2025 at 5:52 PM
But someone at some point had to design each and every one of these parts for the laptop to boot, and there is no single person who understands how exactly even the finished product works.
But that's different than the engineering perspective: what do I need to know to recreate this? In this case even assuming you're not on the manufacturing side and someone can figure out how to make things given detailed description - which tbf also reveals what abstraction layer you're used to.
September 25, 2025 at 5:52 PM
But that's different than the engineering perspective: what do I need to know to recreate this? In this case even assuming you're not on the manufacturing side and someone can figure out how to make things given detailed description - which tbf also reveals what abstraction layer you're used to.
The organic parts are a lot more difficult and there is some recursive complexity in understanding the underlying science there that probably can fairly quickly reach the limits of our knowledge.
September 25, 2025 at 5:52 PM
The organic parts are a lot more difficult and there is some recursive complexity in understanding the underlying science there that probably can fairly quickly reach the limits of our knowledge.
The specific of their manufacturing? Probably. But I suspect there is fairly limited design outside of that - I suspect there are food chemists who could pretty accurately state how roughly one would make all ingredients and what the additives put in them are for and how they work.
September 25, 2025 at 5:52 PM
The specific of their manufacturing? Probably. But I suspect there is fairly limited design outside of that - I suspect there are food chemists who could pretty accurately state how roughly one would make all ingredients and what the additives put in them are for and how they work.
There is certainly a lot that goes into making an egg mcmuffin sandwitch at scale, ans I imagine there is some interesting chemical and culinary engineering that goes into the product, but I'm not quite sure if the ingredients are really that diverse.
September 25, 2025 at 5:52 PM
There is certainly a lot that goes into making an egg mcmuffin sandwitch at scale, ans I imagine there is some interesting chemical and culinary engineering that goes into the product, but I'm not quite sure if the ingredients are really that diverse.
Yeah, and I would say that the complexity of our world is generally underappreciated, but in this case I was focusing on the functionaloty parts actually designed by humans to facilitate that process.
September 25, 2025 at 5:52 PM
Yeah, and I would say that the complexity of our world is generally underappreciated, but in this case I was focusing on the functionaloty parts actually designed by humans to facilitate that process.
Making something like a modern laptop is only possible by delegating a ton of that complexity to other people, who will also delegate a ton of their complexity even further down.
September 25, 2025 at 8:48 AM
Making something like a modern laptop is only possible by delegating a ton of that complexity to other people, who will also delegate a ton of their complexity even further down.
I imagine you didn't consider the chip design level as part of the "challenge" here because it was not your speciality and I'll add that there is probably a ton of stuff I'm not considering here either. The point is: modern devices are extremely complex.
September 25, 2025 at 8:48 AM
I imagine you didn't consider the chip design level as part of the "challenge" here because it was not your speciality and I'll add that there is probably a ton of stuff I'm not considering here either. The point is: modern devices are extremely complex.
And a modern SoC is a lot worse than the components I mentioned there, integrating a ton of hardware IP into a single package and typically including parts specifically designed to resist reverse engineering efforts (for sometimes good security reasons), closing even that hard path to learning.
September 25, 2025 at 8:48 AM
And a modern SoC is a lot worse than the components I mentioned there, integrating a ton of hardware IP into a single package and typically including parts specifically designed to resist reverse engineering efforts (for sometimes good security reasons), closing even that hard path to learning.
A lot of it would require someone outside the vendors that designed them to conduct difficult and expensive silicon-level reverse engineering efforts to learn their inner workings.
I certainly doubt that there is a person who has even a basic grasp of all of this at once.
I certainly doubt that there is a person who has even a basic grasp of all of this at once.
September 25, 2025 at 8:48 AM
A lot of it would require someone outside the vendors that designed them to conduct difficult and expensive silicon-level reverse engineering efforts to learn their inner workings.
I certainly doubt that there is a person who has even a basic grasp of all of this at once.
I certainly doubt that there is a person who has even a basic grasp of all of this at once.
And this is just for chipsets, so mostly just I/O, you'll also need things like disk controllers (modern nvme drives have a lot of logic on board!), multiple PMICs on board of various components (e.g. your RAM), battery management, and probably a lot more I can't think of now
September 25, 2025 at 8:48 AM
And this is just for chipsets, so mostly just I/O, you'll also need things like disk controllers (modern nvme drives have a lot of logic on board!), multiple PMICs on board of various components (e.g. your RAM), battery management, and probably a lot more I can't think of now
And even then I'm not fully sure if they have a single person who would know all of the I/O standards and their specific implementations well enough to give a fairly detailed explanation there. And I'm assuming they don't use any black-box vendor IP.
September 25, 2025 at 8:48 AM
And even then I'm not fully sure if they have a single person who would know all of the I/O standards and their specific implementations well enough to give a fairly detailed explanation there. And I'm assuming they don't use any black-box vendor IP.
I mean, even without going that far down into physics, describing in any detail what even a chipset actually does in that process at even slightly higher level than HDL would probably be impossible for anyone who hasn't worked on that at AMD/Intel.
September 25, 2025 at 8:48 AM
I mean, even without going that far down into physics, describing in any detail what even a chipset actually does in that process at even slightly higher level than HDL would probably be impossible for anyone who hasn't worked on that at AMD/Intel.
Does anyone who doesn't work for AMD PR team call them AMD? Genuinely never heard anyone talk about AMD FPGAs even with specific products, at most I heard "AMD Xilinx".
September 22, 2025 at 12:13 PM
Does anyone who doesn't work for AMD PR team call them AMD? Genuinely never heard anyone talk about AMD FPGAs even with specific products, at most I heard "AMD Xilinx".
It's honestly the only area I don't like traversing in this game, so at least from enjoyment perspective (which tbf seems to be mostly intentional) Bilewater is at the very bottom and has no real competition.
September 19, 2025 at 7:30 PM
It's honestly the only area I don't like traversing in this game, so at least from enjoyment perspective (which tbf seems to be mostly intentional) Bilewater is at the very bottom and has no real competition.