HNHacker News
TopNewBestAskShowJobs

kragen

39,314 karma · joined November 27, 2008

http://canonical.org/~kragen/
submissionscomments
kragen··on Greg Kurtzer's Ginux is now a userspace kernel
No, I writing that in response to "If you can't trust Linux because it is gigantic, you can't trust a whole another new kernel. Userspace is not blackmagic." The nature of Ginux is irrelevant to that proposition, and, as I think I have clearly shown, it is not well thought out.
kragen··on Can a Linear Program Do Weight-Space Programming?
No, the article we are discussing here used a GPU solver implementing the primal-dual hybrid gradient method, which cannot do any of those things. (This is not the same thing as the simplex algorithm from the 01940s. I was wrong about that.) The LLM discussed in the previous work was not in fact able to do that itself. Like a normal person, it wrote a Python program to do it. (Probably using the simplex method, I'm not sure.)

While the apparent advent of AGI is indeed really crazy, the point of this article is sort of the opposite.

kragen··on A great old-timey game-programming hack (2013)
Thank you!
kragen··on Fossil CO2 emissions decrease in 2026
Photovoltaic modules produce a voltage or current on their own. It's true that you have to connect them to a load for them to do work such as washing your laundry or heating your oven, but that load can be very cheap. Like, a wire. None of the dramatically more expensive things you're naming are necessary.

In the US you have a negative 100% subsidy on solar panels in the form of tariffs, and the planning and permit prices you're talking about are additional similar negative subsidies, not actual costs. They don't exist in Pakistan, and they could be set to zero tomorrow at the stroke of a pen, which is what has happened with balcony solar in several European countries. That's why you're talking about nonsense prices like US$30k for a household system.

kragen··on Fossil CO2 emissions decrease in 2026
Those latitudes are pretty close, but that doesn't tell the whole story.

At noon on the equinox, Amsterdam gets 80% of the sun that NYC does, and Fairbanks gets 56%. But, at noon on the winter solstice, Amsterdam gets 55% of the sun that NYC does, and Fairbanks gets 7% ofthesun NYC gets. So, in terms of insolation above the atmosphere, Fairbanks is dramatically worse than Amsterdam. If the sine of the sun elevation was the whole story, to survive the winter on solar, you'd need 8× the solar panels in Fairbanks you'd need in Amsterdam, which would need less than 2× New York.

In fact, cloud cover is very important, and I think it narrows the gap a bit.

kragen··on Fossil CO2 emissions decrease in 2026
You may be computing Carnot efficiency backwards; heat pumps are less impactful for producing higher temperatures, not more. Also, there are practical difficulties with trying to build a heat pump that can function at several hundred degrees.

Aside from pfdietz's good point about regenerators, resistively melting iron with electricity doesn't require any high-temperature exhaust air.

kragen··on Greg Kurtzer's Ginux is now a userspace kernel
Of established languages, Rust, OCaml, F#, and Lean seem to be promising for this kind of thing, but people seem to be vibecoding a lot of new languages right now with this use in mind.
kragen··on A great old-timey game-programming hack (2013)
The PLA is sort of like microcode?
kragen··on A great old-timey game-programming hack (2013)
People wrote magazine articles and books with stuff like this. Dr. Dobb's, early BYTE, Don Lancaster's books, etc. I even remember typing in DOS TSR .COM files in DEBUG.COM from PC Magazine. The problem was, writing magazine articles was a much more limited channel than Usenet, BBSes, blogging, Stack Overflow, or GitHub. Dr. Dobb's published less than one article per day, for the entire world. And things like CACM weren't interested in things like the PUSH-U/PUL-Me.

So orality was more central.

kragen··on A great old-timey game-programming hack (2013)
I thought a lot of games just use Godot, Unreal Engine, or that other one that decided they were going to change their licensing terms to demand per-copy fees from free-to-play games that their customers had already written with it. Does your company have an in-house Godot alternative, or are you having to do things like this on top of one of the existing engines, or what?
kragen··on A great old-timey game-programming hack (2013)
This is a very nice algorithm, used by agrep for fuzzy string search. Wikipedia says:

> The bitap algorithm for exact string searching was invented by Bálint Dömölki in 1964 and extended by R. K. Shyamasundar in 1977, before being reinvented by Ricardo Baeza-Yates and Gaston Gonnet in 1989 (one chapter of first author's PhD thesis) which also extended it to handle classes of characters, wildcards, and mismatches. In 1991, it was extended by Manber and Wu to handle also insertions and deletions (full fuzzy string searching). This algorithm was later improved by Baeza-Yates and Navarro in 1996.

But you know what all those guys didn't invent? They didn't invent BBSes and XMODEM. Ward did.

kragen··on $30M Raised to Redo the CPU
The Alpha left out byte and halfword memory access instructions, and ARM left out halfword access instructions; both had to correct the omission later. I said RISC-V was determined not to repeat "what they saw as the blunders of past designs", and you are quoting from that list of blunders, so it had byte and halfword memory access instructions from the beginning.
kragen··on Can a Linear Program Do Weight-Space Programming?
That's not what's being described here; that's described in another paper that this is a followup to.
kragen··on Can a Linear Program Do Weight-Space Programming?
No, it describes a linear optimization algorithm from the 01940s editing a Qwen model with billions of parameters.
kragen··on Greg Kurtzer's Ginux is now a userspace kernel
Another new kernel might be 1/6000 the size of Linux and therefore have 1/600 times the exploitable bugs—or none, using formal methods. The question is how much a new user-mode kernel can shield Linux from attacks. Usually the new kernel depends on Linux to separate it from the malicious code!
kragen··on Fossil CO2 emissions decrease in 2026
That may be true wherever you live (probably in the US), but it isn't true worldwide. US$30k buys 200 kilowatts (peak) of unsubsidized mainstream photovoltaic modules. Even in the US it buys half that. That's a utility-scale power plant, not a household rooftop installation.

Meanwhile, household rooftop and balcony installations continue to proliferate rapidly in Pakistan, Germany, Australia, here in Argentina, etc.

kragen··on Fossil CO2 emissions decrease in 2026
The Netherlands is much worse than average, but they're not an extreme edge case. Norway is the extreme edge case, and they're also focusing on wind. (Parts of Antarctica get more sun than most of Norway.)

I have no idea why China has such a dismal capacity factor! I suspect they have a shortage of transmission capacity between the sunny places and the populated places. Let me know if you find out.

kragen··on Fossil CO2 emissions decrease in 2026
You have some good points, but I think there are some flaws in them.

Heat pumps are probably now economic only in niche scenarios, even at current PV module prices; they cost dramatically more than the solar panels they displace. And they're really only applicable to climate control, not to other applications of heat, such as cement kilns, electric arc furnaces, or baking bread. I do use one to heat my bedroom in winter, but I don't think they are really relevant to the future.

It is true that electric cooking is more efficient, especially when the heat can be produced in close proximity to the food, as in induction stovetops, crockpots, and microwave ovens. I don't know how to size this, but I don't think it's very large.

40% of global shipping by weight is fossil fuels, not just oil (https://billmckibben.substack.com/p/the-happiest-number-ive-...) but that's by weight, not energy consumption. The vast majority of fossil-fuel shipping is shipped by shipping ships, not by railroads, tractor-trailer trucks, stepvans, small boats, or personal cars, and, if I'm understanding https://unstats.un.org/unsd/energystats/pubs/documents/2026p... correctly, only 2.5% of the global total energy supply ("TES", the UN's term for world marketed energy consumption) is the bunker fuel that powers those ships. (I had guessed 3%.)

So, completely eliminating fossil fuel shipping will save about 1% of total energy consumption.

kragen··on $30M Raised to Redo the CPU
This could be wrong, but https://tracxn.com/d/companies/mill-computing/__5BCN3HKJOoke... says they raised US$100k in debt, and https://www.crunchbase.com/organization/mill-computing-inc says You Have Been Blocked but apparently told Google they were pre-seed funding. https://pitchbook.com/profiles/company/169749-55#funding claims that they raised US$3.2M (US$1.2M in angel funding in 02004 and US$2M in a seed round in 02016). I looked for press releases about funding rounds and didn't find any.

Even in 02016, I don't think US$2M was enough to tape out a silicon prototype of a working high-performance CPU. MPW ("shuttle") programs can give you working prototypes for as low as US$3000, but only in process nodes a decade or more behind the state of the art. (This is fine for analog, but for high-performance digital designs it's generally fatal.) US$30M is a much more plausible budget for the multiple rapid-fire tapeouts you need for anything close to a state-of-the-art process.

kragen··on $30M Raised to Redo the CPU
Don't forget scan! You know. Prefix sum.
kragen··on $30M Raised to Redo the CPU
RISC-V intentionally contained almost no innovation of any kind. The whole point was to have the most boring, standard RISC design imaginable, without anything clever in it. That's what you want for introducing students to their first RISC architecture, and also as a basis for doing computer architecture or CPU fabrication research: on zero-overhead loops, compressed instruction sets, scalable vector instruction sets, IoT machine learning, indium gallium zinc oxide thin-film transistors printed on kapton, zero-knowledge virtual machines, or whatever.

They specifically wanted to avoid the kinds of "quirks" that vendors like Intel and ARM used to make people subject to their patents.

They did try to avoid what they saw as the blunders of past designs, like not leaving enough opcode space for compressed instructions, making instructions conditional, leaving out byte and halfword memory access instructions, having a flags register, having a complicated page table structure, having register windows, having delay slots, etc. None of that was innovative. The actual design of the compressed instruction set and the vector instruction set were innovative, but that came later.

The Mill was the opposite extreme.

kragen··on $30M Raised to Redo the CPU
That's a fun idea! RV32E is 512 bits of general-purpose registers, RV32I is 1024, and RV64I is 2048. AVX-512's ZMM registers total 16384 bits. So you could have, say, a 4096-bit register which you could access as any of 13 different sizes, from 1 bit up to 4096.

I think the on-chip bus interconnect and ALU are probably going to have some finite width, though. Above that number of bits in an operation, your instructions will get slower. Maybe that's what you have in mind, and you just want to abstract that away?

kragen··on Ultorg: The User Interface for Relational Databases
Ultorg is mostly UX research that Bakke published in 02009 when he was working on his Master's Thesis at CSAIL. I haven't heard of any patents on it, but any patents that cover the version of SIEUFERD he published at the time would have had to be applied for by 02010 (02009 outside the US), meaning they'd expire by 02030 (02029). And I think it would have been pretty difficult to get a patent on the things they added later, like adding crosstabs to it.

Having published it so long ago means that it's excellent prior art for patents others might try to use against Ultorg or Ultorg clones.

kragen··on Fossil CO2 emissions decrease in 2026
I think you mean "primary energy", and I agree about that; we discussed that in https://news.ycombinator.com/item?id=50439371.

Rooftop solar has both economic efficiency disadvantages (generally poorer siting, being dangerous and expensive to install) and advantages (no transmission losses, no added load on transmission and distribution systems, and in some cases elimination of inverters and other power electronics). It isn't clear to me whether it comes out to be overall more or less economically efficient, and I suspect that it probably depends on local circumstances, and especially on whether you can reduce or eliminate inverters.

However, the most persuasive case for rooftop solar is basically political rather than economic: it reduces or eliminates the building's dependency on utility-scale generation, transmission, and distribution. This dependency can bite in the form of unreliability (like California's intentional power outages during the now-yearly summer brushfire season, or the "load shedding" rolling blackouts South Africa had for several months last year), in the form of extortionate energy prices due to monopoly rent extraction (California being the poster child here at the moment, extorting 29.75¢/kWh from resident), or in the form of systemic fragility (currently there are blackouts in Kyiv because Russia is blowing up the power stations, which amplifies the coercive effect of Russian attacks on Ukraine).

I want to say, I greatly appreciate your help in getting these figures right.

kragen··on Semi-transparent, colored cadmium telluride photovoltaic panels in agrivoltaics
This is not a good dream. I want to wake up.
kragen··on Fossil CO2 emissions decrease in 2026
Yes, single-axis trackers often get capacity factor over 25%, and solar farms in sunny places very commonly use them. But the picture isn't so sunny everywhere; solar capacity factor in China, Germany, and the Netherlands has been a dismal 10%. So 20% is a ballpark figure, and it's possible to do better under favorable circumstance.

It is true that it would be insane to expect the US to lead the transition to abundant renewable energy, but that doesn't mean the transition won't happen. It just means it will leave the US behind. The US's electricity production (though not energy production!) doubled in the 01960s and then stopped growing, and China's electricity production has doubled over the last 10 years and continues to grow rapidly. Worldwide, we should expect cheap solar panels to lower the cost of electrical energy, opening up new uses for it.

kragen··on Fossil CO2 emissions decrease in 2026
You're right, that's less than a fifth. Thank you.

So, the other 82% of primary energy consumption could be reduced by about 60%, except for the part that's already used for heating (how much is that?), and by about 75% when it's being burned in the inefficient internal combustion engines commonly used in cars and the like.

Ballparking, this seems like fueling the current economy purely with electricity, and counting "primary energy usage" as the output in the case of solar panels (rather than "the sunlight hitting them" as you advocated elsewhere, which is a defensible convention but not one anyone uses) would cut primary energy usage by about a third.

kragen··on Fossil CO2 emissions decrease in 2026
Thank you. One minor detail: producing 20TW per year is in order to have 400TW of generation capacity installed, but only producing 80TW because of the 20% capacity factor.
kragen··on Fossil CO2 emissions decrease in 2026
This is a good point; in particular, electrification of transport will reduce the figure substantially, especially because car engines are usually less than 30% efficient. But electrical power is already about a third of the total, isn't it?
kragen··on Fossil CO2 emissions decrease in 2026
20TW is roughly current world marketed energy consumption. But, on Earth, solar power plants have a capacity factor around 20%, so you need 100TW of solar panels to produce those 20TW. Moreover, we should expect human energy consumption to continue to grow, not suddenly become constant, especially since solar energy is so much cheaper than older sources of energy; 20 years from now we might need four times that much, depending on how AI works out. In that case you'd need to produce about 400TW÷20 years = 20TW/year.

According to the IEA, we currently have only about 1.5TW/year of manufacturing capacity: https://www.solarplaza.com/resource/13581/article-global-sol...

Hydropower is already scaled nearly to its limits, and wind energy is useful and important, but is dropping in cost much more slowly than solar, and it's tapping a much smaller resource.

← PreviousPage 2 of 34Next →