Sorry for lazy question but some people here know this off top of their head so asking. Memory bandwidth of this chip?
Last time I check an NVidia situation was for DGX Spark (the GB10 chip), it has regular LPDDR5X which by JEDEC standard cannot go beyond ~270 GB/sec, ie 8533 Mbit/s on a 256 lanes bus.
So yeah Lemire seems to go "OMG unified memory, they're following Apple path..." ok, but Apple pulled off a much faster interconnect, 800 GB/s ballpark, and I'm trying to understand (not really, I'm asking you to try understand, he he) how is this laptop faring in that regard.
It's bland and generic because it's a manifesto. Author (and HN submitter) is Loris Cro, aka @kristoff_it, VP of Community at Zig Software Foundation.
In his role, devising as set of general guidelines to use as compass when things (inevitably! and often!) get very very muddy and Right v. Wrong is hard to tell apart -- both objectively, and also from the point of view of being a community leader with ton of vested interest -- is essentially one half of his job. Other half is abide to said guidelines.
So @kristoff_it last week sat down, came up with three simple rules short enough he can print on a business card (or hang on his office wall or whatever), and posted them here to test if they make sense to the wider community.
TLDR: yes can seem bland / generic but within context it makes sense to me author needed to distill his ethics in a nutshell.
It's ok to post stories from sites with paywalls that have workarounds.
In comments, it's ok to ask how to read an article and to help other users do so. But please don't post complaints about paywalls. Those are off topic. More here.
I'm sorry I don't understand. From the way you frame it, and the sentiment of the replies, seems like this is some scary big number. MacBook M3 Max is a beefy machine and doing inference means it's going at full send. 50W is... what tiny appliances consume. Sure it's more than reading emails but... it's still not a number to be shocked at. An on-the-go laptop has a TDP (max rated power) of 45W. Regular work laptop is 70W. Gaming laptop 230W. The servers I have in the lab on which I run benchmarks counting syscalls per seconds for days on end (you know, performance engineering!) are now going north of 1kW.
Washing machine 900W. Hair dryer 1500W. Pizza oven 2000W. So yeah, you say 50W, yeah sure same as video rendering or gaming I guess, yet not really an OMG-level number.
And frankly I'm not quite sure there's anything like economy of scale where it gets more efficient if you serve more users (like some sibling comments seem to imply).
Last thing, and I know many know but also many others don't or have forgotten: Watts is a rate of consumption, not an absolute amount. That is Joule, energy. So you say 50W, but what you pay for (or the planet pays, whatever) generally is the amount of energy, hence you need to say for how long that consumption was sustained. 50W over 2 hours, that's 100 Joules, the actual resource you consumed and paid for.
Power (watts) is like speed (m/s). You say 50 miles an hour, need to say how long was the drive, so we know how far you got.
50 watts over 2 hours is 100 watt hours (Wh) which is 360 kJ. A joule is a watt second. For reference, battery capacity is often measured in Wh and household electric power use in kWh.
Also, datacenter scale devices are almost certainly designed to minimize energy use per operation given comparable latency. You can still compete as an on prem consumer by (1) repurposing your existing hardware, which saves on high CapEx costs, (2) increasing latency, getting your answer computed in a longer time, which probably saves at least some power by design if you can leverage e.g. NPUs, or (3) running smaller or more bespoke models that aren't worthwhile for the bigger players to serve at scale.
There's also a likely gain in serving more requests in parallel, but it may have more to do with successfully amortizing memory access for model weights than any inherent increase in efficiency. Anyway, I've argued in sibling comments that you perhaps can also leverage this on consumer hardware for the special case of DeepSeek V4.
> _50 watts over 2 hours is 100 watt hours (Wh) which is 360 kJ._
Yes of course that was a brain fart of mine. Watt is Joule per second not certainly Joule per hour. I made the point of "lecturing" readers on power v. energy since Antirez (OP) wrote _"50W of energy usage..."_ (instead of power consumption) and it's a mistake people often make. So my side point was: ok 50W but for how long.
The other thing I'm arguing is 50W is nothing to be shocked by. I would like to see an argument for the opposite. I'd like to know what's the power consumption of playing eg. Baldur's Gate for a couple hours on a gaming rig and I wager we surpass that by a margin.
Now, the data center economy of scales. You're saying they almost certainly exists. Okay whatever I don't know. Requests served in parallel. Amortizing memory access for model weights. Likely. I'm writing this with some thinly veiled dismissive attitude because I believe that it would be very useful to have hard data on whether or not serving many users v. just one user makes LLMs more efficient. It's an important point with wide ranging implications.
If there is scale, like you claim, and one day a wealthy patron gifts me a 40k USD rig where I can run a frontier LLM locally, then I'd still be making selfish use of the commons (energy, which belong to the planet, all of us, that kinda stuff) because the efficient/responsible choice is to pool and use a cloud vendor (or pool your rig with neighbors etc).
But saying a machine can be more efficient if it serves many users sounds to me a bit like nine women making a baby in a month.
Keep in mind, I said serving many requests in parallel, not just many users. In fact it's even more efficient if you can batch the requests of a large subagent swarm in parallel since this allows for sharing a big chunk of context/KV cache not just the model weights. That's why I raised the possibility of leveraging this same efficiency with DeepSeek V4. If as a user I can get into the habit of just firing off a request to be cranked on in the background and be completed whenever, and I reach a compute-limited performance workload (just like the big inference labs that serve many users concurrently, only on a smaller scale since the overall compute bottleneck hits sooner) that's quite new wrt. local models. It used to be that we could only do that by spending huge amounts of money on very fast RAM and/or scaling out to multiple nodes.
A big cloud vendor does not face the same opportunity, they cannot leverage the repurposing of your own existing hardware. And they'll definitely want to minimize latency in order to get maximum throughput/utilization from the hardware they did buy, even at an emergy cost. That's why I was careful to note latency as a possible factor before.
Ah ok, sharing context/KV cache, I can see that helping. I need to learn more about DS V4, you seem to hint it has some advantages over previous generations in this respect. I haven't followed that closely to quite catch this argument, I'll check it out.
The basic argument is that its KV cache is roughly an order of magnitude more compact than previous Chinese models, which were already very compact compared to the likes of Gemma 4 (though that example is a bit of an extreme). If you pair this with the basic facts of how to maximize LLM inference performance at scale (this was recently talked about in a video lecture on the Dwarkesh Patel YouTube podcast) the case for doing slow batched inference on prem with DeepSeek V4, perhaps even with memory offload, becomes, as I see it, quite obvious. Of course, I'd like to be proven wrong!
Right, Dwarkesh's episode with Reiner Pope. Didn't watch the full video but as soon I saw both going to an old school blackboard with an actual chalk in hand I could tell they meant business hehe :) Thanks for recommending the vid and for the info about DS V4.
Right. "Anarchists are simply people who believe human beings are capable of behaving in a reasonable fashion without having to be forced to. It is really a very simple notion."
That's one of those definitions that's so broad as to make the word being defined meaningless. It's always silly when one re-phrases their position into something trivial that no one would disagree with.
Exactly. Of course they're capable of it. That doesn't mean they will. They have a lot of incentives to behave badly, and there's no way to eliminate them all.
Even under our decidedly non-anarchic regime, people STILL find reasons to behave poorly. I can't imagine removing the disincentive of state punishment would benefit society very much.
By "bad behavior," I mean robbing and murdering and the like, so no need for scare-quotes. Framing the average criminal as the victim of their own circumstances -- which seems to really be in vogue -- is entirely unconvincing to me.
> people STILL find reasons to burn to a crisp.
You make it sound as if turning to crime is less the criminal's decision and moreso nature's.
Yes, but not nature's—the built environment and socially constructed institutions of modern civilization.
Conservative political scientists like James Q. Wilson have historically argued that the root of crime is an essential moral and cultural failure, rather than just a byproduct of poverty. They maintain that social programs squander investments on those who will simply continue their destructive ways, and that society instead needs punitive mechanisms to regulate inherently destructive human urges.
On the other hand, sociologists and criminologists argue that while the decision to commit a crime belongs to the individual, the conditions that make that decision likely are structural.
Criminologists have long studied "social disorganization" as an engine for bad behavior, analyzing why certain neighborhoods suffer from persistent vandalism, street crime, and violence even as the specific individuals living there change over the decades. Critics of this theory often share your skepticism—arguing that high-crime neighborhoods might simply be the result of "birds of a feather flocking together," and that individual choices or family nurturing are far more important than neighborhood effects—but, ultimately, research demonstrates that people are profoundly motivated not only by their own choices, but by the circumstances and choices of those around them. When community social capital is high, networks of trust enforce positive standards and provide mentors and job contacts. When those adult networks and institutions break down, individuals are left to their own devices, making them far more likely to act on shortsighted or self-destructive impulses.
While it doesn't explain 100% of crime, this is just true. You change people's circumstances such that crime isn't rational, and they're less likely to do it.
There is a reason that crime goes up a ton when existing tools for survival disappear (e.g. disaster scenarios). When people have paths to prosperity, the need to do crime goes down. When the marginal value of crime is low, people don't do it. You can get there with draconian punishments, but you can also get there with, like, a strong social safety net and general prosperity.
While not the only reason, one reason that my coworkers won't steal my wallet if I leave it somewhere is that the $20 is mostly irrelevant to them given the general level of prosperity at my office.
I'm willing to bet most burglars aren't motivated to do crime due to suffering from starvation-level poverty; there is hardly ever a "need" to do crime -- i.e., a scenario wherein doing something criminal is the only way to survive. You totally neglect the moral angle and reduce it to a barebones cost/benefit sort of judgement, which is reflective of this popular view of criminals as hapless victims of fate or of society, and who are almost righteous in their choice to do crime. Oh, and the only solution is more welfare.
I said that for many people crime is a rational approach to more prosperity. That doesn't mean folks are near starvation and have no other choices, it just means that criminal options may be more appealing than other ones. If you create accessible, non criminal pathways to prosperity, crime decreases..if you remove them, it goes up.
it's heavy lifestyle restrictions that lead to anti-social behavior in the first place. by far the most common crime is property crime, people usually commit it out of desperation and lack of opportunity. the degree of personal freedom in a capitalist state is defined by wealth, which creates a natural incentive to steal. then when they do, those people are put in prison, where they connect with other labeled criminals, all of whom face significantly lower chances of being hired, making sure that doing anything else in their life except crime will be as difficult as possible. aren't those heavy lifestyle restrictions enforced on people by government?
As the person who posted the quote, gonna be direct: no idea.
I have to say, I don't identify myself as a anarchist (maybe a bit of a sympathizer), yet I'm middle aged and finding myself a little dissatisfied by many things I see around me, so if I see people making the equation anarchist = degenerate, my immediate reaction is "yeah let's slow it down shall we."
Fair. But I think that statement isn't meant as a strict and precise definition (eg. Stanford Encyclopedia of Philosophy or whatever), more like a "gateway" description directed at those who associate anarchism only with utter chaos and "burn the house down" kinda attitudes.
Now, I'm aware that when you need to say something is "gateway" that's a bit of a red flag, i.e. "milk before meat" (describing something as friendly and innocent at first, then only later showing the more aggressive indoctrination) is exactly what cults do. Having said that, I'd grant that the late David Graeber is quite the straight shooter so I think he's in the clear here.
When I recognize this pattern (reducing one's beliefs to a line of common sense) in someone's writing, I usually take that to be evidence that they're not a quality thinker. I've skimmed the rest of the article you linked from Graeber, and I think my first impression holds up. Like, take this snippet:
> Everyone believes they are capable of behaving reasonably themselves. If they think laws and police are necessary, it is only because they don’t believe that other people are. But if you think about it, don’t those people all feel exactly the same way about you?
Woah, mindblown! If you think about it, aren't you kind of a huge hypocrite and elitist for doubting that others can control themselves? Well, no! We know that plenty of people do, in fact, decide to act criminally and selfishly of their own accord. This line, and many others in Graeber's article, are goofy and I wouldn't take him seriously on this topic.
> "Anarchists are simply people who believe human beings are capable of behaving in a reasonable fashion without having to be forced to. It is really a very simple notion."
If it were that simple, then every FOSS project would be considered to operate under Anarchists principles. After all, the license and software forkability made it so that no one is forced to conform to whatever social structure is used to maintain a given project. But in real life, Anarchists will still argue that a Benevolent-Dictator-For-Life governance approach is wrong, even if it applies to digital artifacts that have zero marginal cost.
There may be plenty of good reasons for them to argue that, but none of them are "very simple notions" as your definition would imply.
> But in real life, Anarchists will still argue that a Benevolent-Dictator-For-Life governance approach is wrong, even if it applies to digital artifacts that have zero marginal cost.
no they won't, FOSS project's governance model has no relevance to anarchist discussion. anarchists are against coercive authority, not leadership in general, and FOSS does operate under anarchist principles, which is why anarchist community is a strict subset of FOSS community.
Anarchist developers have argued that, a lot, on public FOSS mailing lists.
I strongly suspect if you asked ChatGPT to pretend to be an anarchist FOSS developer, it would argue about your FOSS governance model because that's the data it was trained on.
If we define "leader" as "someone who commands by force or by some other means the obedience of a group of people" then Anarchy is a society without leaders. It doesn't mean a society without order, but it presupposes that people can behave reasonably and that that is enough to ensure order.
Your "Other means" could almost be an essay prompt.
There's distinctions between power and violence (see Hannah Arendt), between social and structural power (see The Tyranny of Structurelessness).
And then there's this ancient Chinese text that has been slopified for a million management manuals:
The best leaders are those their people hardly know exist.
The next best is a leader who is loved and praised.
Next comes the one who is feared.
The worst one is the leader that is despised.
The best leaders value their words, and use them sparingly.
When they have accomplished their task, the people say, "Amazing! We did it, all by ourselves!"
To me this essay was an eye-opener, both because it's well argued and also because it's so obvious once you read it. Even outside the specific niche of feminist groups in the US, who hasn't witnessed this phenomenon in action? Those supposedly flat groups where everyone has a voice, yet it's always the same subset of people who are heard and ultimately influence or direct all decisions? And the unwritten rules who are both invisible and "the law".
It sounds great until you see what kind of actual people operate under the banner of anarchism. Then it might turn out their definition of reasonable fashion may be quite different from yours.
> almost all of your daily behavior is an anarchistic expression. How you deal with your neighbors, coworkers, fellow customers in shopping malls or grocery stores, is often determined by subtle processes of negotiation and cooperation.
Exactly. I see elsewhere in this page people comparing this project to Linus Torvalds starting an OS in his dorm room while studying CS. Like these were "young and clueless" devs writing an OS for fun.
From the looks of it, this seems like a serious corporate backed project made by employees of the Ant Group, the chinese fintech giant. A more fair comparison would be with Google's Fuchsia OS (defunct) or Huawei's HarmonyOS. It may succeed, it may fail, but it's nothing like a couple of kids doing a passion project to learn Rust.
I've never used one, but I assume there are professionals for hire to do this.
In fact I've read one of the advantage of going with a publisher (as opposed to self-publishing) is they give you an editor. But again, I'd expect a freelance market to exist.
If I could add one prescription to TFA, it would be to avoid using "just" (the adverb, as in "simply") at all costs.
"A monad is just a monoid in the category of endofunctors."
Maybe it is, maybe it isn't, but none of that is simple. There's that math joke about proof methods, and this would be "proof by intimidation".
When describing a process: "To measure the inverse reactive current in unilateral phase detractors, just use an ordinary turbo encabulator". Why "just"? Are there other methods? For what reason is this the preferred one?
When giving advice: "Why don't you just use a bash script?" This implies your suggestion is simpler or more economical than my proposed approach, therefore better, but you aren't supporting its alleged superiority with arguments I can counter, only implying it.
I see. It was the first sentence that came to mind. The point I'm trying to make is, in my experience, "just" is often abused, as it conveniently relieves one from providing a sound argument.
Yes, good suggestion. I've also stopped using "just" when speaking or writing in the last few years. You can almost always drop it and retain the same meaning without sounding rude or dismissive.
I've noticed this in at least one other language I'm loosely familiar with (Bulgarian) and it holds there too. When the speaker adds "просто" it has the same effect of sounding rude and dismissive.
What I meant by it is "the fine article", meaning the post. Actually using acronyms goes against "Don't assume knowledge", incidentally a point in Eva's list (the article).
I've learned the acronym here on HN, googled for a second and found "the fine article" as explanation, but now that you ask I've checked wiktionary... and apparently the commonly accepted meaning is derogatory (the f*king article, like "RTFM", "read the F-ing manual).
Lesson learned, won't be saying "TFA" again unless I mean f-ing
Thanks for the references. I'll check them. I'm extremely biased against the business use of passive voice, to the point of getting so worked up when someone does it with me on mail or slack, that I need go to a quite room and relax for 5 minutes.
The purchase order has been made.
Your virtual machine has been created.
I see this a lot in corporate environments, and it's by low level managers thinking every bit of information they're trusted with is so delicate and confidential, they must go above and beyond to reveal the least possible amount of it. Note, in the examples above, it was the manager themselves doing the action. What's wrong with "I did this", "I did that"? This is LARPing as CIA agents (or whatever), and they like the sound of it (I want you to know I have information I can't share, you little shit!).
"I've just created your virtual machine" peasant
"Your virtual machine has been created" special ops elite force management
Mind you, 99.999% of the time, the concealment of "who did the thing" is totally unnecessary. It's only there to reinforce status.
"I've just created your virtual machine" - 3 year old, "mummy, mummy, /I/ did this! praise ME!". Got to insert yourself as the first thing in the sentence, as if the customer cares who did it.
"Your virtual machine is ready" - Waiter, assistant, comfortable out of the limelight to let the focus be on the customer and what they are interested in.
"I've just created your virtual machine" - You didn't make the deployment pipeline, you didn't build the hosting, or write any of the hypervisor or guest OS, you ran one script and now you're taking all the credit. Embarrassing.
Jon Hall was essentially Linus Torvalds' agent in the 90s. While at DEC, he got Linus to the right places and meet the right people, so that his "hobby, nothing big or professional" OS entered the trajectory to become what we know today.
I'm about 30 years younger than Jon Hall, so I couldn't be familiar with his accomplishments other than by oral/written accounts. Since he hasn't written big hit books I could read or software I could use (alright alright, Linux for Dummies), I constantly saw people calling him a legend but never understood why. I finally asked around in the Linux kernel community, and was explained the extent of his contribution: he was Linus' mentor in a way. When they met in 1994, Linus was a 25 yo student and Jon a 44 yo DEC marketing manager. I like to think of their conversation as something like "Listen to me kid, this is what you gonna do".
With this in mind, a line in Jon Hall's wikipedia bio stands out: It was during his time with Digital that he initially became interested in Linux and was instrumental in obtaining equipment and resources for Linus Torvalds to accomplish his first port, to Digital's Alpha platform.. Another one in his linkedin work history reflects this view: Senior Marketing Manager, DEC, 1983-1998: In 1994 met Linus Torvalds, recognized commercial value of Linux, obtained funding for port of Linux to 64-bit Alpha processor, opening up a billion dollar line of Linux-based High Performance Computing Super Computers.
Last time I check an NVidia situation was for DGX Spark (the GB10 chip), it has regular LPDDR5X which by JEDEC standard cannot go beyond ~270 GB/sec, ie 8533 Mbit/s on a 256 lanes bus.
So yeah Lemire seems to go "OMG unified memory, they're following Apple path..." ok, but Apple pulled off a much faster interconnect, 800 GB/s ballpark, and I'm trying to understand (not really, I'm asking you to try understand, he he) how is this laptop faring in that regard.
reply