A movie can be duplicated indefinitely. There's no guarantee your song will be appreciated as art. I'm not sure why you say you can't print out an image and hang it in your living room; we do that all the time at home.
I've personally never dabbled in NFTs, but I don't think it's fair to ascribe the inherent conflict between information and scarcity uniquely to them.
I could imagine an AI sidekick that does all this work for you, and always has the last word because it'll never give up.
A place like Meta or Microsoft would tell you to pound sand, but an aligned army of collective-bargaining agents might succeed in removing a specific term from a smaller service.
Plenty of enterprise server hardware (racks, servers, RAM, disks) does have an active secondhand market after 3-5 years of use, but I think GPUs are too specialized for it to be viable. I doubt anyone has the setup to run a H200 in their home rig.
I also don't think companies are going to have mandatory replacement cycles for GPU hardware the same way they do for everything else, because:
1. It is an order or magnitude (or more) more expensive.
2. It isn't clear whether Moore's law will apply to the AI GPU space the same way it has for everything else.
Unless Nvidia can launch a new chip every 2-3 years with massively improved performance-per-watt at a lower price no one is going to rush to recycle the old one.
> I doubt anyone has the setup to run a H200 in their home rig.
There are PCIe versions of these right? And another comment is saying there are PCI adapters too. It "only" requires 600 to 700W. It's not out of reach for everybody.
If the used regular server market is any indication, you can find, after a few years, a lot of enterprise gear at totally discounted prices. CPU costing $4K brand new for $100 after a few years: stuff like that.
A friend has got a 42U rack and so do some homelab'ers. People have been running GPU farms mining cryptocurrencies or doing "transcoding" (for money).
It's not just CPUs at 1/40th of their brand new price: network gear too. And ECC RAM (before the recent RAM craze).
I'm pretty sure that if H200 begin to flood the used market, people shall quickly adapt.
> Unless Nvidia can launch a new chip every 2-3 years with massively improved performance-per-watt at a lower price no one is going to rush to recycle the old one.
I agree with that. But if they resell old H200s, people are resourceful and shall find a way to run these.
Would it even require a particularly high level of resourcefulness? Purchase the GPU along with the mobo that slots it. It's not as though companies typically swap out CPU and GPU while keeping the rest of the box.
They should max out a bit below 6 kW? The H100 SXM5 is 700 w which would place the system at 5.6 kW plus change. Too much for a standard circuit but well within the bounds of a residential appliance.
It's a monolithic 8U rackmount appliance so perhaps a dishwasher would make for a decent size comparison?
Definitely no good if you rent but homeowners should have little to no difficulty. The sort of people interested in such gear usually have multi kW racks already.
> Unless Nvidia can launch a new chip every 2-3 years with massively improved performance-per-watt at a lower price no one is going to rush to recycle the old one.
That's exactly the point.
Performance/watt is increasing so much gen-to-gen that it makes no longer sense to run older hardware.
Are you saying that there hasn't been massive improvement in performance per watt metrics for datacenter GPUs, which directly affects performance and profitability of said DCs?
You said "I guess they must be the expert", sarcastically, implying NVIDIA's CEO either didn't know or was wrong about the idea of perf/watt improvements. If you have evidence this is wrong, it would be great to present it. Otherwise, most reasonable people will accept that Jensen's statement is accurate, even if he's not a neutral 3rd party.
you can absolutely run e.g. datacenter-level A100 at home, there are adapters from the SXM to the PCIe socket. Haven't seen people running SXM versions of H100s this way but this could be due to the price factor only
Well technically true, I would wager that the home lab is going to require increasingly distinct and unusual adaptations to retrofit the hardware to home use.
New stuff is all liquid cooled by default and that's a paradigm shift for your average home lab.
I'm less aware of exactly what's happening on the power side of things but I think some of the architectures are now moving to relatively high voltage DC throughout and then down converting it to low voltage right before it's used. So not exactly just plug-and-play with your average nema15 outlet.
Well by the time the become obsolete you can run that computing on a Mac with no special cooling so I really doubt they will be of any use. Maybe in some parts of the world where electricity is cheap. If someone wants to really find out perhaps watching the crypto ASICs stories could help.
Last I checked AWS is still offering g4dn instances that run on NVIDIA T4 GPUs, which were first released in 2018. I think most people underestimate how long superscalers can keep these things running profitably after they depreciate, and you probably don’t want anything they throw away.
My last employer is still running a bunch of otherwise discontinued g3 instances with 2015 era GPUs.
It's likely the GPU boards are designed for water cooled data center racks and might not fit in a regular PC case. It's also possible the PCB the GPU's are mounted to might not be standard PCIe cards that fit into an ATX case.
I bought a used NEC SX Aurora TSUBASA (PCIe x16 board that looks like a GPU board) and realized it has no fans. The server case it is designed to fit into is pressurized by fans forcing air through eight cards on a special 4 + 4 slot motherboard. I have to stack and mount three 40mm fans on the back.
Honestly, I bought it more for collection than actually using. If I ever do anything with it, it will be from Plan 9. However, searching around there is plenty of documentation on the architecture in addition to a github repo: https://github.com/veos-sxarr-NEC
They are build to physically last 5-7 years in 24/7 datacenter use, but they have effective lifetime just 3-4 years, then their value has deprecated and electricity and infrastructure cost dominates. Meta did a benchmark where 9% of the chips failed every year, 'infant mortality' is much higher in the first 3 months of use.
9% is an absurd failure rate for solid state electronics. Particularly considering the profit margins. I assume it's related to the power densities involved. Would you happen to recall the source?
Depending on the elemental composition, it could definitely be worthwhile to recycle wherever scale is practical. For giant datacenters and companies using hundreds of thousands or millions of gpus, that adds up to a lot of gold and other valuable elements.
In order to take advantage of that, someone needs to be positioned to process all that material economically, and to make the logistics achievable by the big players. If it costs Facebook $10million to store and transport phased out gpus vs just sending them to a landfil, they're not going to do it. If they get $100k for recycling - probably not going to do it. If they pocket $5 million, they will definitely contract that out, especially if it costs $50 million to build out the infrastructure to handle it.
Probably a good company idea - transport, disposal, refurbishment of out of cycle GPUs and datacenter assets, creating a massive recycling pipeline for recapturing all the valuable elements is a pretty good niche.
I've written about this elsewhere but I predict there will be a significant secondary market for repurposing parts of datacenter GPUs (for example, RAM chips) by desoldering them and soldering them onto new PCBs that fit PC/consumer use cases.
I previously ran 150,000 AMD gpus in all conditions at 100% utilization for years. I currently have a multi-million $ cluster of enterprise AMD GPUs.
A couple real world points:
1. They generally don't just fail. More likely a repairable component on a board fails and you can send it out to be repaired.
2. For my current stuff, I have a 3 year pro support contract that can be extended. Anything happens, Dell goes and fixes it. We also haven't had someone in our cage at the DC in over 6 months now.
I have to maintain our GPU's. Generally the worst parts are the watercooling pressure, the HVAC, and the power. I can run it stable only at 300W per CPU, the normal max is 310W. Now with throttling to 300 it's stable for a year, before it burned two mainboards already, with lots of downtimes.
My experience is that power problems stem from not having good power and/or poor airflow.
I'm convinced that this is why we haven't had any issues in our current location. Zero outside air, zero dust, insanely well built zero expense spared airflow and power supply / management.
You send them back to Nvidia or a third party e-waste recycler at end of life. Sometimes they're resold and reused, but my understanding is that most are eventually processed for materials.
Consistent, low-impact enforcement is better at preventing unwanted behavior than the rare and severe but also capricious enforcement performed by human police.
It can also give permission for unwanted behavior. Cf. the Haifa study, where the rate of late pickups increased when daycares added a fine. One explanation is the fine turned a complex moral obligation into an ordinary financial transaction.
The Haifa daycare study can’t be used to extrapolate much.
They fined parents (IIRC) ~$3 per late pickup. Rerun the study with a $300 fine and let’s see how it pans out. It’s an interesting finding, but that then people take it to mean that fines don’t work (no matter their size) is insane.
Well, If the staff got stiffed on the fee "many times", and the parents were allowed to bring their kid back.. the place didn't charge $1 per minute late. They just bluffed and got called on it.
(apologies for the immediate edit, changed my wording)
I understand that you could not keep the child till you were paid (kidnapping and ransom shouldn't be a business plan!), but you could refuse them future service until they paid.
The nursery has clear times that it is "open". I must drop off my child between A and B O'clock in the morning, and I must collect them between X and Y O'clock in the afternoon. Like a shop - they are allowed to have opening hours.
The issue is that they can't just close when they want if there's a child still there. So they have to have some way of enforcing these rules on parents. A "per minute" fine seems appropriate, so that it's more the later you are. And you need the fine to be enough that it is punitive enough, when considered against the income of your parents. Otherwise it provides no incentive. Ours is not $300 (more like $30), but it seems fair.
That's largely in line with unplanned or off-hours work for many professionals in the area of a city. If you want for example, plumbing done after normal business hours $300 per hour is a typical rate. In at least one case I paid $50 just to get a supply shop to open their doors after hours to get the needed parts to repair my own home.
It sounds like a great policy. Good news is that you can choose to be on time or pay the penalty or choose another provider who hasn't decided to implement this...yet.
I'm shocked by your question. I honestly would like to hear why you think this should not be acceptable. Why should they continue working overtime and cut into their own personal/family time because of the parent's failure?
Let's say you have a job interview. You're 5min late, so they either don't hire you, or the receptionist says the interviewer is now not available. Are you now due the salary, because you being this late 5min cost you a lot of money?
If you in a private contract reject the terms of paying $5 per minute late, well then the other party now knows you plan to be late a lot, so they'll be glad if you take your business elsewhere.
Keeping people from being able to go home after their workday, effectively forced overtime, is incredibly disrespectful. And even if "it's not your fault", you are the only one that could have prevented it. So incentives should be in place that you don't. $5 per minute sounds fair.
If you force me to stay late for a full hour you'd BETTER pay me triple digits. But in this case the $300 for an hour may have to be shared among several people.
It's too late to edit my other comment, but it's shocking to me how the people downvoting that comment can have such a lack of empathy and respect for people working in daycare.
I can't understand how one can treat people like servants, forcing them into unpaid overtime, to wait until I'm good and ready to show up. And to be upset and call it "unacceptable" to compensate people when you mistreat them.
A $3 fine is a good portion of someone's disposable income
and a $300 fine is not much of someone else's.. A civil
penalty of that nature almost guarantees some part of
the population will view it like the $3 fee.
This is exactly why license points (leading to suspension) are better than fines.
If the ticketing decision made by an automated camera system is deemed acceptable when issuing mere fines akin to parking tickets, but deemed unacceptable when issuing other penalties (which don't have this wealth inequity issue we are discussing now, at least not exclusively), that's effectively a poor tax.
More broadly, I think it's important to distinguish (more than we do) what aspect of justice a fine is supposed to be for, particularly between restorative versus punitive. The first is what it costs to fix measurable damage-done, the second is what we need to ensure the person cares to change their behavior.
The government operating automatic camera citation systems, almost never is interested in improving safety or even minimizing undesirable behavior- often the placement of such cameras is done to maximize revenue (as when red light cameras are placed at long-cycle-time intersections vs intersections with a history of accidents). And it’s been documented that some cities have reduced yellow light times (which almost always leads to more citations) rather than increasing yellow light times (which usually leads to fewer people running the yellow, because people are less likely to take a chance after the light has been yellow a long time).
There’s a lot of compelling evidence that these systems are just revenue machines.
I don't like paltry sums like this. $300 is a significant financial impact to someone who's barely making ends meet, and absolutely nothing at all to a billionaire. "If the penalty for a crime is a fine, then that law only exists for the poor."
Make it a tiny % of net worth, with a modest minimum and watch EVERYONE obey. Or at least something meaningful to everyone, or don't make it a fine. Use some other carrot or stick.
> I don't like paltry sums like this. $300 is a significant financial impact to someone who's barely making ends meet, and absolutely nothing at all to a billionaire.
One thing about daycares is that you will essentially never find "someone who's barely making ends meet" and "a billionaire" with kids at the same daycare, so a surcharge for out-of-normal-expectations service does not need to be designed to address both cases.
(In fact, you'll probably not find a billionaire with kids at any daycare, their hired childcare workers won't be shared with other people, and will probably be adequately compensated up front in a way which anticipates a fair degree of schedule variability.)
OTOH, with red light cameras, you also don't need to scale the fine to work with both, because the entire purpose is to bind the lower classes while exempting the upper from any substantial burden. (The least cynical explanation is that it is to discourage behavior which might incur liability grossly exceeding the mandatory level of insurance company by those least able to cover the cost of that liability, thereby avoiding uncompensated harms, but the realistic explanation is...not so generous.)
Something I heard from someone who worked at the Palo Alto Apple Store two decades ago:
Steve Jobs's kids drew on an eMac with markers or something. He made them make an appoint for the store's Genius Bar and wait in line to have it looked at like everyone else. I don't know anything about how the staff tried to clean it or the outcome.
X% of net worth is still a bigger deal to someone with a net worth of $20 than to someone with a net worth of $20M, even though the latter may get some sticker shock. And it's possible (if rare) to have a reasonably middle-class lifestyle and an actually negative net worth. Presumably you would not make it possible to pay off student loans by repeatedly violating a red light, although it would be very funny.
You’d be surprised at the numbers of people living middle-class lifestyles with negative net worth. Credit card debt, car loans (with a too-small down payment, a car purchase can easily cause one’s net worth to decline the second you take delivery on the car), underwater mortgages, not to mention student loans.
But that makes wildly different incentives to enforce, depending on the target. We all know this stuff is all about revenue enhancement, and in that capacity, the targets will become the whales.
Or how about the curb weight of the car? Higher mass means you're doing a lot more damage in an accident. People might think twice about buying an F250 for their grocery getter.
I mean as a much greater "study", look at the UK - government introduced fines for parents of kids missing school, and the rate of absenced increased - because parents see it more as a cost that you just have to pay to go on holiday during school year.
I get your point, but I doubt the fine could have been ethically higher. Domino's drivers killed dozens of people in speed-related accidents before they ended their 30-minute guarantee.
I don't think our society is ready for the combination of automatic enforcement and truly punitive penalties. We readily demonize the accused; just having your mugshot taken can end your employability. Yet many of us break laws daily -- speeding, jaywalking, watering the lawn during the day, even plugging in a microwave oven without a building permit in some jurisdictions -- and society still works because we don't expect much enforcement. We are heading toward a future where everyone will have marks on their permanent record, but today our society tut-tuts, or much worse, at anyone who does.
"Domino’s confirmed it knew of 20 people who died in crashes involving its drivers in 1988 (the National Safe Workplace Institute would later claim Domino’s delivery drivers had about the same death rate as miners, who had a fatality rate of ~35 per 100k)."
It is understandable that someone who only lived in the United States or a low-enforcement place would have this world view. I'm more sanguine about the trajectory of our society.
Australia, Netherlands, Germany, Austria, etc, have well-enforced traffic laws. Speeding is the exception rather than the rule, unlike the United States, where one can expect the flow of traffic to be 10-20 miles per hour over the posted limit. Yet these societies don't suffer from an excess of enforcement or consequences in other areas. For example, it is legal to walk around in public with a bottle of beer in virtually all of Europe.
What we have seen in the United States is a reduction of many hardly-enforced laws. Jaywalking and minor drug possession have been decriminalized in several US states. This is due to voter interest. It will continue to be up to the public to decide what do to when enforcement can catch up to excessive laws.
In the Netherlands fines are insane. If you pick up your garbage bin two hours too late from the street corner, you pay 210 euros.
On the other side, if you sell drugs to kids and have a weapon, nobody touches you.
Yes, you can sometimes walk around legally with a beer bottle here.
I don't think you did get my point (my fault perhaps), as my only point is that if you take away from the Haifa study that fines will automatically increase the prevalence the targeted behavior in all situations, that's an insane conclusion to draw. There are lots of variables at play: the size of the fine, how consistently and strictly it is enforced, the ability of the finer to collect the fine, the social context, and so on. The Haifa study examines none of these. It does does highlight an interesting phenomenon, but without further studies that control for these variables, I don't think we can just blindly assume that the outcome in the Haifa day cares will apply to all situations where a fine is levied.
I see all the time on the Internet (and even IRL once) people make claims like, "oh, carbon taxes will just increase CO2 output, you know like in that Israeli daycare study." Drives me nuts.
Are fines the best possible solution to this particularly traffic problem? I have no idea. I'm not an expert in this area. But I am highly confident that whatever relation it has to the Haifa daycare study is so incredibly tenuous that it is not worth mentioning.
We might be talking past each other. Call Haifa a parable, if you will. I understand why you find fault with the study, but I invoked it to call out (quoting myself) that a fine can also give permission for unwanted behavior. That point adds to yours and doesn't contradict it.
The reason I said anything in the first place is that I object to automatically administered punishment. Either separately can be OK. Automatically administered? No problem, that's called a tax (including use taxes like tolls). Punishment? Then we'd better have due process, and yeah, it's going to be expensive and labor-intensive to administer, but that's critical in a free country. That's why I called out the "is better than" quote. I think it's strictly worse.
Sure. Then the bill requires that all those fines you pay go towards street calming infrastructure, eventually making it physically impossible (or at least very uncomfortable) for you to continue speeding.
Kind of like if enough parents paid the late pickup fee, eventually the daycare could afford a van for dropoffs.
It seems like this rarely happens. The fines become another stream of income, and reliance on that income kills any incentive to fully eliminate the behavior the fines are ostensibly meant to discourage.
Random thought: this also accurately describes the financialization of home ownership. It was supposed to provide stability in shelter, and instead created a market that's completely unaffordable to the prime home-buying generations, in favor of protecting those who've come to depend on unconscionable valuations.
No one, gun to their head and hand on a Bible, should defend a status quo where the only way to afford a median house is to have twice the median income.
Given the many restrictions on how the income can be used in this bill, I find it unlikely that will apply here. Feel free to check back in at the end of the pilot.
There are multiple well-researched and practical interventions that can be done to make driving safer for drivers, pedestrians, and cyclists. Implementation in the US is regularly scuttled by insane self-styled experts who "audacity" their way to public trust and influence, inexplicably.
I've seen quite a few roundabout projects over the years that seemed to work out well... Although I really detest any that have multiple concentric lanes.
I grew up in Fremont, CA, which pioneered the use of red-light cameras and terrible red-light camera practices (e.g. shortening yellow light times to increase revenue and giving a cut of the fines to the companies installing cameras). I hated cameras, the idea of speed cameras felt like big brother, and the basic principal of attributing a violation to a car and not a person (and thus requiring a person to rat out the driver) felt like a huge civil rights issue.
I then moved to Amsterdam and became the biggest fan of continuous, always-on ANPR speed cameras. On some freeways, your car is recorded at certain checkpoints and EVERYONE driving over the speed limit ALWAYS gets a fine.
Why? Because they are properly implemented (only high-risk areas), very well communicated (tons of signage), consistently applied (no crying your way out of a ticket, no racial profiling), purpose targeted (you get a speeding ticket, not a bunch of other fines at the whim of a cop), and correctly incentivized (ticket revenue does not immediately go to the local police or city).
The best thing about the average speed cameras is that between the checkpoints all cars drive at almost exactly the same speed. No one trying to overtake, just 5 lanes of traffic at 1km/h below the speed limit
I don't have a phrase to describe this concept. But it's when we blame one thing for a problem caused by another. Because placing the blame where is belongs is inconvenient.
You create a transport system where you're mixing incompatible modes of transport and carnage is what you get. What I see is instead of placing the blame on that everyone wants to place the blame on someone anyone else based on whatever moral anxiety they have.
Consider some people are drunks, you create a system where they have to drive you get drunk drivers. Don't want a bar walking distance from your quiet suburban neighborhood, again drunk drivers. Zoning that separates businesses and stores from low density housing. Driving, accidents. Mix pedestrian, bicycle, bus and car traffic, you get carnage.
No one really wants to blame the system and spend money to fix it. So blaming other individuals is what we do.
How about you prove your claim by providing some evidence that the richer you are the more speeding tickets you get. Otherwise you're solving for a problem that doesn’t exist.
I read the grandparent comment's point as being about suggesting %-based fines.
> prove your claim by providing some evidence that the richer you are the more speeding tickets you get.
How/where did the grandparent comment claim that the rich get more speeding tickets? Even if the rich speed at a lower rate, would that make %-based fines a negative improvement?
> a problem that doesn’t exist
My assumption was the speeding is a problem no matter whether rich or poor, and that both exist. Is there disagreement there?
Instead, I think their point was that even a $100 fine for a poor person may impact their ability to pay for groceries, childcare, etc, whereas for someone who has $10 million, etc., even a $1,000 fine will not impact their ability to pay for groceries, childcare, etc as they still have $9,999,000.
Expensive cars tend to accelerate faster, and it can be vastly harder to feel the speed. It would be unsurprising if up until some limit there was a correlation between wealth and the frequency of getting speeding tickets.
> One explanation is the fine turned a complex moral obligation into an ordinary financial transaction.
Income proportional fines solve the problem. The fine should be a deterrent, but not create crippling debt. That is impossible without taking into account the income of the infractor.
This isn't legally "tax evasion", it's well-established law: corporations have some degree of personhood, and as a corporate officer, I am allowed to disburse payments to those in my employ as I deem fit. Individual states have very little authority to look into foreign (other state) corporations.
Good luck contending with stare decisis and all of the implied interstate commerce issues to try and prevent this.
> It can also give permission for unwanted behavior. Cf. the Haifa study, where the rate of late pickups increased when daycares added a fine. One explanation is the fine turned a complex moral obligation into an ordinary financial transaction.
You know what they say: a fine makes something legal for an amount of money.
And for daycares, I think a lot of parents saw it just like that: a cheap way to keep the kid away from them for longer.
Do you think it's time for version numbers in filenames? Or at least a sha256sum of the merged files when they're big enough to require splitting?
Even with gigabit fiber, it still takes a long time to download model files, and I usually merge split files and toss the parts when I'm done. So by the time I have a full model, I've often lost track of exactly when I downloaded it, so I can't tell whether I have the latest. For non-split models, I can compare the sha256sum on HF, but not for split ones I've already merged. That's why I think we could use version numbers.
Thanks! Oh we do split if over 50GB - do you mean also split on 50GB shards? HuggingFace XET has an interesting feature where each file is divided into blocks, so it'll do a SHA256 on each block, and only update blocks
That might be the answer -- something like BitTorrent that updates only the parts that need updating.
But I do think I'm identifying an unmet need. Qwen3.5-122B-A10B-BF16.gguf, for example: what's its sha256sum? I don't think the HF UI will tell you. I can only download the shards, verify each shard's sha256sum (which the HF UI does provide), llama-gguf-split --merge them, and then sha256sum the merged file myself. But I can't independently confirm that final sha256sum from any other source I trust.
reply