jmb99

Have you ever been to India? In particular Delhi and Gurugram, and the surrounding manufacturing/industrial areas.

Probably the same way they have for the last 30 years: poorly, in a dozen different ways, depending on what that particular hiring manager thinks has correlated with successful hiring in the past

I’m a big proponent of writing simple, understandable code, so I’m playing devil’s advocate here a bit, but: who cares?

A significant (majority?) portion of developers have been shipping JavaScript/node applications for the last decade that contain hundreds of MB to GB of code from god knows where doing god knows what with dependency trees the size of redwoods. It’s not like your average mediocre dev really knew what was going on behind their gluing of frameworks together - at least from what I’ve seen.

If you have remotely competent tech leadership that enforces relatively intelligent patterns (a good one I’ve found is “write everything backend in rust”) you can make AI churn out monstrous amounts of code that… isn’t all that bad? And if you enforce it writing and updating a docs/API.md on every commit/PR you’re probably doing better than 80+% of devs I’ve ever met. Up until a few years ago it wasn’t uncommon to roll up to a new job that was a “legacy” pile of garbage concocted over 20+ years with no comments or API docs and a readme that tells you to ask for help from someone who has been dead for 5 years. At least AI code is full of comments (some of which might even be accurate) and there’s a finite (relatively low!) cost to figuring out “wtf is this doing and how is it doing it”

By the time my plan’s done I’m usually between 200 and 350k context. Even if keeping that around gives a performance bump for the implementation (which I haven’t noticed to be the case) it balloons the cost. I would much rather put everything in a plan file and start fresh.

Plus, I usually plan with a more expensive model and guide implementation with a cheaper model (with smaller validation calls back to a more expensive model)

Back of the napkin math says a bare minimum of ~2000kg of solid fuel in three stages to “get to” GEO using a 3-stage rocket, to deliver 1kg of “something.” You’d need a gravity assist highly elliptical orbit which is a touch harder to pull off (but probably not harder than getting 2000kg of ammonium perchlorate). Plus building the actual rocket, guidance computer, etc.

Not trivial but not impossible. If you’ve got a few hundred million people with nothing better to do, I’m sure a few dozen/hundred could pull it off.

Well, the alternatives are:

> It would be really cool if someone built an extension to show GitHub status live

"These already exist, why wouldn't you search before posting?"

> There exist quite a few extensions to show live GitHub status

"Why would you not post them if you know they exist?"

Or, what actually happened:

> Here are some extensions that might work for you to show live GitHub status

"Why would you recommend extensions to do this?"

Seems to me like, if your goal is to pick apart someone suggesting something, you'll find a way to.

My uneducated guess says, not much. For running massive models you still need a ton of high-bandwidth interconnects between many individual chips/GPUs/etc since you need to do math across a few TB worth of weights. That's simply going to require more power (and more die area in I/O, and therefore more cost). Being able to run small models in tiny power envelopes is incredibly useful to people, but I believe it will be covering a different niche than what datacenters can provide. Likewise, you'll still need crazy amounts of high-end memory to populate whatever goes in these datacenters.

The only thing that will crash prices is reduced demand (duh) or, more interestingly, increased production. In particular, if CXMT is able to get their DDR5 fabs up to a reasonably high yield, that could add some downward price pressure (as could government subsidies). As well, if Micron/Kingston/Hynix think that CXMT is going to start cutting into their market share, they might be willing to either increases supply or drop prices. Unfortunately CXMT looks to be taking quite a while to get their new fab up to max capacity so that may take a year+ before anything manifests.

If you're interested in following the (publicly available) info on these sorts of things, check out what companies like Axelera, DeepX, and MemoryX are doing today and have on their roadmaps, as well as the sorts of chips/SoCs Qualcomm, Kinara (now NXP), and Ambarella currently have announced (or have on the market). And remember, that pretty much all of these chips on the market today were in initial development more or less when ChatGPT first launched. If you knew what you knew today (or a year ago) about what requirements current- and next-generation models would have (from a silicon perspective), what might you do differently? Think for instance, host system interconnects, amount and speed of on-package or on-die memory, image/video decode capabilities, int8 vs fp8 vs fp16 vs bf16 compute units, etc. And, consider that most "AI" stuff in development a few years ago was all 15nm or 12nm - because who was gonna pay big money to get fab capacity at 3nm to run some object detection models? So most of the stuff on the market today is on very old nodes and therefore not super power efficient.

I'm not who you responded to and I don't have any info on Google. Nor can I explain in detail due to NDAs. But multiple major players are working on something along the lines of what the parent is alluding to.

The "edge" AI landscape (in particular, what you can do with ~5W) is going to be nuts in about 18 months.

How is “all RAM sold in consumer PCs must be registered ECC” equivalent to age verification laws? I’m not saying I agree with the premise but that is a bit of a leap

> ECC DIMMs are available only for the memory speeds guaranteed to work by the manufacturers of CPUs, like Intel and AMD, e.g. for up to DDR5-5600 in UDIMMs or DDR5-6400 in RDIMMs.

This is not true[1]. These DIMMs are not guaranteed to work with all Threadripper CPUs, but they are guaranteed themselves to be stable at their advertised frequencies, which are higher than JEDEC. (And from my experience, 9000-series threadripper is perfectly happy to drive 8-channel 1Rx8 at 7200MT/s.)

AMD (and possibly Intel, I don’t have any experience with their recent offerings) also lets you overclock memory completely on WRX90. So if you buy JEDEC sticks, you’re welcome to try to push them as hard as you can.

[1] https://v-color.net/products/ddr5-ocrdimm-amd-wrx90-workstat...

There are absolutely fast DDR5 ECC DIMMs[1]. Just break out your wallet (and piggy bank, and rainy day fund, and..)

The main reason there aren’t more options here is that a) high-frequency memory usually isn’t sufficiently stable, unless binned, and b) there’s very, very, very little demand.

[1] https://v-color.net/products/ddr5-ocrdimm-amd-wrx90-workstat...

I’ve been running a Supermicro X10SLH-LN6TF as my router, and it’s very well suited for the job. Power use is very high for a home router, but reasonable for a 1U router - especially considering the era it’s from and its capabilities (routing 5-8Gbps per stream depending on firewall config, able to handle a couple active streams, and line rate routing with no firewall).

I wouldn’t suggest anyone buy something like this today for home use, but 6 years ago when I purchased it there were not many 10GbE routers to be had period, let alone for the price ($300 roughly if I remember correctly ). Even considering the cost of power over the last 6 years, I still come out much cheaper than if I’d bought something more power efficient new at the time.

Do you think it’s legal to shoot a firework at someone’s house?

> O2 sensor in cars compare the O2 concentration between the outside air and exhaust gases, it needs outside air as a reference

Source?

In ~300k km worth of diy oil changes, I’ve yet to change a crush washer, and yet to have a drain plug leak.

I always replace them on friends’ Toyotas, because they seem more important, but on every car I’ve owned it hasn’t mattered. And if you take the least amount of effort to google “how to change oil on ________” (fill in the blank for your year, make, model), some forum or video will probably tell you exactly what steps to take, including whether or not changing a washer is necessary.

Changing oil requires

> a place where you have sufficient access to the vehicle to drain it

Probably the only valid argument for people who park on the street.

> the right equipment

One $5 wrench, one $10 filter wrench (optional). One set of ramps ($40), or jack stands ($30) if you already have a jack. One drain pan, $10 (or free if you're resourceful). Total cost max $65. Cheaper if you look for deals, buy used, borrow from a friend. If you can't afford $65 once to save money in the long run while owning a car, you probably should've bought a cheaper car.

> the right disposal solutions

Every oil change requires a jug of oil to be purchased. You can drain your used oil into this jug and then dispose of it along with your other household hazardous waste. This is not hard.

> Most people who have cars do not have that.

I might believe this for a place to do an oil change, maybe. I struggle to believe most, but I would believe many. Aside from that, if you don't have those things, you are choosing not to have them.

Which is kind of the point. None of these things are hard, at all. The majority of car owners 100 years ago could adjust their own timing, clean distributor points, replace belts, etc. because if they couldn't, they'd be calling for a tow truck every few hundred miles. Those are all harder, and things have only gotten easier with time. If you can't do them, you are choosing not to, because there's an even easier solution - spending more money and getting someone else to do it for you.

I am typing this on my return flight home from a business trip in Delhi. There are many other areas the Indian government needs to be focussing on first.

I had a similar view to you ~2 weeks ago. Spending some time there very quickly made me realize that there’s a lot of other things that are much more pressing.

I’ve tested this recently (this post week). Had a dns entry up and pointing to an nginx server for ~12 hours, zero requests. 17 seconds after the letsencrypt cert was issued, the floodgates opened. Over a dozen of requests per second.

I agree. I ran mid 16s in 8th grade, and was in the 14s in high school, with the only training being whatever we did in gym class. But I do also look at the sheer number of overweight kids these days and figured, well maybe mid-16s is actually a reasonable average point.

> Very few mere mortals could run that fast for even 100m.

That works out to roughly a 16.7-second 100m. While certainly not crawling, that would be a fairly average pace for a fairly fit middle- to early-high-schooler with a bit of practice.

Yes that’s insane to maintain for a marathon, but it’s not even remotely out of reach for 100m for most relatively-fit people at some point in their lives.

I’ve mentioned this as an option in other discussions, but if you don’t care that much about tok/sec, 4x Xeon E7-8890 v4s with 1TB of DDR3 in a supermicro X10QBi will run a 397b model for <$2k (probably closer to $1500). Power use is pretty high per token but the entry price cannot be beat.

Full (non-quantized, non-distilled) DeepSeek runs at 1-2 tok/sec. A model half the size would probably be a little faster. This is also only with the basic NUMA functionality that was in llama.cpp a few months ago, I know they’ve added more interesting distribution mechanisms recently that I haven’t had a chance to test yet.

I stopped getting scared of `if` and `of` about a decade ago when I started explicitly saying (in my head) "input file" and "output file" rather than "if" and "of." You still can mess up the order, but imo no more easily than you can swap `cat in > out` for `cat out > in`.

> Friends don't let friends use `dd` where `cat` can do the same job.

Technically yes... but I like being able to explicitly set block sizes and force sync writes.

The linked post currently has demonstrated that 6 years in some reasonable-ish condition is perfectly fine.

> they might not last one year.

> Definitely not a medium to passively store anything long term without power!

Do you have any evidence to back up this claim? I'm much more interested in data than fear mongering.

You can however set the block size to something quite large, which means you write the same random pattern spread out over multiple blocks repeatedly. If you pick an "odd" block size (like say, your native block size multiplied by 47), it's highly unlikely your disk under test will be swapping around "groups of 47 blocks." (I usually just do a nice multiple, like 4K16, but if you're super paranoid a weird multiple should be pretty much good enough). You won't get reporting of which exact* blocks on the drive are failing, but these days, that isn't really useful information - if any blocks are failing, warranty or ditch the drive.

Yes, that’s right, don’t buy any new car, any phone, any television. Hell don’t buy any x86 laptop or desktop computer, since you can’t disable out replace Intel ME/etc.

How is that better?

How does SSH become an arbitrary user without effective root?

How on god’s green earth is `sudo apt install telnet` sufficiently challenging to be a driving factor to creating your own distro??

> I have a 21 yo car and a 12 yo car and will eventually have to get something 'modern' (worse) that forces spyware/subscriptions on me just to get from point at to point b

I daily a 30 year old car. There exists a sweet spot of reliability, safety, and comfort (probably the early-mid 2000s) that in theory, you should never have to buy a vehicle outside of, newer or older. There will always be clean old cars in good shape you can buy, you don't need a new vehicle.

Unless you can't buy gasoline anymore. But that's still quite a long ways away imo.

> He finally found a safe spot and successfully pitted the car to a stop.

No such thing as a safe spot to PIT someone, ever, let alone while they're asleep at the wheel. This is a great example of why people hate all cops, anyone with two brain cells to rub together would get in front of the car and gradually slow to a stop.