So a business that is entirely dependent on ARM IP and, for the most part, Android should "remind" the company they're dependent upon? Let's do a thought experiment - Qualcomm switches to RISC-V while the other Android SoC makers (MediaTek, Samsung, Google, Xiaomi, etc.) stay on ARM. Who buys the new Snapdragon RISC-V phone?
They can design chips. What Qualcomm might not be able to do is deliver solid software support for their riscv64 hardware. Android decelerated their efforts toward support recently.
The network experience on Nintendo devices always seemed janky and home-grown. I feel like they built everything from scratch at corp HQ complete with wonky edge cases.
Lots of companies need skills they don't know how to ask for. So they advertise for the wrong things. For example a medium size real estate firm probably needs someone to deal with the computers. Its a grind but cold reach out can go far
this seems like bad example.. a commercial company in the US almost always has Windows.. Windows is a merry-go-round of virus and intrusive, debilitating updates. Professional consultants in every big city exist to do nothing but get paid to babysit Windows. Why spend money on training a junior that may leave, instead of hiring the consultants that know how to babysit Windows? .. Secondly, real estate is close to law and banking, where there is a culture of semi-arbitrary seniority.. juniors do as they are told by their managers or else you are replaced. It is a situation that most adults quickly try to leave, but there are infinite numbers of new semi-desperate people to replace the juniors, as there have been for the last several centuries in those professions.
For those two reasons, a real estate office is a bad example of the dynamics of hiring junior (high skill) CS graduates IMHO
I agree that companies do a poor job of advertising and screening for skills that they really need. Guessing, it might have to do with HR or worse, bad-intention managers, copying the habits of other hiring practices blindly.
It's one of these areas where people (including medical professionals) hold strong beliefs, but then it turns out that there are other highly-developed countries where this is not routinely practiced, and the outcomes aren't necessarily different.
Routine wisdom teeth removal is not a thing in most of Europe. Another random example are colonoscopies and routine flu vaccines (except for the elderly).
I've generally assumed the simplest explanation is that many of these weakly-supported procedures are regular, consistent income streams for the people who perform them in the US: my four wisdom teeth (that were causing me serious issues at age 19) cost $2k to remove nearly 20 years ago, and I know colonoscopies are billed to insurance in the thousands. There's not much incentive to move to cheaper tests or wait-and-see, when you can just do it to everyone who reaches a certain age by default.
Presumably flu shots are good business for the manufacturers, though I'm not sure about the science. After having the flu as a healthy late-twenty-something a while ago, which was...intensely horrible, I've chosen to get it ever since.
It's not always true though. My dentist in Europe pushed me to get my wisdom teeth removed early "because you're going to need to get them removed eventually anyway" at a government clinic with no profit incentive.
Not sure which country you're talking about, but in Germany, the public health insurances cheerfully pay for annual flu and COVID boosters for everyone.
Flu may not be too dangerous in people who aren't elderly but it still sucks. Can a non-elderly, not in any other high risk group, person get seasonal flu vaccination in Europe if they ask for it, and is it covered by European health care systems?
They're not only covered by the gesetzliche Krankenkassen (German public health insurance providers, mandatory if you don't meet the criteria for private insurance), mine gives me a little bonus if I prove I've gotten various vaccines, including annual COVID booster and flu.
This guy infamously has a problem with each and every HN user and chooses to display an NSFW image saying such if your request's referrer header has news.ycombinator.com. Don't click.
If they struggle with first stage re-usability for a while, that merely adds cost. Further they probably want to iterate and scale anyways, so in the short term they're gonna be building a lot of first stages anyways.
Second stage reentry is necessary for this thing to ever carry people, ostensibly the mission the ship was designed for. It is a hard requirement.
That's not good enough - everything attached to the computer (including network gear, unless it's fiber) must be powered by the same UPS, or you have pathways for lightning current to enter.
Some UPS for consumers even have an ethernet through-port so you can connect your internet cable from the wall to the UPS before it touches your router.
That's functional isolation. Yeah, they're individually tested to 1500 Vrms (or so the factory in china claims), but that insulation is still just the very thin lacquer of the windings and maybe lacquer on the core. There are more expensive (and physically much larger) Ethernet transformers for medical devices and such, these have actual double/reinforced insulation and are tested to much higher voltages.
Typically the Ethernet shield is also only alibi-insulated from the device's ground. Sometimes not at all. (Personally it's always funny to me to see the fat 1kV/1nF capacitor from the shield to the device ground and then the metal of the socket is just bunched up with basically no clearance against the metal case or something like that). This can of course be avoided by simply using unshielded cable.
A online UPS has in line: a rectifier with its transformer and output-smoothing capacitors, a huuuge battery, surge capacitors, power transistors, another transformer and more capacitors. And only then comes the actual device with its PSU.
It takes an awful lot of power to smash through these multiple layers of insulation and bypass all the capacitors, at least if the PSU is properly designed (specifically, clearance between the various power / ground domains).
In contrast, Ethernet transformers are tiny small little things.
Yes, but only the transformers will provide the galvanic isolation. And if the windings are not physically separated then you are basically in the same situation.
I'm in the business of mi300x. This comment nails it.
In general, the $2 GPUs are either PE venture losing money, long contracts, huge quantities, pcie, slow (<400G) networking, or some other limitation, like unreliable uptime on some bitcoin miner that decided to pivot into the GPU space and has zero experience on how to run these more complicated systems.
Basically, all the things that if you decide to build and risk your business on these sorts of providers, you "get what you pay for".
Distributed training data creating & curation is more useful and feasible. Training gets cheaper 1.5x every year, but data is just as expensive, if not more, given that the era of "free web crawls of human knowledge" is over.
I agree with you, but as the article mentioned, if you need to finetune a small/medium model you really don't need clusters. Getting a whole server with 8/16x H100s is more than enough. And I also believe with the article when it states that most companies are finetuning some version of llama/open-weights models today.
