New GPUs are quite a bit better than previous ones but perf oer dollar has been flat for a while now.
Also, if you're talking gaming GPUs old ones work fine given there has not been a new PlayStation or Xbox in many years. Min spec for many games is 5 years old tech
Yep. The whole fake frames thing seems like a witch hunt to me. If you can actually get a 5090 and were previously fully utilising a 3090 or 4090, it is a significant upgrade, as long as it doesn't burn your house down.
The generated frames thing is interesting as it's an obvious example of the whole discussion around halo products (the x080 or x090 cards), "the rich get richer", it takes up a lot of air in the room but is only appropriate for a small vocal proportion of the market.
As you say you need a good framerate to start with before generated frames make sense so either you're already running well with a high tier card and are further able to show off the combination of framerate/resolution/detail level, or running a lower tier card at lower settings turning the graphics down further which can be a very obvious trade off. Less demanding games which are generally online competitive usually wouldn't do as well with any extra latency introduced, and the situation where I've heard it would be a good fit is emulators or any kind of game where the CPU is the limiting factor
I just don't think the 5090 is for gamers; it's for hobbyist AI people who want to buy multiple to run high parameter LLMs. Fake frames are irrelevant to this crowd, even if they take one of their 5090s aside to game once in a while.
Sure they are better... but once you factor for die size, core count, wattage and so on... the improvements being made are less impressive than they first seem.
My alarm clock (old school radio alarm clock with big red digits) is plugged in and resets to 00:00 when there's no power. I would notice even the smallest outage this way (unless it happens exactly at midnight and lasts less than a minute).
That has happened maybe twice over the last 16 years. In both cases the outages were a few minutes at best. I live in Berlin; power outages are extremely rare here.
That 13.7 minutes is not for all house holds. In the rare case an outage happen, the very limited amount of households affected by that would experience that duration on average. And lets face it if something knocks out a major power line, which does happen, that single incident might take a few hours to resolve and would probably drag that average way up. Which means that I expect the median outage duration is probably a lot lower than the average; in the order of seconds or minutes at best.
"The number of interruptions per customer in 2023 was 0.34, which means that each customer is only affected by a disruption once every three years on average."
That sounds more like it. I generally only adjust the time on my alarm clock when daylight saving requires me to; twice a year.
Yes, super rare and short in Berlin. I'm originally from a rural area where overground high voltage lines between villages are common. There it's typically 1-2 hours of outage every 3-5 years or so. The local utility company has some truck-mounted generators to help out in smaller outages.
As an anectdotal data point, I live in Poland (Warsaw) and I can't remember the last time I had a power interruption at my home. I don't use UPSs at all, and my NAS gets years of uptime, unless I decide to reboot it.
This gets much worse if you live in the countryside, for obvious reasons, and I would guess that the German average is mostly driven by countryside, not big cities.
I had similar experience in the UK. Maybe 3-4 outages in the first nearly 30 years of my life. Moved to the states and even living in the outskirts of Seattle I used to get multiple outages in a year.
The funny thing is, every time people from not-the-states talk about how rare power outages are, americans feel this bizarre urge to defend their power companies and grids, coming up with incredibly contortions to explain why it's not even remotely possible to do power the same in the states as elsewhere in the world. One memorable conversation here on HN ended up with the poster, facing the fact that yes, even in countries with lower population density still manage to bury their power cables (because they were claiming people were too far apart), somehow decided that it was because the states didn't have the expertise or equipment for burying power cables. Apparently no one here has diggers, and things like sewage pipes and gas pipes just run over the surface.
When I lived in the Czech countryside, we used to have about one outage per year, and it was almost always a planned one, when the grid required maintenance or expansion/upgrade (new houses being built).
In the city, there is something like one outage per 4 years, usually due to an extreme thunderstorm or floods. And it usually lasts under 20 minutes.
Reliability of the grid is a major indicator of infrastructure quality and I am somewhat surprised by the fact that Americans tolerate so many outages and consider them somehow natural.
The typical German experience could be described as "every decades your street loses power for two hours at a time" (which comes out to about that average). As you say, it's worse in rural areas and better in urban areas, though maybe less extreme than in Poland. But power interruptions are typically short and highly localized. Nobody would even think of getting a generator in case the power goes out, and outside a server room nobody has an UPS
Anecdotally, everyone I know in the US who has a generator didn’t get it because of random power outages, but because of natural disasters such as hurricanes, tornadoes, earthquakes, and blizzards.
At least in my area (NJ) we lose power around a week on a good year, approx 2 weeks out on a bad year. Outages usually caused by trees falling during thunderstorm/wind, sometimes snow, not hurricane. Oh and drunks slamming into poles. Everyone has generators or other forms of backup power in my area.
But problem in recent years was the dead ash trees from emerald ash borer, they took a few years to fully die and weaken but now they just collapse into lines on a good wind gust.
Localities that have their own utilities never go down (Madison, NJ), my area is covered by Firstenergy that cheaps out and keeps only a skeleton crew in NJ, so when stuff like the poles snap at end of my driveway it takes 3-4 days to send crews from texas or ohio to replace it.
> At least in my area (NJ) we lose power around a week on a good year, approx 2 weeks out on a bad year. Outages usually caused by trees falling during thunderstorm/wind, sometimes snow, not hurricane. Oh and drunks slamming into poles. Everyone has generators or other forms of backup power in my area.
I had no idea your basic infrastructure was so bad.
Why not demand buried cables? Trees can't fall on them, drunk drivers can't knock them down.
On the other hand, here in Europe I've never had any problems remotely so common and severe, not in any place I lived, even with overhead power lines, including the tiny remote (by local standards) Welsh hamlet: https://maps.app.goo.gl/AtGFM9C5xJ6GrMJz5?g_st=com.google.ma...
I'm not an American, but Australia has a comparable number of power outages. A big factor is cost, and population density is what causes cost to be a concern. The EU has a population density that's three times higher than the US; underground lines are three times more expensive than overhead.
Australia is of course ten times less dense than the US, comparable to Idaho, but we have a unique combination of moderately dense land in the east and south west plus shockingly sparse areas with little to no infrastructure at all.
Likewise Australia, the outback is about 70%-85%-ish of the country depending who I ask, with about 607k people in it, so if the Aboriginal Australians were to secede with it, the rest suddenly gets x3-x6 the population density with no actual change to their experiences.
I think what would be useful is a histogram of population densities, or for example a percentile based metric: population density covering 99/95% of population.
I'm thinking something like "From the average person's perspective, how many people are within 100m, 500m, 1km, 5km, 10km, and 50km?"
The questions that seem to matter, and for which population density is a proxy, are: how crowded does it feel, and how easy is it to get to the socioeconomic advantages of being near other people (infrastructure, commuting, shopping, community sports, faith buildings, pubs, etc.)?
Those sorts of natural disasters are happening over once a year in many metropolitan areas these days, including places where they used to be incredibly rare.
Ah, but a power outage is kind of fun. It is like a little bit of survivalism, but you know the power will be back fairly soon. There’s a whole ritual, fill the tub, get some candles, figure out how to keep the food cold.
The links actually cover this, since EIA tracks major events in power disruptions and separates them in the graph. US network is still orders of magnitude worse than Germany.
If I read the provided source correctly, if you exclude those events the average US customer still has two hours of power outages per year. That's a lot better than five, but still nearly an order of magnitude worse than 13 minutes.
Atlanta, GA has underground power transmission through much of the city (and specifically my neighborhood within the city) and we can still lose power in storms. What isn’t underground currently is being moved there in much of the city. However, the transmission equipment is still outside and still has physical limits. There’s only so much water, so much wind, and so much lightning electrical systems can take without someone going dark. Underground trunks can also flood and short so they’re not a full panacea in areas that have heavy storms.
We’ve had a mild storm season the past two years and we’ve maybe been down seconds in total in my area.
While the average I’m sure is correct, the distribution is going to depend a lot on what is going on with nature. When I lived in California I lived in an LADWP area so I didn’t experience rolling blackouts. As a kid I had friends that would come over because they lived in an Edison area and play time in air conditioning at my house was much more enjoyable in August. If you were to find the average downtime during that period of time I expect it would saddle everyone with 10-15 minutes of outages even though my area never went down and my friends lost power for a few hours each week during heat waves.
I wish that was the case. Coincidentally, I live in Atlanta, also within the city (ITP, close to Piedmont Park) and all my neighborhood has above ground power transmission. Coming from Europe, it's infuriating. Not a year passes that we don't get two or three blackouts and a fair decent number of brownouts. That includes the past two years.
Any time the Santa Ana winds pick up in so cal they turn the power off to prevent fires now so yeah the US power infrastructure is shit compared to Germany. Over there it’s all underground because when they rebuilt after ww2 they realized it’s dumb to have it be so easy to blow up your power grid with bombs.
Much harder to hit though, except with modern precision bombs if you happen to know where they are.
That said, I also doubt that bomb resistance was the main reason for underground power lines. The sentiment in Germany at the time was, and mostly still is today: "No more wars!"
If you have any residential overhead distribution, that’s gonna drag you down compared to everything buried. It still counts as downtime of a tree did it.
I am so happy I am not this jaded after over 20 years in software. I have never felt like I have hit a ceiling in terms of salary or skill. I still learn each and every day. I would get bored if this were not the case. Would I recommend it to everyone? Hell no but I wouldn't choose anything else.
Has your compensation been up much inflation adjusted each year or even three years over that time span?
Yes, you can keep learning and you must to stay competitive. But most will plateau if they are aggressive over the first decade especially if they don’t jump to the FAANG side of comp.
As someone that spent quite a bit of time with table-transformers, I would definitely not recommend it. It was one of the first libraries we added for parsing tables into our chunking library [1] and the results were very underwhelming. This was a while back and at this point, it's just so much easier to use an LLM end to end for parsing docs (Gemini Flash can parse 20k pages per dollar) and I'm wary of any approach that stitches together different models.
It isn't like US public system, medicare, is great. I still end up buying my folks supplemental. This narrative of pubic vs private misses most of the nuance.
there's so much nuance in an insurance company having billions left over every year after subtracting payouts from premiums! it's sooo complex and nuanced
I think a lot of people would consider alternatives if there were some. Sourcing AMD hardware is a PITA, even just renting for validation. On the consumer side, AMD announced they are done with top end GPUs for the time being.
I agree with you on marketing but Nvidia's relationships with the big cloud providers and big companies is more important. It locks AMD out.
> On the consumer side, AMD announced they are done with top end GPUs for the time being.
You reminded me of the many "let's hope that the next AMD GPU is super fast so Nvidia cuts the prices on their XXX GPU" I regularly see on hardware boards.
Haha, I think that hope is officially dead. Although, Nvidia hardly cares about graphics right now so AMD might release a good gaming card even if it isn't "the best". Rising prices on the "best" will mean gamers never have them anyway.
Surprisingly enough, several large systems on the top500 list use AMD (and even intel?!!) GPUs.
To me this indicates that AMD and intel have the potential to compete with Nvidia, but they haven't managed to translate that into the general AI/non-HPC market.
I have used some of those systems, namely frontier (AMD). In the case of frontier, it is the US gov hedge their bets. The super computer before was Nvidia and the next is intel. Bunch of politics there which don't translate to business but make a ton of sense for a government.
I am in Switzerland where you have 25Gbit/s for about 70$/month so I understand that this might be an exception. But even if it is 10 000$/month, it is still widely cheaper than cloud bandwidth.
If you don't mind me asking, to where? That is, what uplink do you see to your nearest AWS or gcloud? In the US, advertised residential speeds don't nessarly translate to realized gains. Just pushes the bottle neck to the ISP.
Agreed that either way it is way cheaper than cloud bandwidth.
Also, if you're talking gaming GPUs old ones work fine given there has not been a new PlayStation or Xbox in many years. Min spec for many games is 5 years old tech