I've been ignoring the inflation talk on the assumption that it's the usual bullshit Republican fearmongering, or, at worst, a return to the mean after the past couple years of unusually low inflation. I Googled "inflation chart" and this was the first hit, so for what it's worth it's almost ex recto but not quite. Over a one-year or even five-year time span, fair enough, today's number looks bad. Before that, though? The current level was exceeded around 2008 or so, which is concerning because of the crisis around then, but then again we're just coming out of a crisis now. Before that, it was very volatile compared to more recent times.
Rental cars are horrible now, if they are available. They weren't in Omaha.
We've talked about the semiconductor shortage here before but I don't think it's really widely appreciated how significant, broad, and long-lasting its effects are going to be. The causes seem to have been a terrible combination of a big COVID-19 hole in the production chain, increases in consumption in China, hoarding by manufacturers due to uncertainty induced by Trump's trade wars, and the inexorably growing appetite of crypto miners.
More, and probably importantly, domestic production capacity will definitely help, but on the order of years. And unless regulatory agencies find a way to kill the crypo mining industries, they're going to be rationally outbidding other consumers for as much production as they can demand.
It's supply chains. Everyone's restocking, prices going up. (And "supply chain disruption" includes in the US swathes of businesses that laid off staff or went bankrupt, which can't help.)
3: AIUI the semi shortage is as much due to bad choices by customers (most notably auto manufacturers) who cancelled orders in early 2020 and now find themsleves at the back of the queue, due to massive competing demand from everything else: massive preexisting trends for data centers, network gear, 5G rollout, multiplied by WFH demand; with seasonal cellphone demand and now restocking demand piled on top. Part of those big trends is "high performance computing" -- how much of that is mining, versus ML and cloud expansion, I don't know, but I'd guess the latter are more important.
No, we didn't get our orders in in time. We're doing that with the booster shots.
Lumber prices rocketed up but have already dropped 40%; the prices are still quite high historically, but everyone is right that this looks like a combination of a demand shock plus supply chain disruptions plus a bunch of politically motivated attacks by people who don't see anything wrong with twenty years of deflation and disinflation because it helps capital at the literal expense of borrowers.
Are journalists assigning blame properly?
I have to assume this was sarcasm.
What Mossy reports in 5 is what I've read -- particularly in regards to car manufacturers canceling their orders as if the demand wasn't going to bounce back a year later. I think the stimulus caught them by surprise.
Trying to figure out how much mining matters in prices, so I did a bunch of back-of-the-envelope calculations, surely too many to actually draw a conclusion (and also some of it is obviously wrong). Get ready for some bad math.
Bitcoin network hashrate is order-of-magnitude 100 EH/s. (source). E = exa = 10^18. No other currency's network comes close in size, even the other valuable ones. Good GPUs for mining are about 25-60 MH/s (source). Assuming hashrate is additive (not sure why it wouldn't be?), that implies there are 2*10^12 GPU-equivalents being used to power bitcoin. That can't be possible so one of my assumptions is absurdly off or I can't divide while goofing off in a meeting.
Try it another way: Bitcoin uses 110 TWh/year (enh, google it, I'm sure the number will be out of date soon). GPUs pull 200-300W, let's say 250W. If they're running full out 24-7, that's 50 million (5*10^7) which is much more reasonable so I'll use that number instead. Now admittedly that's the total install base, not annual network growth; I don't know how often miners replace their cards. In 2017 the hashrate was a minuscule fraction of what it is now (source) but grew pretty rapidly after that point, so let's say it did that 50 million over 4.5 years. So that's about 10 million cards per year assuming no replacements (and hence is a lower bound).
In 2020 41.5 million discrete graphs cards were sold, so the total bitcoin hardware install is about one year of world GPU sales. That same page says the total market for such cards was $14.8 billion last year for an average sale of $350. However, that $350 is across all cards; market prices for popular mining GPUs have MSRPs closer to $700 (and $900+ on the secondary market).
If my wildass 10 million number is right, that would mean $7 billion in GPU sales a year for bitcoin mining. mining might actually be the majority of the GPU market.
Wikipedia has the total semiconductor industry at $481 billion, so GPUs are a small part of the total. I assume GPU fabrication equipment is pretty specialized and you can't just switch over something that produces e.g. low-energy embedded equipment or general-purpose CPUs over to make GPUs, although GPU sales might drive future capital investment choices. So my guess is that this is mostly just affecting GPUs. So: bad for gamers, bad for machine learning, probably not particularly related to car manufacturing.
I don't think Bitcoin mining hardware and GPUs are necessarily interchangeable. GPUs are used for Etherium but Bitcoin uses specialized chips that aren't really the same as GPUs. You could say there is some equivalence but its a stretch - the chips in bitcoin mining rigs can't be used anywhere but bitcoin mining rigs.
Probably the squeeze on GPUs isn't so much increasing demand for Etherium mining but rather the tendency for Etherium miners to burn through GPUs at an outrageous pace.
dalriata: I don't follow BTC or other crypto (b/c "knaves and fools") but I thought that BTC used custom ASICs? So not GPUs?
On the other hand, its all silicon, and there isn't that much of it.
12,13: Oh, my bad--I don't pay enough attention to crypto so I've probably confused Ethereum and Bitcoin chatter. What's the fab process for those ASICs like? It's still massive amounts of matrix math, so I assume they're much closer to GPUs than CPUs or anything else. Do GPU manufacturers use basically the same tools? If it's highly substitutable, that should still affect GPU supply if not demand.
Anyway, using the same numbers for Ethereum, which uses about 44 TWh of energy a year, that's more like $2-3 billion ignoring card replacement. Still a pretty significant fraction of the GPU market, although that's much smaller than what we're interested in.
A tree fell on our car a year ago, and now that the world is opening back up, we're finding that we need two cars -- just as used car prices have gone nuts.
At least you got the house done before that went nuts.
Kind of. We moved back in three months ago, but the contractor still hasn't finished the destroyed room (though it's close).
We also started renovations on two bathrooms a month before we moved back in. One of the bathrooms is almost done after a series of screwups by various players. We're optimistic that the second bathroom will be done promptly, though. The biggest problems were getting the furnishings -- which we did for both bathrooms simultaneously -- and getting our contractor to coordinate sub-contractors. (The plumber has to show up before the tiler, for instance.)
In 11 days it will have been a year since the tree came down.
You know, these guys are pretty careless with their truck keys.
On the other hand, I've got one of those combination-locks-with-a-key-inside things hanging on my front door so the contractors don't rely on us for access. If it came down to a theft contest, they could do some damage ...
TBC, I happily agree crypto is a significant resource suck, but I think other factors are much bigger, and that we're fast approaching a bitcoin collapse, for external reasons.
The PRC is coming down hard on coal-fired miners in their north and northwest, which is most of them, for the sake of emissions targets. Some of that is migrating toward cheap hydro power in the southwest; but at some quite near point that excess hydro is likely to be mopped up by general demand; and meanwhile the tech wars mean the country will continue to have limited or shrinking access to high-end fabrication, which Beijing will be increasingly unwilling to squander on mining-optimized silicon.
And across the ocean, Colonial might be a turning point, with the DoJ of all people hacking Darkside to recover the ransom. Bitcoin has made itself a US national security problem, and I don't see it surviving that kind of attention for long.
It's worth remembering that our other national security problems aren't exactly endangered.
Crypto is one big stupid scam, it's all just a silly con.
It's a really good con. The Q fuckers barely make any money.
Your other national security problems don't consist of giant data centers connected to and earning all their income on the global internet.
dalriata: "It's still massive amounts of matrix math"
Is it? I thought for BTC, it was basically "sha256 this block of bytes and check that the bottom N bits are zeroes"[1] or something like that? [been too bloody long since I cared about the details, hope I don't get this wrong] I would be surprised to learn that it isn't easy to massively parallelize. Hard to believe it needs advanced fabs, either. [deleted off-the-cuff bullshit argument for why that's the case, since it really boils down to "gee, I think so"]
I don't know what ETH uses for its PoW function.
[1] And before doing so, you stuff either a consecutive or random value (the "nonce") into the first word of that block (or something like that).
My R "friends" all know it is all dueto people are refusing to work because the government is giving them too much money.
I don't know how much inflation is around the corner, but I remain convinced that more would be better. For years now inequality has the biggest economic problem the U.S. faces. INflation reduces inequality, especially bewteen the richest and the middle class. Inflation dropped substantially during the Reagan years, and has stayed lower ever since. Inequality rose substantially at the same time, and has also kept rising ever since.
Who gets hurt the most by inflation? Lenders who extended long-term, fixed rate loans. That is, banks and their shareholders. And hedge funds that buy loan portfolios, or derivative products like collateralized debt obligations, from banks.
Who is completely unaffected by inflation? the working or non-working poor, who live hand-to-mouth, spending money as soon as it comes in, unable to borrow any significant amount and unable to save. Since income and outgo are equally affected by inflation, it makes no difference.
Who comes out ahead? Borrowers with long-term fixed the obligations: home owners with large mortgages; college graduates with education debt; and to some extent car owners and lessors. Some loans are variable rate, but many aren't, and many variable rate loans have upper limits. Basically, the middle quartiles of the economy, especially those under 45 or so. Anything thast gives to this group and takes form the banks and hedge funds should be encouraged.
The greatest trick the banking industry played on the American public was convincing them that inflation was bad for them, even as it was reducing the cost of their mortgages every month.
Opposite of a shortage: an overwhelming surplus of Girl Scout cookies.
https://www.nytimes.com/2021/06/15/us/girl-scouts-cookie-sales-membership.html
This isn't a thing in Scouting on knifecrime island, although when I was one, we did all kinds of charity tin rattling. Whenever I saw the phrase I always imagined them hawking cookies they'd, like, *baked themselves*? Which would be completely normal for a ultra-wholesome civic initiative here.
I had seriously never considered they would be distributing millions of boxes of standardized industrial baked goods but, well, of course, what could be more American than being really really enthusiastic about FMCG logistics?
There's an open box of them in the room with me now.
Anyway, my sister's troop sold all but six boxes, one of which is here.
29: Huh, thanks, I guess I really should've looked into it before spouting off. I had no idea it worked like that. That's way more boring than I thought.
30: Well, they sure suck.
I largely agree with 31, but a few quibbles; 31.3 depends a lot on wage stickiness (that's probably not the right term as we're talking about raising wages) doesn't it? Just because prices go up doesn't mean you can negotiate higher pay. And even if a worker can, that might take a few months to work its way through the system, making the worker poorer in a way they feel. And they'd certainly have trouble under stagflation circumstances (not that we're likely to hit that anytime soon).
31.4 assumes that banks won't build inflation expectations into their calculations, either by preferring-variable rate loans or greatly increasing the rate on fixed-rate loans. But it'd definitely be good for those long-term borrowers who bought on fixed terms under the old low-inflation situation.
Since income and outgo are equally affected by inflation, it makes no difference.
Wow this seems like it depends on some large assumptions, most of which seem false (especially in the short term).
35: Indeed. But they are very, very sure that they are correct.
What was that Latin or Greek word that clever economists were using in the early 2010's as an explanation for why jobs weren't coming back as fast as a standard model predicted? I think it literally meant like "stretching out"?
Speaking of shortages...
https://doomberg.substack.com/p/we-are-about-to-run-out-of-some-stuff
35: It's called sticky wages either way.
Which is what his-tiresias said before he became her-tiresias.
34: Did you steal that box?
No. Just ate a few. It was just short bread.
Given current inflation trends probably better to be long bread.