Pricing per gram please
It is many decades since I measured anything in ounces. I expect that will be true of many/most readers.
For as long as I have been a reporter and analyst in the IT sector, November has always been supercomputing month. Way before there was a TOP500 ranking of supercomputers in June 1993 but just as I was leaving university, the first Supercomputing Conference was held in Orlando in 1988. And that November SC show set the cadence …
The US chooses to use archaic measurements with arbitrary multipliers and divisors. The rest of the world laughs at them, while the US can be irrationally proud of their choice.
Most other countries use a normal system of measurements that's easily computed and communicated, and they are proud of that choice.
And then we have the UK who joins in with the rest of the world when laughing at the US but then goes and uses most of the same measurements anyway, with the bonus that they reinvented several of them, kept the same names but changed the numbers. The difference between the UK and the US is that the UK seems to lack the will to do what they claim to do when comparing themselves. They describe height in feet, weight in stone and pounds (and hence ounces), volume in pints (and also ounces, but different ones), and distances in miles. There are of course people in both countries who eschew the old units either because their work (almost anything scientific, for example) or their personality prefers metric. If you're outside both countries, you can easily look down on both, but if you're in the UK, be careful where you aim your judgement.
I've grown to like the US "archaic" units because they're human-sized and scaled imho. An ounce is the perfect size for a shot of whiskey, and, in the case of water, a liquid ounce is equivalent to a solid ounce, unlike the litre that weighs a kilo-gram (1000x mismatch).
Plus, 16 ounces to a pound is a nice 2⁴ scaling (an hexadecimal nibble), and fractions are omnipresent in gallons, halfs (1/2), quarts (1/2²), pints (1/2³), and even fifths (1/5) for the decimally oriented.
A mile is about as far as one can walk before being tired and turning back (a kilometer is way too short for that), and 12 ounces is the perfect size for a beer bottle (not sure which complete moron invented the 25 centilitre beer bottle that is just way too small for human satisfaction ... it may only be fit for a cat or dog or somesuch, very disorienting, and a terrible sign of creeping inflation!).
I'm not sure decimal is made for humans (cowboys?), maybe it's best suited to high-efficiency terminator-like robots (civil servants?) ... ;)
Opinions can vary, but just to correct a misconception, an ounce of water does not weigh one ounce using US measurements. In the US, one fluid ounce of water at 4°C weighs 1.043 ounces. It's a little different in the UK, where if you combine two different laws that redefined units, then one ounce of water does weigh one ounce, except that the volume unit was defined in 1824 and the weight unit in 1963 and before 1963, the pound and ounce were different, thus meaning that anyone previously using the ounce as volume had to decide whether to change to match the new weight units or retain the old volume units where one ounce of water would no longer weigh one ounce. If they're useful enough that you're going to keep counting things in them, I'm almost certainly not able to convince you otherwise, but be careful when deciding these work when precision is needed unless you've been very careful about which ounces from which country and year you're using and given the same information to anyone relying on that.
There's no mismatch at all, actually, because the litre is a unit of volume and the SI unit for volume is the cubic metre. 1 litre is 1/1000th of a cubic metre, just as 1 gram is 1/1000th of a kilogram (the SI unit for mass), so they have the same relationship to their respective base units. It's an easy mistake to make, though - most people assume the litre is the base unit of volume because it's exactly 1kg of water.
The french (and others) used to use pounds, ounces, inches and feet too
The problem is that all of them differ slightly and even the US/UK inches used to be fractionally different until the "engineering inch" was defined in the 1930s (Napolean is supposedly 5'2" - In french units yes. In English units he was around 5'6")
Like troy vs avoirdupois ounces, US/British gallons are based on different standards - and in this case it's the BRITISH using the new-fangled method whilst the Americans kept on with the older one
Until 1824 the wine gallon was 231 cubic inches(3.78 litres) and the ale gallon was 282 cubic inches(4.62 litres)
After that date the British switched to defining gallons as the volume of 10 pounds of water at 67F(17C) - for obvious reasons two different standards WITHIN THE SAME INDUSTRY is problematic in an age of inccreasing mechanisation
There's also the "corn gallon" (which had 5 different sizes from 1497-1688 and is also the USA dry gallon of 270ci/4.4 litres), the "coal gallon" whilst Ireland/Jersey/Guernsey all had their own definitions.
The important thing about grams and metres is that they're the same everywhere and that matters for international trade. You CAN sell a pound of bananas in Britain but the pricing is per kg and the weight has to be stated in kg along with whatever other measure used
"They describe height in feet, weight in stone and pounds (and hence ounces), volume in pints (and also ounces, but different ones), and distances in miles. "
Nah, it's a generational thing.
I was bor and decimalisation came in so I'm
5"10, weigh 67kg, measure short distances in mm, cm and m, but drive miles to work, but I did used to run a 5k most mornings. Couldn't tell you the MPG of my car but I know the price of petrol by me is £134.9 a litre.
I drink pints in a pub, but do like the odd 500ml can of beer from time to time.
Then you could buy a litre, a 2 pint or 2 litre bottle of milk at the same time as buying your 500g of cheese to put on your 1/4lb burger.
Ask my kids and they'll tell you their height and weight i metric, They measure all distances and weights in metric and will do the mental gymnastics to convert long distances to Km
Gold is in Troy ounces
I might guess the GPUs were in kilos or grams and converted to Avoirdupois (= US customary units.)
Troy oz. 31.1034768 g
Customary oz, 28.3495231 g
Which I think would slightly undervalue the hardware ~28.35/31.10 (.91)
Cattle in cwt (live weight?) reduced to $/oz — only in el Reg.
Worked for a software company who supplied software to the police. We received a request for an icon for an extremely ie *lot* more than gold, item which criminals were selling. It wasn't drugs or precious metals but ... bull semen! If you can get a kilo of the stuff from a pedigree bull, you can buy yourself a very nice house.
... a different kind of white gold moo juice indeed, frozen in a canister kind of like a milk jug, and worth some $300 to $1,500 per 0.5 ml vial (straw?) - $18,000+ per ounce ... (other calcs gave me $3,600/Oz but at this rate ... who's counting!)
Can we get the pricing in reg units please, or maybe body parts? How many kidneys/eyes/lungs/testicles do I need to harvest - I mean lose - to get a NVIDIA DGX Spark or an RTX Pro for my home Workstation.
Seriously, prices for even basic GPUs for today's games is extortionate.
Sure, now lets actually compare to the silicon, not the silicon plus a bunch of electrical supply and plumbing.
A 300mm wafer weighs about 90g and is worth about 20K to 40K. That's $220 to $440 per g, or 6K to 12K per oz.
Known good die from that will be more expensive by another factor of two or so.
And it looks like gold is way over-valued today, much as palladium was in 2021, so it'll probably come back down to half its current value -- don't buy gold now, sell it instead if you have some. Same with GPUs, best to sell them, plus (as noted) they do depreciate as fast as brand new premium cars driven fresh off the lot, so ... if needed, it's best to buy second-hand ones imho.
Second-hand cattle on the other hand ... ;)
But you get real products from a herd of cows. Unless you can make a server farm produce good milk and good meat...
Regarding those mini nukes: Russia and China already have, Canada seems to be the first to start building one (for civil energy production), USofA is currently too busy with internal affairs.
mini nukes (10W radiothermal units) are horribly inefficient, not to mention dangerous as hell around civilians, Sooner or later someone tries to pull them apart for the scrap metal value
SMRs are too small to be economic (you need your steam plant sized to drive 1GW generators and the limiting factor is the size of the turbines - if they could be made bigger they would be) because whilst a smaller nuke might be easier to build, the biggest cost driver after construction is maintenance of the steam turbines - a problem exacerbated in water-cored nukes by only being able to make "wet" steam at a maximum 350C which tends to cause turbine blade pitting as droplets condense, vs the dry/supercritical steam made in combustion plants (or molten salt reactors)
Just like trying to haul cargo with a Ferrari vs a Kenworth, the upfront price of a SMR might be half that of the bigger plant, but the added operating costs will eat you alive in short order
The most promising mid-term nuclear prospect is China's TMSR series. They've gone where the USA ventured/abandoned in 1965-69 whilst other countries are still playing around with engineering drawings. More importantly, they've demonstrated that thorium can be used as the fuel without external processing, which pretty much upends the entire nuclear weapons industry
Enriched uranium as used in LWR reactors was originally the unwanted waste product of nuclear weapons manufacture. Nobody builds bombs from U235 because it's too espensive (a U235 bomb would be ~$3-20 billion depending on design whilst plitonium ones are a few tens of millions apiece. Weapons-grade uranium is around half a billion dollars per kg, which is one of the reasons that even the USA only has a few nuke boats - the fuel costs way more than everything else onboard (and the boat) put together
Enriched uranium was used on the first LWR reactor because thorium wasn't available, they knew it would work and enriched uranium allowed them to build a reactor small enough to use in a submarine hull (Nautilus)
Nobody would produce 3% enriched uranium for LWR designs if there wasn't a market for the 89% "wasted" depleted uranium (DU is the feedstock for making weapons-grade plutonium). After all, you can use more natural uranium to achieve criticality in a ,uch larger shore-based reactor and at $150/kg for natural uranium vs the $60-85,000/kg of 3% enriched uranium you can afford a lot of it.
Or to put it another way: "LWR nuclear power is a figleaf covering up your weapons programs."
All enriched uranium nuclear power has a weapons program behind it.
All solid-fuelled reactors designs can be abused to make weapons-grade plutonium.
Thorium (U233) bombs aren't practical because by the time you've finished extracting enough U232 from your U233 to have a workable weapon, the cost is even higher than U235 weapons plus you have a wildly radioactive (as in "lethal to be near") processing plant whilst if you don't remove enough U232, your bombs fizzle AND kill the ground handling crews
Spent nuclear fuel isn't weaponisable because it would cost more than weapons-grade U235 and would leave you with the same processing issues as thorium
It all comes back to depleted uranium being the practical feedstock for nuclear bombs, whilst the USA has actively discouraged/blocked systems capable of using thorium or natural uranium because divorcing civil nuclear power from the dependency on nuclear weapons waste would render all processing a military activity (instead of dual use) and even make uranium mining dubious because thorium is available in massive quantities as the unwanted waste product of rare earth mining (There's more thorium in monazite than all the lanthides comobined and extracting it is required to get at everything else, so you have it whether you want it or not)
I've fairly confident that by the mid 2040s China will be the global hyperpower thanks to its ability to sell energy systems to 6 billion people - and the fallout associated with moving away from an economy based on energy scarcity will be interesting. (Renewables can't meet the energy requirements for full decarbonisation and in any case, molten-salt thorium looks set to undercut the costs of renewables along with most other power generation methods currently in use)
I suppose it depends how we're making herd sizes equal so we can compare them. If it's equal numbers of cows or servers, then we also need a good idea for what a single server is, but they end up being surprisingly similar.
We can try a quick calculation to estimate this. A relatively average cow might consume enough food to generate 50 MJ of energy per day, which if divided equally produces an average power consumption for one cow of 579 W. An NVIDIA H200 can consume 700 W. If they were equally efficient, then the cows should be about 15% less CO2-producing than an equal number of H200s.
But are they equally efficient? This starts to get into weird areas. For example, if I count the emissions from constructing the datacenters and piping in water to cool them, which I can find numbers for online, then shouldn't I also count the construction of farms and irrigation systems used to make the feed for the cows and the water they drink directly? I'm too lazy to do that, so let's go with something a little more complex than the last estimate but that I can finish doing today. One page gives us an average methane production rate of 100 kg/year or 274 g/day while another broadly agrees with that number (they used pounds, though) and also gives us an average CO2 emissions rate of 6,137 L/day (about 11.2 kg). Using this calculator, we get an average CO2-equivalent emissions rate of 0.019 t/day for a cow. Our 700 W H200 probably needs about the same amount of cooling power usage, not counting the water as described above. How does a 1.4 kW load look? Using the same calculator, we can use an average power emissions rate from the US which gives us 0.013 t/day for that load. So it looks like cows are less efficient than servers, which isn't that big a surprise given all the other biological stuff involved.
... and which is more intelligent? (hint: one of them exhibits "distinct personalities", "social learning", "self-agency", "cognitive judgment", and "are able to make sophisticated discriminations among not only objects but humans and conspecifics" ... the other one is more like dead meat by comparison I think!)
"a penny in the United States now takes about 4 cents to make"
and it's been copper-plated zinc for at least 35 years, with the "silver" coins being nickel-plated copper
The UK has used copper/nickel plated steel in the 1/2/5/10p coins for the same period (20 and 50p are nickel plated copper)
Other countries simply discontinued their "copper" coins in the late 1980s-early 90s and round to the nearest 5c (or equivalent 5 centiunit value)
The ironic part is that zinc is almost as expensive as copper and 3-12 times the price of steel (steel price depends heavily on composition)
The mention of the 6000 series brings back memories of college. My school had a 6500 I learned assembly on (and RATFOR, a structured fortran) as well as several other languages. The 60 bit word, with 6 bits used for characters. Machine was in the 2nd down basement and students got access to card readers on the 1st level basement. School my sophomore year traded up to a 6600. By then I'd already started EE classes and that area had take the VAX/UNIX road.(11/70) Terminals were so much better than card readers!
By my quick and possible faulty calculation, GPUs are now about 10x more expensive by weight than HP printer ink. So you can't really call them cheap.
On the upside, there's no forced retirement of GPUs where after 1 year of work, the chips go into limp mode and require an electron refill for 300% of the original purchase price to recover their original performance level.
Well, not yet at least.