I don’t understand how baumol’s cost thing is any different than lower supply of labor causing price of labor to rise, and why it would merit a special name or Wikipedia entry.
It's not just about a lower supply of labor, really. It's not cleanly captured in terms of supply and demand. The reason a string quartet costs more now isn't because there are fewer people who can play instruments, it's because other things pay better.
In other words: if the average worker salary suddenly rose by 100% in every area except classical musicians (let's say there's some magical new productivity technology that increases productivity for every job except music, and further say that the workers capture that value) then the classical musician salary would also rise, because otherwise people would stop becoming musicians at the same rates. Nowhere in this thought experiment did the supply of labor change.
The supply of labor would have changed. See below where you wrote:
>because otherwise people would stop becoming musicians
It's about as cleanly supply and demand as you can get. The supply of musicians willing to work for $x decreases, since they now have the option to do other work that is more preferred than being a musician for $x. Therefore, you now have to pay more than $x to continue incentivizing a musician to be a musician.
Going back to your string quartet example, which I do not know if it's true or not, but let's suppose it is:
>The reason a string quartet costs more now isn't because there are fewer people who can play instruments, it's because other things pay better.
The reason a string quartet costs more now is because there are fewer people who can play instruments (relative to demand, since price is where the supply and demand curves intersect). Because other things pay better, fewer people (again, relative to demand) might choose to play instruments, causing less supply (relative to demand), causing prices to rise.
My point is that it's all still just simple supply and demand curves. If the demand for corn skyrockets, causing the price of corn to increase, and farmers choose to plant corn instead of wheat, then causing a decrease in the supply of wheat, then causing the price of wheat to increase, is that anything other than supply and demand?
All I can say is that it seems more like a second-order consequence of supply and demand, and an unintuitive one for many people. The counter factual/subjunctive “the musicians are paid more because otherwise they would have chosen other jobs” is not the way in which people usually talk about supply and demand. You could also say that the rules that govern the supply and demand of labor are not quite the same as the rules that govern supply and demand of goods, and people’s general failure to realize this results in the Bamoult effect being surprising.
> You could also say that the rules that govern the supply and demand of labor are not quite the same as the rules that govern supply and demand of goods, and people’s general failure to realize this results in the Bamoult effect being surprising.
I wouldn’t say that because I disagree there is any difference in the application of supply and demand curves between labor and goods. If anything, Baumol’s effect clearly demonstrates that price (wages) is set by supply and demand just like goods, and it’s entirely unsurprising.
As you reduce the supply of laborers for labor type A because those laborers have better options, then the price for labor type A rises. That’s what Baumol says. That’s what supply and demand says. I fail to see the significance.
On the contrary, "dark silicon" instead suggests that separating fp32 and int32 (now in GA102/104, fp32 and int32/fp32) data paths at the cost of more die space usage currently makes excellent sense. (See also: tensor cores, ray tracing cores.) Jensen Huang very briefly alluded to this when during the GA102/104 announcement he mentioned the end of Dennard scaling.
But the GA102/GA104 doesn’t have seperate execution units for INT and FP32 because the INT also does FP32. So I don’t see how that shows that separating FP32 and INT hardware makes sense.
Fun fact: Alex Krizhevsky's cuda-convnet was also an early adopter of CHWN tensor layout. Basically, having the batch size N as the major dimension limits you to running batch sizes that are multiples of the warp size (typically 32), but then you also have an easier time of implementing fast kernels for all your neural and tensor ops, including tensor convolutions, without getting nearly as stuck in the weeds of microarchitectural optimizations.
rustc does have a working nvptx target today, though it’s not supported nearly as well as the mainstream cpu targets, and some things you would really want for gpu programming (e.g. shared memory address space) are not currently exposed in the rust language. But kernels written in rust can compile to ptx; you’ll still need to write glue code.
Yeah this would be about extending it to columnar analytics funcs, like `df['x'].apply(f)`, `df.query("x > 10 && y < 10"). I realized I may be wrong about the compiler speed part, not sure if it'd be faster than numba for codegen nowadays :)
Her short story collections are quite accessible. If you want to dive into the Hainish cycle, _Four Ways to Forgiveness. Or for a mostly Hainish but more diverse set of stories, _The Wind's Twelve Quarters_.
