Describing the issues behind the cost prohibition of material science experiments and discovery, then handwaving it by just saying we can use supercomputing is incredibly ignorant of how inaccessible and coat prohibitive supercomputers are. Not to mention, even if your supercomputer has an idea you still need a lab that can confirm the idea.
They spend a lot of text describing the fact that it wouldn’t be possible without Frontier, so I think they are aware of the cost. The whole ecosystem of little, easy to apply for compute credit grants, and then larger grants (like the one they mention applying for at the end) is part of the solution to that problem. It doesn’t seem like ignorance, but rather a choice to emphasize in the writing that they are part of the solution.
Physical laboratory time, materials, and workers are also scarce and expensive. And physical lab tests can consume a lot of calendar time.
So, even with the constraints and costs, using the supercomputer to run 1,000 or 10,000 tests and then physical lab tests on only the few most promising results can still less costly in both funds and especially calendar time than the costs and time to run a similar number of physical lab tests to produce the results.
It is literally a balancing of the scarcity and costs of each mode of getting to the results, and those experts in the field, with their own labs, are making the claim that this is the way their research should proceed. Do you have actual information and specific knowledge that in fact proves them wrong (b/c that comment seems a lot more hand-wavey vs theirs)?
> Not to mention, even if your supercomputer has an idea you still need a lab that can confirm the idea.
I think this is the critical aspect. You don't need to own a supercomputer to run simulations in supercomputers. Sometimes you don't even need supercomputers at all. It might be a challenge but it's not prohibitive.
What's clearly a blocker is the verification & validation work to assess if a specific model actually works. You need to run experimental tests to check if your module is any good, and you need to run experimental tests to check if the output of your computational models correlate with reality. That costs both time and money.
Supercomputers are becoming more accessible. There's a large gap between research and real-world use, such numerical models need to be thoroughly tested for robustness before they can be deployed.
And according to experience, such studies are usually conducted simultaneously with Laboratory experiments, or already have Lab results they can compare with. If not, such study acts as a new source of synthetic MD data which can be picked up by the experimental folks.
I do agree that people who write such blogs tends to sell these ideas or studies as finished products which they are far from. The product of such studies are usually a paper or in other cases a patent. An idea others can build upon, like lego bricks.
Carbon Fiber is the new Asbestos. Ask anyone who works with the stuff, and I wisened up after I used to print with Carbon Fiber PLA, one day I looked at my hands under a microscope and was horrified to see microscopic little carbon fibers embedded all over my hands. Washing my hands just spread them around. They were embedded deep in my hands, all over work surfaces and probably floating in the air. Horrifying!
Very interesting, thanks for sharing your experience. But maybe this is the case for carbon fiber based filaments only (i.e. this is likely <1% of carbon-fiber as the use-case of 3D printing carbon fiber based hybrid plastic filaments is a new thing), and that most real-world carbon fiber is in the form of flat sheets whereby I doubt there are strands coming out.
The durability and uniformity of carbon fiber makes it an attractive alternative material for mass-production of high-quality musical instruments. The difficulty is getting it to vibrate enough to sound like wood. Meanwhile, the prices of tropical hardwood instruments are becoming solidly prohibitive. The improvement described in this article, making carbon fiber thinner, lighter and stronger, may be significant in that market.
10 comments
[ 3.6 ms ] story [ 38.4 ms ] threadPhysical laboratory time, materials, and workers are also scarce and expensive. And physical lab tests can consume a lot of calendar time.
So, even with the constraints and costs, using the supercomputer to run 1,000 or 10,000 tests and then physical lab tests on only the few most promising results can still less costly in both funds and especially calendar time than the costs and time to run a similar number of physical lab tests to produce the results.
It is literally a balancing of the scarcity and costs of each mode of getting to the results, and those experts in the field, with their own labs, are making the claim that this is the way their research should proceed. Do you have actual information and specific knowledge that in fact proves them wrong (b/c that comment seems a lot more hand-wavey vs theirs)?
I think this is the critical aspect. You don't need to own a supercomputer to run simulations in supercomputers. Sometimes you don't even need supercomputers at all. It might be a challenge but it's not prohibitive.
What's clearly a blocker is the verification & validation work to assess if a specific model actually works. You need to run experimental tests to check if your module is any good, and you need to run experimental tests to check if the output of your computational models correlate with reality. That costs both time and money.
And according to experience, such studies are usually conducted simultaneously with Laboratory experiments, or already have Lab results they can compare with. If not, such study acts as a new source of synthetic MD data which can be picked up by the experimental folks.
I do agree that people who write such blogs tends to sell these ideas or studies as finished products which they are far from. The product of such studies are usually a paper or in other cases a patent. An idea others can build upon, like lego bricks.