[onion2k]

If his conclusion were true companies like Shapeways would only use cheap hobby printers. But ofcourse it is not.

It's far from a perfect analogy, but with servers Google demonstrated really well that the cost of high-end 'professional' equipment can be a waste of money where cheap, commodity kit can be used at scale instead. It's not completely unreasonable to think that a farm of cheap 3D printers could be used in place of one high-end model eventually.

[jib]

People are freaking expensive. I think that is something that you miss easily.

200k sounds expensive for a printer, yet that is the cost of 1 guy for three years (assuming western country).

The 200k machine just has to be a small bit better than the 2500 machine for it to pay off so fast it is ridiculous. A commodity approach to printing sounds like it would involve a lot of quality checks and other really pricy manual things that would eat up that 200k in under a year.

If you can remove all human costs, then yes, commodity gear is likely better, because you are paying for gear that knows that humans are expensive, so even small savings in humans translate to a ton of cost, and so the professional gear is priced accordingly.

[onion2k]

The 200k machine just has to be a small bit better than the 2500 machine for it to pay off so fast it is ridiculous. A commodity approach to printing sounds like it would involve a lot of quality checks and other really pricy manual things that would eat up that 200k in under a year.

Owning the 200k machine has to be better than owning 80 of the cheaper machines for it to be better. In a lot of scenarios that will be the case - if you need the improved resolution, reliability, etc that the 200k machine offers then the cheaper machines just aren't an option. In time though, as the cheaper machines improve, the expensive machine may stop being a better option.

[bonzini]

> Owning the 200k machine has to be better than owning 80 of the cheaper machines

No, his point was that with the cheaper printer you might need to hire one guy to look after the quirks of the 2500 machine and customize each part so that it works fine with the cheaper machine. If the 200k lasts more than 3 years, it's already paid for itself, because after 3 years the cheaper machine would cost you 2500 for the printer and 200k for the guy.

[lawless123]

"In time though, as the cheaper machines improve, the expensive machine may stop being a better option."

Or in time the expensive machine gets cheaper.

[onion2k]

That might also happen, but in general cheap things get better faster than expensive things get cheaper.

[baq]

the difference in 3d printers and servers is rather hard to compare: different servers can compute the same result, faster, slower, but it'll be the same. not exactly true with 3d printers, where a result isn't binary.

[dpark]

Servers aren't actually binary, either. Your cheap server might serve the same result (binary), but performance might degrade under load more rapidly (nonbinary), the failure rate might be be higher (nonbinary), and the support for cheap hardware might be nonexistent (nonbinary). There's also the issue of SKU drift/mix as your cheap hardware fails and gets replaced with a different/newer SKU (adds maintenance cost, is also nonbinary).

At the end of the day, the cost difference plus the ability to mitigate/hide the problems via software makes the commodity hardware the winning option, but it's not just a matter of "same bits come out; good enough".

Commodity hardware is also not really cheap. These are 1U/2U servers that cost thousands of dollars. These are cheap relative to mainframes, but they're not consumer devices.

[TD-Linux]

In fact, the $2000 Lulzbot printer used in the article is made of lots of 3D printed parts, manufactured on a farm of Lulzbot printers: https://www.youtube.com/watch?v=v_jUObUGLTA

[spacecowboy_lon]

I think your missing the point that big data commodity systems are not the $199 pc's ypou buy at walmart.

For hadoop a comodity box is a dual xenon white box with 192G of ram plus infiniband networking kit.

[timlyo]

The difference here is reliability. Commodity kit is generally reliable, just slower; whereas cheap 3d printers are more likely to fail a print/produce junk.

[seizethecheese]

In which case, the junk can be tossed and tried again.

[VLM]

Meanwhile the assembly line and workers expensively wait for the assembly jig to be printed. And the contract due date doesn't automagically move out a day because you had a printing problem, so now you're paying (more?) overtime once you finally get a working jig.

There is a solution to that, when your labor and capital costs are huge but the cheap printer is 1/10th the cost of the fancy printer... simply install and use 3, 4, maybe 5 cheap printers, keep the best print, and pocket the savings.

[gh02t]

Yeah, but 3D printers are also really slow. The part in the article took 18 hours (on both printers) and in my lab we've had similar experiences.

Not sure where the break-even point is, but the cost of a failed print is pretty high in terms of sunk time for most prints. On an industrial scale, a machine that prints correctly 99% of the time might be a better investment than two that print say 75% of the time.