For definitions of 'very strong' equal to 'weaker* and more crack prone then conventionally cast parts, much weaker then parts machined out of forged billets, much, much weaker then forged, rough machined, heat treated, final machined''.
Nope; better than cast, equivalent to wrought static properties (final machining is required to get nice surface finishes and tight tolerances, as it would be with casting). "Sintering" is a misnomer; "selective laser melting" would be more accurate. The parts are fully dense.
Unless I was already 80 they aren't getting any implants into my joints made out of anything short of Titanium.
Want Titanium for that hip, old man? Try i.materialise 3d-print service...
Repeat after me: '3d printers are not able to make full strength metal parts. It is extremely unlikely they ever will be able to. 3d printers are not able to make full strength plastic parts. It is extremely unlikely they ever will be able to.'
On the 3D printing front, gimme one that prints steel, aluminium alloys, etc. with the structural integrity of their conventially produced equivalents (i.e. not sintered) and I'll start to take this discussion seriously.
EOS, 3DSystems and Arcam make systems that make Steel, Aluminum, Inconel and Titanium parts with mechanical properties equivalent to (and in some cases surpasing) wrought parts. The "sintering" in the trade names is a misnomer. The process is actually micro welding with lasers or electron beams. Straight out of the machine the parts are fully dense and useful.
In one week I can do experiments that 5 years ago would have taken 10 people a full year to perform. With such throughput it isn't necessary even to formulate a hypothesis. You just test every possible variation and let the data speak for itself. Machines are more consistent than people, don't get tired, if the make mistakes the mistakes are systematic and easy to troubleshoot.
I have to admit, the high-throughput stuff is pretty neat though. I think the great part about it is that it frees up the human to think more about test design and hypothesis formulation and worry less about the mechanics of putting drop A into beaker B.
The annotated ITAR indicates (121.1 Category XV (c)(2), pp50 in that pdf) that there is a speed and altitude restriction: "Designed for producing navigation results above 60,000 feet altitude and at 1,000 knots velocity or greater".
Hopefully they will get credit because their receivers worked at low-speed and low-altitude (on the way down), and they've already integrated their accelerometer data to get very reasonable velocity and position estimates.
"one of the requirements is GPS data over 100k. Even with four separate GPS systems, we were not able to get a high altitude fix." With no tangible record of the rocket's soaring ascent, it's unlikely that Deville and his friends will score the cash. Amateur Qu8k Rocket...
One of the significant hurdles for Carmack's prize is the ITAR speed/altitude restrictions on most GPS receivers. It will be interesting to see what sort of receiver they used; hopefully at least one of the four was an unrestricted receiver.
Rather that a full up Navier-Stokes solver (I was going to say openFOAM too), since it's a short section of a course, maybe just have them play with the NACA airfoil potential flow solver; that's pretty neat. You can explain the simplifications to the governing equations between "real" CFD and potential flow, and show them that the simple models can still be useful in certain situations.
Griffin was counting on it being too big to fail, so thought he'd get additional funding to cover the overruns.
Absolutely right, he tried to do a Pentagon-style program without a significant Iron Triangle backing him up, so instead of being too big to fail, it was just too big.
I still don't understand how the orbiter would have ever made sense though, every pound of air-frame is a pound of payload given up. Paying to get it up in the first place is expensive, you should bring back as little as possible (teeny-tiny re-entry capsule, separate cargo / orbital lab / what-have-you). Just because the shuttle was initially part of a larger program doesn't make it smart or modular.
The core tenant of design for a long, long time has been modularity and leveragability.
Does long, long time mean 'since the total cluster that was the shuttle design'? Hauling all that tile and structure was a really horrible design decision, Zubrin's criticism of it was spot on.
Not surprised Griffin is trying this. He's always had some agenda.
Griffin is / was such a douche bag. I do not understand why everyone kissed his ass so much, 'oooh, he wrote teh book!!1elventyone!!!'. Yeah, guess what? He spent all that time getting degrees and writing text books and not building or flying hardware!! When the boss draws a rocket on a napkin (Ares I) and ram-jams it down the organization's throat, he is a total and unrepentant jerk-off and should never be trusted with any position of authority ever again. Unfortunately this is exactly the kind of well-credentialed, but worthless asshole that gets promoted in the government, and NASA happens to be particularly flagrant with this sort of buffoonery.
Ah, ranting feels good, especially on the internet where it lasts forever...
Funding NASA helps fund the research and development that allows for the possibility of creating that infrastructure we so desperately need up in space in order to do any of it.
I'd argue funding NASA prevents the creation of infrastructure (gun / laser launch, systems of tethers / rotovators) because NASA can afford one-off rocket shots which result in no residual infrastructure, whereas private industry would have to be smarter (to be affordable).
No more external fuel tanks are being manufactured, the rest of the parts chain is shutting down. When the shuttle is gone, America loses manned access to space
There are still enough residual ETs to do some interesting prototyping things, such as a shuttle derived heavy.
The Constellation program sounded like a real soup sandwich. Cancelling it would be a good thing if it paved the way for something done right. But that's not happening
The shuttle was just farting around in LEO.
See SpaceX's first Falcon 9 in the hangar at the cape, at least it's got the possibility of farting around in LEO more cheaply. Diverting money from Constellation to COTS is paving the way.
If this sounds like hell to you, hop in your El Camino, crank up the Metallica, and head to McBurgerndy's-Fil-A-Bell. Buy three triple whopper chicken bacon cheese towers, a SuperJumbo Coke, a sixty ounce curly mayonnaise french fry bucket, and of course thirty dozen cinnamon twisters. (Don't forget your blood sugar! Your kidney dialysis isn't until next week.) Stuff two of the burgers into your mouth, gorge on the fries...
You know what I’m gonna do
I’m gonna get myself a 1967 Cadillac Eldorado convertible
Hot pink, with whale skin hubcaps
And all leather cow interior
And make brown baby seal lions for head lights (yeah)
And I’m gonna drive in that baby at 115 miles per hour
Gettin' 1 mile per gallon,
Sucking down Quarter Pounder cheeseburgers from McDonald’s
In the old fashioned non-biodegradable styrofoam containers
And when I’m done sucking down those greeseball burgers...
I haven't the foggiest idea how a temperature gradient can cause matter to climb out of a gravity well.
Thermophoresis causes particles in a fluid to move because of a temperature gradient. The similarity parameters (Reynolds / Mach / Knudsen) for a planetesimal in an accretion disk are probably similar to the aerosal particles in air that the wiki article talks about.
To make any statements, you would first need to make some observations
Or you could be a Bayesian, make some assumptions, include a priori info in the analysis (which you should probably do anyway even if you have data); before you get up to check if there is an elephant in the kitchen assigning equal priors to the two hypothesis is a sound maximum entropy sort of method. You can then update your 50/50 state of knowledge after observing zero or many elephants in your kitchen.
"the data comes from an underlying normal distribution"
You mean we often assume the residuals are normal; the data could be any distribution at all, that's why we fit models. There's plenty of parametric stuff you can do with different distributions on the residuals too; Google "R glm Poisson", "R glm binomial", or "R glm family".
You might already know all this, but ever since that black swan book came out there's a bunch of statistical-illiterates running around saying, "the whole world's not normal", without understanding that everybody who understands the world and statistics understands that already too.
Take a problem that is all 64-bit integer math and has a branch every couple hundred instructions and GPUs will do for crap on it.
So would a Cray; supercomputers and GPUs are made for the same sorts of problems (exploiting data parallelism). Now if by 'supercomputer' you mean 'a cluster of commodity hardware', then ok, you've got a point, that heap of cpus will handle branches plenty fast.
Well, I'm not sure about most of your criticisms, but they use Jacobi instead of Gauss-Seidel because SSOR is not data parallel, but Jacobi is.
That would make the performance the same as for the GPU system.
Really? Care to share any results that support that? I'm quite sure the peak flops you can achieve on the GPU are much higher than the limited SIMD capability of the CPU.
Note that I am being generous here and actually ignoring the program setup time when they need to copy the data to the GPU.
Sure there's communications overhead, but that's true of any parallel processing problem, the trick is to find problems that have a big computation to communication ratio (which happens to be most of computational physics and these tomographic reconstruction problems that TFA mentions as well).
you can get absolutely incredible performance out of off-of-the-shelf GPUs these days.
I had heard this from folks, but didn't really buy it until I read this paper today. They get a speed-up (wall clock) using the GPU even though they have to go to a worse algorithm (Jacobi instead of SSOR). Pretty amazing.
Slashdot Mirror
GE is going to production with 3D printed fuel injectors and uses 3D printed turbine blades for testing right now.
See also this Direct Metal Laser Sintered (DMLS) 3-D printed regeneratively cooled engine: http://rocketmoonlighting.blogspot.com/2011/07/testing-update.html http://rocketmoonlighting.blogspot.com/2013/09/seeing-triple.html
For definitions of 'very strong' equal to 'weaker* and more crack prone then conventionally cast parts, much weaker then parts machined out of forged billets, much, much weaker then forged, rough machined, heat treated, final machined''.
Nope; better than cast, equivalent to wrought static properties (final machining is required to get nice surface finishes and tight tolerances, as it would be with casting). "Sintering" is a misnomer; "selective laser melting" would be more accurate. The parts are fully dense.
Unless I was already 80 they aren't getting any implants into my joints made out of anything short of Titanium.
Want Titanium for that hip, old man? Try i.materialise 3d-print service...
Repeat after me: '3d printers are not able to make full strength metal parts. It is extremely unlikely they ever will be able to. 3d printers are not able to make full strength plastic parts. It is extremely unlikely they ever will be able to.'
Why would he want to repeat that? Fully dense metal parts with equivalent-to-wrought properties out of the machine and with a bit of stress-relief heat-treatment: http://www.eos.info/en/products/systems-equipment/metal-laser-sintering-systems.html http://production3dprinters.com/slm/spro125-direct-metal-slm-production-printer
On the 3D printing front, gimme one that prints steel, aluminium alloys, etc. with the structural integrity of their conventially produced equivalents (i.e. not sintered) and I'll start to take this discussion seriously.
EOS, 3DSystems and Arcam make systems that make Steel, Aluminum, Inconel and Titanium parts with mechanical properties equivalent to (and in some cases surpasing) wrought parts. The "sintering" in the trade names is a misnomer. The process is actually micro welding with lasers or electron beams. Straight out of the machine the parts are fully dense and useful.
In one week I can do experiments that 5 years ago would have taken 10 people a full year to perform. With such throughput it isn't necessary even to formulate a hypothesis. You just test every possible variation and let the data speak for itself. Machines are more consistent than people, don't get tired, if the make mistakes the mistakes are systematic and easy to troubleshoot.
I have to admit, the high-throughput stuff is pretty neat though. I think the great part about it is that it frees up the human to think more about test design and hypothesis formulation and worry less about the mechanics of putting drop A into beaker B.
The annotated ITAR indicates (121.1 Category XV (c)(2), pp50 in that pdf) that there is a speed and altitude restriction: "Designed for producing navigation results above 60,000 feet altitude and at 1,000 knots velocity or greater".
Hopefully they will get credit because their receivers worked at low-speed and low-altitude (on the way down), and they've already integrated their accelerometer data to get very reasonable velocity and position estimates.
Yes, they know how to launch legally.
They didn't get a lock with any of their receivers at apogee.
"one of the requirements is GPS data over 100k. Even with four separate GPS systems, we were not able to get a high altitude fix." With no tangible record of the rocket's soaring ascent, it's unlikely that Deville and his friends will score the cash. Amateur Qu8k Rocket...
One of the significant hurdles for Carmack's prize is the ITAR speed/altitude restrictions on most GPS receivers. It will be interesting to see what sort of receiver they used; hopefully at least one of the four was an unrestricted receiver.
Rather that a full up Navier-Stokes solver (I was going to say openFOAM too), since it's a short section of a course, maybe just have them play with the NACA airfoil potential flow solver; that's pretty neat. You can explain the simplifications to the governing equations between "real" CFD and potential flow, and show them that the simple models can still be useful in certain situations.
Griffin was counting on it being too big to fail, so thought he'd get additional funding to cover the overruns.
Absolutely right, he tried to do a Pentagon-style program without a significant Iron Triangle backing him up, so instead of being too big to fail, it was just too big.
I still don't understand how the orbiter would have ever made sense though, every pound of air-frame is a pound of payload given up. Paying to get it up in the first place is expensive, you should bring back as little as possible (teeny-tiny re-entry capsule, separate cargo / orbital lab / what-have-you). Just because the shuttle was initially part of a larger program doesn't make it smart or modular.
The core tenant of design for a long, long time has been modularity and leveragability.
Does long, long time mean 'since the total cluster that was the shuttle design'? Hauling all that tile and structure was a really horrible design decision, Zubrin's criticism of it was spot on.
Not surprised Griffin is trying this. He's always had some agenda.
Griffin is / was such a douche bag. I do not understand why everyone kissed his ass so much, 'oooh, he wrote teh book!!1elventyone!!!'. Yeah, guess what? He spent all that time getting degrees and writing text books and not building or flying hardware!! When the boss draws a rocket on a napkin (Ares I) and ram-jams it down the organization's throat, he is a total and unrepentant jerk-off and should never be trusted with any position of authority ever again. Unfortunately this is exactly the kind of well-credentialed, but worthless asshole that gets promoted in the government, and NASA happens to be particularly flagrant with this sort of buffoonery.
Ah, ranting feels good, especially on the internet where it lasts forever...
Funding NASA helps fund the research and development that allows for the possibility of creating that infrastructure we so desperately need up in space in order to do any of it.
I'd argue funding NASA prevents the creation of infrastructure (gun / laser launch, systems of tethers / rotovators) because NASA can afford one-off rocket shots which result in no residual infrastructure, whereas private industry would have to be smarter (to be affordable).
No more external fuel tanks are being manufactured, the rest of the parts chain is shutting down. When the shuttle is gone, America loses manned access to space
There are still enough residual ETs to do some interesting prototyping things, such as a shuttle derived heavy.
The Constellation program sounded like a real soup sandwich. Cancelling it would be a good thing if it paved the way for something done right. But that's not happening
The shuttle was just farting around in LEO.
See SpaceX's first Falcon 9 in the hangar at the cape, at least it's got the possibility of farting around in LEO more cheaply. Diverting money from Constellation to COTS is paving the way.
Here's a round-up of the recent news with links describing that stuff.
If this sounds like hell to you, hop in your El Camino, crank up the Metallica, and head to McBurgerndy's-Fil-A-Bell. Buy three triple whopper chicken bacon cheese towers, a SuperJumbo Coke, a sixty ounce curly mayonnaise french fry bucket, and of course thirty dozen cinnamon twisters. (Don't forget your blood sugar! Your kidney dialysis isn't until next week.) Stuff two of the burgers into your mouth, gorge on the fries...
Dennis Leary did it better:
You know what I’m gonna do I’m gonna get myself a 1967 Cadillac Eldorado convertible Hot pink, with whale skin hubcaps And all leather cow interior And make brown baby seal lions for head lights (yeah) And I’m gonna drive in that baby at 115 miles per hour Gettin' 1 mile per gallon, Sucking down Quarter Pounder cheeseburgers from McDonald’s In the old fashioned non-biodegradable styrofoam containers And when I’m done sucking down those greeseball burgers...
What tools? Well, I'm not sure, but I've narrowed it down to two: it's either Emacs or Vi...
Plus even if you do get a sizeable exposure (it needs to be ingested to have any effect), you just need to drink a sixer of cheap watered down beer.
I haven't the foggiest idea how a temperature gradient can cause matter to climb out of a gravity well.
Thermophoresis causes particles in a fluid to move because of a temperature gradient. The similarity parameters (Reynolds / Mach / Knudsen) for a planetesimal in an accretion disk are probably similar to the aerosal particles in air that the wiki article talks about.
To make any statements, you would first need to make some observations
Or you could be a Bayesian, make some assumptions, include a priori info in the analysis (which you should probably do anyway even if you have data); before you get up to check if there is an elephant in the kitchen assigning equal priors to the two hypothesis is a sound maximum entropy sort of method. You can then update your 50/50 state of knowledge after observing zero or many elephants in your kitchen.
"the data comes from an underlying normal distribution"
You mean we often assume the residuals are normal; the data could be any distribution at all, that's why we fit models. There's plenty of parametric stuff you can do with different distributions on the residuals too; Google "R glm Poisson", "R glm binomial", or "R glm family".
You might already know all this, but ever since that black swan book came out there's a bunch of statistical-illiterates running around saying, "the whole world's not normal", without understanding that everybody who understands the world and statistics understands that already too.
Take a problem that is all 64-bit integer math and has a branch every couple hundred instructions and GPUs will do for crap on it.
So would a Cray; supercomputers and GPUs are made for the same sorts of problems (exploiting data parallelism). Now if by 'supercomputer' you mean 'a cluster of commodity hardware', then ok, you've got a point, that heap of cpus will handle branches plenty fast.
That would make the performance the same as for the GPU system.
Really? Care to share any results that support that? I'm quite sure the peak flops you can achieve on the GPU are much higher than the limited SIMD capability of the CPU.
Note that I am being generous here and actually ignoring the program setup time when they need to copy the data to the GPU.
Sure there's communications overhead, but that's true of any parallel processing problem, the trick is to find problems that have a big computation to communication ratio (which happens to be most of computational physics and these tomographic reconstruction problems that TFA mentions as well).
you can get absolutely incredible performance out of off-of-the-shelf GPUs these days.
I had heard this from folks, but didn't really buy it until I read this paper today. They get a speed-up (wall clock) using the GPU even though they have to go to a worse algorithm (Jacobi instead of SSOR). Pretty amazing.