How do I convince my employer to buy me a supercomputer?
Name:
Anonymous2015-01-30 4:50
I don't want to own an old and busted supercomputer either. I want a Cray XC30. I'm done some research, and it seems that a good price would be $20,000,000. Now, I know it's going to be hard to convince them to sign for that, but I've compiled a document detailing the benefits. Firstly, all my programs would compile almost instantly. I estimate that it takes almost ten minutes for the current build server to recompile my team's project from scratch. That means my entire team suffers a collective hour each time that happens. I estimate that that means a loss of $23482346 per year to company productivity. Additionally, it would be good for morale, which I estimate to improve productivity by 12512%. Finally, compared to the cost of per-seat license that they are already paying for, it's change to them.
What examples have I missed? Please help me /prog/. If I get my requisition order approved, I promise that I will give you all shell accounts on it to access during non-work hours.
haha are you fucking 12 years old nigga?? hahaha SUPERCOMPUTER
Name:
Anonymous2015-01-30 12:39
stack my anus
Name:
Anonymous2015-01-30 13:16
This thread smell like oldfart: there are no more supercomputers dude, and Cray is almost dead... what you really want is CLOUD COMPUTING! with AWS you just pay for what you use, and everything is on da cloud, so you will never have to touch dirty hardware anymore :D
Name:
Anonymous2015-01-30 14:57
>>1 Super computers are suitable for solving only massively parallel tasks. Programmers don't really need them. The only thing programmer needs a fast build cycle, which solved by not using C/C++. For example, Symta supports separate compilation, where you can change a class method or add data members, without recompiling entire class.
The same way all decisions are made in business. Find the guy who makes the decision and suck his dick. If he still won't buy it for you, kill him and suck his replacements dick.
>>11 Changing a field in a C++ class's *.h file requires full program recompilation anyway.
Name:
Anonymous2015-01-31 1:04
>>12 No. You need only recompile the individual units that include the modified header. With a properly written makefile, most of the time for an incremental build is spent checking the filesystem and re-linking (mandatory operations that do not parallelize well).
Name:
Anonymous2015-01-31 1:14
I have 12 towers of xboxes in my room, each 10 units high, all connected via ethernet.
You need only recompile the individual units that include the modified header.
C++ code frequently uses some CObject, which gets included by every *.cpp and frequently modified, because it is a central class to program. Examples would CRasterLayer, CWaveform, CMesh, CGameEntity, CText, etc...
Name:
Anonymous2015-01-31 1:55
>>16 And how often do you change the class signatures of code that's already written and already mature as opposed to subclassing that code and overriding the old behavior?
Name:
Anonymous2015-01-31 2:47
And how often do you add mayo to the burger signatures of grease that's already cooked and already assembled into a meal as opposed to cumming into that burger and overriding the old taste?
Name:
Anonymous2015-01-31 3:41
Wait until Q3 2015 and build yourself your own supercomputer for under $10,000 USD. If you can't afford that, than your programming job is shit.
Why Q3? Because Intel is releasing the Skylake CPUs (16nm process, w/ DDR4 RAM and PCIe 4.0 capable motherboards as standard) and AMD is releasing the R9 3xx series GPUs.
A single R9 390x will have 4GB of HBM (3D stacked memory with 4x throughput of GDDR5), will be watercooled out of the box, and will be capable of 10 TFLOPs, with a price tag of around $600 USD.
Now, buy 8 R9 390x GPUs, and split between two Skylake based systems, so that each system has 4x GPUs, and connect both systems over CAT6 on your home network.
You now have an 80 TFLOPs compute cluster, which is equivalent in performance to a standard IBM Watson system, and for under $10,000.
Program the GPUs in OpenCL C or SYCL (essentially C++14 with OpenCL language extensions).