How would you build a rig for machine learning?
>rig
>>58600996
GPUs. As many as you can fit in a 4U rackmount. Get a motherboard with as many PCI slots as you can find, which are pretty much gauranteed to be dual socket. CPU doesn't matter as much. Just get cheaper xeons.
>>58600996
>L3 Dollar
>>58600996
Stuff it with the best Nvidia GPUs you can find. Sadly, some of the best GPU machine learning libraries are exclusive to their CUDA jewery.
>>58600996
An FPGA that is basically 95% DSPs and some LUTs to glue them together.
>>58601472
FPGAs designed as neural networks are technically classified as neuromorphic computers, and are banned under U.S. weapons export laws. Other countries have similar laws.
>>58601489
responding to myself becuase I wanted to mention that the reason why its under weapon export laws is because its likely they have been using neural networks to control patriot missles and the early predator drones since the '90s
>>58601260
>L3 cash, aka cache
>>58601521
>>58601489
Don't have such shitty law here.
>>58601489
Gonna need sauce on your claims. Machine learning via FPGA is one of the few ways we can proceed with artificial intelligence, it would be stupid to ban that.
>>58601489
>>58601521
false
>>58601018
>>58601260
these
>>58601489
>>58601521
Bullshit, both posts.
If you just throw computer resources at your ML project you're probably not doing something interesting.
>>58601574
True.
www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA465592