The first supercomputer to break the TeraFLOP barrier was Intel's ASCI Red in 1996. Building it cost more than $46 million (about $71 million in 2017). Today, we can have fifty times that processing power on our desks for just over $3,000. An AMD Ryzen 7 and two Vega Frontier Edition cards form the heart of such a system.
What Can $3,100 Buy?
What can you buy with $3,100? The answer is more complicated than we might think, because we all have different tastes and needs. Travel, home repairs, car payments, settling debts, replacing appliances, investing… the options are many. Buying a computer definitely qualifies, but there's something else. Those $3,100 buy an impressive amount of processing power, especially when compared with systems built at the beginning of the century. According to Carlos E. Pérez, $3,100 equals 50 TeraFLOPs (FP16), and you can have it on your desk today.
Inside the Build
The big question is “what for?” Well, if your work involves artificial intelligences, neural networks, algorithms, and deep learning, you need all the firepower you can get. AMD offers it through its latest hardware, starting with an eight-core Ryzen 7 1700 and two Radeon Vega Frontier Edition cards, optimized for this type of work (not for gaming; the consumer version will debut soon). The system is backed by 32 GB of DDR4-2400 RAM (not ideal, since Ryzen prefers higher frequencies), an SSD (525 GB) and a hard drive (5 TB), an X370 motherboard (which enables overclocking and dual PCIe x8), and a giant 1000-watt power supply to feed it all. The Rosewill case costs less than $50.
The Hardware Arms Race
Of course, hardware is an arms race. Within 18 months (or less), $3,100 will buy more TeraFLOPs. AMD has already launched EPYC and Threadripper, Intel has its Core X and Xeon, and don't forget Nvidia with its Tesla series. I insist that computers are experiencing one of their best moments. If you want new hardware, don't hesitate.
For the full story, check out the original build guide: Official site:
