Big Sur: This is the AI computer of Facebook, become open source
The computer that performs calculations of machine learning for Zuckerberg’s social network is now open source, and can draw all its secrets.
Facebook is releasing the hardware design of a server that is now used to ” train ” the software of artificial intelligence, so that other companies can explore this world and to design similar systems. Codenamed Big Sur, Facebook use the server to run their own programs of learning machine, a type of software that learns how to use the user and becomes more effective and better without any further interaction from developers.
On machine learning, they are focusing almost all the big names in the computer industry, including Google, Apple, Microsoft and even NVIDIA. The specific technology, which can be commonly called artificial intelligence is used in various areas: for instance allows to locate things and people analyzing faces and objects in a photo or a video, recognizes words in a sentence dictated and also to detect spam email or any attempts at fraud.
Facebook had already released sections of its code of AI programs in the past, but this is the first time offering its hardware as open-source. Big Sur is based heavily on GPU computing, specifically with NVIDIA Tesla, extremely efficient in these areas compared to a traditional CPU. The server can integrate up to eight high-performance video cards that each consume approximately 300Wh and can be configured to each other in different ways through PCIe connectors.
According to Facebook said the new GPU-based Big Sur is about twice as fast as previously used hardware and the ability to distribute computations across eight different GPU enables engineers to ” increase the speed and size of networks of a Another factor of two. ” Another very interesting feature of Big Sur, due to the use of GPU and CPU traditional, is related to the cooling system needed to keep in check the heat dissipation.
The hardware of Facebook does not require a ” cooling systems special ” or specific infrastructures. By contrast, other high-performance server can generate a lot of heat, with an additional outlay required for specific cooling systems, such as in extreme cases tanks with special fluids in which servers are immersed to keep the temperature of the electronic components under control.
Using Big Sur, Facebook can rely on a system of air cooling rather common (as you can see from the pictures published), giving up special industrial solutions and maintaining very low installation costs and maintenance. Like most hardware Open Compute, the Facebook server is rather simple to structures, with components that are designed to be compatible with the needs of society.
” We have removed the parts that were not used often, and components subject to breakage frequently – such as hard drives and DIMMs – can now be replaced in seconds, ” said the company. Facebook has also sought to simplify the maintenance work, ” coloring ” of green all the elements that can be used to repair or replace a defective part. Even the motherboard can be removed in a matter of seconds.
Big Sur is almost entirely tool-less, with the screwdriver that is intended only to remove the radiator which cools the CPU.
While Big Sur to offer open-source community can be an altruistic move, the other this will allow the company to get advice and support to improve the technology and make it more effective. The company said it will triple its investments in the field of computing through the GPU, so that he may use the machine learning algorithms on a higher number of services and sharpen its reliability on those that already actively exploiting them.
