Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124

Such distributed computing makes sense because “AI computing can and should be distributed at the edge, placed on smaller platforms closer to the user environment,” he said. Benjamin Leecomputer programmer and engineer at the University of Pennsylvania, in correspondence with Ars. “This approach can result in much lower workloads on the team because the simulations require fewer GPUs, unlike studies that require thousands of GPUs to work together,” he said.
However, AI simulation tasks can be as diverse as text questions and answers, software code generation, and multitasking conversations—each with different computational requirements and performance expectations, Lee cautioned. Therefore it will be important to ensure that the computing nodes can provide the necessary service for each task, as well as to maintain network connectivity between the nodes.
Lee also questioned whether it is necessary to scale data centers down to “the granularity of a few GPUs” to reduce their burden on the power grid. He said replacing 20-megawatt conventional data centers with 1-gigawatt hyperscale data centers would be equally beneficial.
Then there is the issue of security. XFRA units that are spread across the region may be more vulnerable to data security threats than centralized data centers. “A lot of lateral attacks want to be close to the machine, which the data center can control,” Lee said. “Distributed GPUs in their own homes are very difficult to secure.”
Thieves may also see XFRAs on the side of buildings as a tempting target, as the Nvidia GPUs inside can sell for around $10,000. Several comment thread on Reddit they have already considered this, while some commentators have suggested that they may be interested in obtaining integrated products for themselves as citizens. “Of course, there is a risk of losing real equipment due to theft,” Lee said.
Any potential benefits and challenges will become more apparent during the SPAN pilot. But at the moment Silicon Valley is buzzing orbital data centers and Ocean-going AI data centersdata center nodes located in rural areas can be very robust – to the extent that landlord associations can identify them.