According to The Information, Apple is working on an inference machine with two or four M8 Ultra chips, aimed at companies that want to host their models themselves.
Apple has not sold server hardware since the discontinuation of the Xserve in 2011. According to information published on September 16 by The Information and relayed by MacRumors, the company is studying a return to this market with an artificial intelligence server based on two or four M8 Ultra chips still to come.
The target would be companies that want to run models on their own infrastructure, with an emphasis on inference — the phase when an already trained model produces its responses — rather than training. To get processors to talk to each other, Apple would look at NVLink Fusion, the hardware and software brick for data exchange between chips offered by Nvidia.
Nothing is set in stone: neither the use of Nvidia, nor the product itself. No launch is expected before 2029 and the project can be abandoned or continued without Nvidia's technology. The mere fact that it is being considered illustrates the warming of relations between two companies that have long been at loggerheads.
Read the full article on the original source
Go to the source →