Meta Platforms, the mum or dad firm of Fb, is reportedly planning to deploy a brand new iteration of its customized chip into its knowledge centres, as per an inside doc reviewed by information company Reuters.
The chip, which was introduced final yr, is part of Meta’s efforts to bolster its synthetic intelligence (AI) capabilities. With this, the corporate goals to minimize its reliance on Nvidia chips and curb mounting bills related to AI improvement.
The transfer comes amid Meta’s purpose to combine generative AI merchandise throughout its platforms akin to Fb, Instagram, WhatsApp, and {hardware} like Ray-Ban good glasses.The corporate has been investing considerably in specialised chips and knowledge centre reconfigurations to run these applied sciences.
What Meta has to say
A Meta spokesperson confirmed plans to begin manufacturing of the up to date chip in 2024, saying that the chips complement commercially obtainable GPUs, which Meta continues to obtain in massive volumes.
“We see our internally developed accelerators to be extremely complementary to commercially obtainable GPUs in delivering the optimum mixture of efficiency and effectivity on Meta-specific workloads,” the spokesperson was quoted as saying.
The report stated {that a} profitable deployment of its personal chip on the scale at which Meta operates might minimize tons of of thousands and thousands of {dollars} in annual power prices and billions in chip buying prices.
Meta to buy 350,000 Nvidia GPUs
Earlier this yr, it was reported that the corporate plans to amass roughly 350,000 flagship “H100” processors from Nvidia by the tip of the yr. The brand new chip, internally dubbed “Artemis,” focuses on inference, contrasting with GPUs primarily used for coaching AI fashions.
Whereas Meta’s earlier makes an attempt confronted challenges, consultants anticipate that an inference chip like Artemis might improve effectivity in processing advice fashions in comparison with energy-intensive Nvidia processors, doubtlessly resulting in substantial price financial savings.
The chip, which was introduced final yr, is part of Meta’s efforts to bolster its synthetic intelligence (AI) capabilities. With this, the corporate goals to minimize its reliance on Nvidia chips and curb mounting bills related to AI improvement.
The transfer comes amid Meta’s purpose to combine generative AI merchandise throughout its platforms akin to Fb, Instagram, WhatsApp, and {hardware} like Ray-Ban good glasses.The corporate has been investing considerably in specialised chips and knowledge centre reconfigurations to run these applied sciences.
What Meta has to say
A Meta spokesperson confirmed plans to begin manufacturing of the up to date chip in 2024, saying that the chips complement commercially obtainable GPUs, which Meta continues to obtain in massive volumes.
“We see our internally developed accelerators to be extremely complementary to commercially obtainable GPUs in delivering the optimum mixture of efficiency and effectivity on Meta-specific workloads,” the spokesperson was quoted as saying.
The report stated {that a} profitable deployment of its personal chip on the scale at which Meta operates might minimize tons of of thousands and thousands of {dollars} in annual power prices and billions in chip buying prices.
Meta to buy 350,000 Nvidia GPUs
Earlier this yr, it was reported that the corporate plans to amass roughly 350,000 flagship “H100” processors from Nvidia by the tip of the yr. The brand new chip, internally dubbed “Artemis,” focuses on inference, contrasting with GPUs primarily used for coaching AI fashions.
Whereas Meta’s earlier makes an attempt confronted challenges, consultants anticipate that an inference chip like Artemis might improve effectivity in processing advice fashions in comparison with energy-intensive Nvidia processors, doubtlessly resulting in substantial price financial savings.
