Meta começa a testar primeiro chip interno de treinamento de IA

[{"selector":"#anim-3ef625ed-6ed7-4a81-bb03-54a94539d112","keyframes":{"opacity":[0,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-1e00f84a-07af-49c3-ab81-b9219a00b619","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-f6e3c3bb-1745-4bde-952b-429309ddbb5d","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-bcfd3993-96a7-44dd-b320-13be67c36148","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] Unsplash

[{"selector":"#anim-1ac017ad-b528-474d-9969-e9fcf88cbc3d","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-a8a49aeb-a9da-46c9-954f-51b29ec413fa","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-7f252cac-bc0f-4127-804a-0f9f241f34b4","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-142d0b98-cdf5-4279-938d-ae18cee388e8","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-dd6957c2-40ca-437e-8fe3-837f0201087a [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(56.591796779002955%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] A Meta está testando seu primeiro chip interno para treinar sistemas de inteligência artificial, buscando reduzir a dependência de fornecedores como a Nvidia Unsplash

[{"selector":"#anim-a10e9f57-7970-4d0d-a13f-5152dfaf020a","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-27f28aac-285c-4741-a6d5-1f92b5c9293e","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-e6c3c5cd-128d-43e1-970d-c02df0c42626","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-8e3c120c-27fd-46c9-8a99-cb912613ccb2","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] O chip faz parte da estratégia da empresa para diminuir custos de infraestrutura enquanto investe em IA Unsplash

[{"selector":"#anim-ef92a27e-3442-42f8-9436-5c4af81e1c94","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-f9a80c69-2808-45f3-b81f-c35537e305f4","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-3c03bec2-0a5f-4367-894c-a1d2badc3ab6","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-da395e04-441e-4b68-885b-328ec43302c5","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-6fc6a209-27d1-48fa-a0a8-db142e0a1bee [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-7.812499744007886%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] Desenvolvido com a TSMC, o chip é um acelerador dedicado, o que pode torná-lo mais eficiente do que GPUs convencionais Unsplash

[{"selector":"#anim-c5de5c0a-83b5-403e-95ad-715172a8c44f","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-297c17b9-57e3-4bd1-93ab-643b01b37f4f","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-699339fb-c348-4253-bd8f-80cfb804611e","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-7e87b7a7-896e-4ed1-b44f-3a415ac4d9fa","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-fb52d2b4-bbb5-4649-8b74-b14858ca9b70 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-32.97668163495715%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] A Meta já usa chips próprios para inferência em sistemas de recomendação do Facebook e Instagram e quer expandir para IA generativa Unsplash

[{"selector":"#anim-bba663ac-c175-4f1b-a12a-818844351b8c","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-ddbb60cc-7322-4d99-9d59-74cf20a2e22e","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-acbbff85-122a-4be5-9b0f-7853748807b7","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-bb3eab52-da4d-4dd3-872a-c26483981dbe","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] A empresa já teve tentativas anteriores frustradas com chips internos, mas considera o desenvolvimento atual mais promissor Unsplash

[{"selector":"#anim-c601e8a9-2680-4a0a-bf64-05cd5a1f1ac7","keyframes":{"opacity":[0,1]},"delay":0,"duration":1500,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-d04873c2-f865-4105-8775-5852ee2221e8","keyframes":{"opacity":[0,1]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-f8c22739-3368-43b6-a197-1c80ffdf3b23","keyframes":{"transform":["scale(0.3333333333333333)","scale(1)"]},"delay":0,"duration":500,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}] [{"selector":"#anim-e83dd1a3-edea-417c-914e-521576c12bc2","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-856b3340-7c2d-4621-9429-0f59ed0ac8f9 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-36.65364572666995%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] Apesar do novo chip, a Meta continua sendo uma grande cliente da Nvidia e acumula GPUs para treinar seus modelos de IA Unsplash

CNNBRASIL.COM.BR

[{"selector":"#anim-59cefdc0-838e-4573-a38b-4f07ce25472d","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-5f56184d-e89f-45a7-a0d5-62667a30c803","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-648a445f-468a-47be-b930-be5c621d22b3","keyframes":{"opacity":[0,1]},"delay":0,"duration":1000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-43944230-ffa5-47af-ba16-5f0bd808c2e4","keyframes":{"transform":["translate3d(0px, 2132.63615%, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-eb061e2b-03ab-4edf-a5ad-c5d952437f1d","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-3940ccab-bf54-4dca-a573-9af16a57193b","keyframes":{"transform":["translate3d(-117.84387%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] Unsplash leia mais em