ʻOiai e hakakā ana nā mea loiloi ʻoihana me kekahi e wānana i ka manawa o ke ʻano o ka 7nm NVIDIA GPUs mua, makemake ka hoʻokele o ka hui e kaupalena iā ia iho i ka ʻōlelo e pili ana i ka "wikiwiki" o nā ʻōlelo pili pili. I ka makahiki 2022, e hoʻomaka ana nā ʻōnaehana kōkua hoʻokele ikaika e pili ana i ka Orin generation Tegra processor, akā ʻaʻole naʻe e hana ʻia me ka ʻenehana 7nm. ʻIke ʻia e pili ana ʻo NVIDIA iā Samsung, nona ka ʻenehana 8 nm, e hana i kēia mau mea hana.
Nā hoa hana mai ka pūnaewele Ua hiki iā mākou ke loaʻa nā ʻōlelo manaʻo mai NVIDIA e pili ana i ke kaʻina hana ʻenehana i hoʻohana ʻia e hana iā Orin. ʻOiai ʻaʻole hōʻike ka mea hoʻomohala i ke komo ʻana o Samsung i kēia mau kauoha, ua hōʻoia ʻia ka hoʻohana ʻana i ka ʻenehana hana 8nm. Hāʻawi ʻo TSMC i kāna mea kūʻai aku i kahi ʻenehana kaʻina hana 7-, 6- a i ʻole 5-nm, a aia kēia kaʻina ma ka chronological order. No laila, kuhikuhi nā mea āpau i ka launa pū ʻana ma waena o NVIDIA a me Samsung i ka hoʻokumu ʻana i nā kaʻina hana Orin - ʻoiai ʻoi aku ka maʻalahi o ka hoʻolālā ʻana ma mua o ka hypothetical hope-generation GPUs, a ʻo ka mea hope paha e hana ʻia e TSMC. ʻO ke koi kiʻekiʻe no nā lawelawe o kēia hui e koi iā NVIDIA e noʻonoʻo iā Samsung.
ʻO ka hoʻolālā ARM Hercules, kahi e hoʻohana ai nā mea hoʻohana Orin, i hoʻomohala mua ʻia no kahi ʻenehana kaʻina 7- a i ʻole 5-nm, akā ʻo ka ʻenehana 8-nm o Samsung e kokoke loa i nā palena. Ma ka hoʻonohonoho kiʻekiʻe loa, e loaʻa i nā kaʻina hana ʻo Orin he ʻumikūmālua cores computing me Hercules architecture, akā hūnā ʻia nā kikoʻī e pili ana i ka ʻāpana kiʻi.
E like me ka wā ma mua, hiki i nā ʻōnaehana kōkua hoʻokele ikaika o ka pae ʻelua o ka autonomy ke hana me kahi kaʻina Orin hoʻokahi. ʻO ka koho maʻalahi me ka pae hoʻohana mana ʻaʻole ʻoi aku ma mua o 15 W e ʻoluʻolu me kahi kāmela hoʻokahi, ʻo ka pae hana e hiki i ka 36 trillion hana i kēlā me kēia kekona. Hiki i ka hoʻololi kahiko o Orin ke hoʻonui i ka hana i 100 trillion hana i kēlā me kēia kekona, hana me ʻehā mau kāmela i ka manawa like a ʻaʻole i ʻoi aku ma mua o 40 W.
E hoʻonui ka Orin processor duo i ka hana i 400 trillion mau hana i kēlā me kēia kekona me ka hoʻohana mana ʻaʻole ʻoi aku ma mua o 130 W. Ua lawa kēia no nā ʻōnaehana kōkua hoʻokele o ke kolu o ka pae o ka autonomy. Ua hoʻohana nā ʻōnaehana like Xavier i kahi kaʻina hana kiʻi discrete, akā ua hoʻopau ʻia a hiki i 230 W, a ʻaʻole i ʻoi aku ko lākou pae hana ma mua o 160 trillion hana i kekona.
ʻO ka ʻōnaehana poʻokela o ka pae ʻelima o ka autonomy e hui pū i kahi pālua o nā kaʻina hana Orin a me kekahi mau mea hana kiʻi NVIDIA discrete. I kēia hihia, e hoʻokiʻekiʻe ʻia ka pae hana i 2000 trillion hana i kēlā me kēia kekona, akā e hoʻonui ʻia ka hoʻohana mana i 750 W. ʻO kahi ʻōnaehana like e pili ana i ʻelua kaʻina hana Xavier a me ʻelua mau Volta generation GPU i hāʻawi i ka hana ʻaʻole kiʻekiʻe ma mua o 320 trillion hana i kēlā me kēia kekona me ka pae hoʻohana mana o 460 W. ʻO ke kakahiaka nui loa e hoʻoholo ai e hoʻohana nā NVIDIA discrete GPU o kēia papa i ka hoʻomanaʻo HBM o ka hanauna like me ka makahiki o ka hana ʻana, akā ʻo ke kiʻi i kuhikuhi ʻole i kēia.
Source: 3dnews.ru