Kamaʻilio ʻo AMD CTO iā Chiplet: Ke hiki mai nei ka manawa o ka hoʻopaʻa ʻana i ka photoelectric co-sealing

Ua ʻōlelo nā luna o ka ʻoihana chip AMD e hiki ke hoʻolako ʻia nā kaʻina hana AMD e hiki mai ana me nā accelerators domain-specific, a ua hana ʻia kekahi mau accelerators e nā ʻaoʻao ʻekolu.

Ua kamaʻilio ka Hope Pelekikena Nui ʻo Sam Naffziger me AMD Chief Technology Officer Mark Papermaster i kahi wikiō i hoʻokuʻu ʻia i ka Pōʻakolu, e hōʻike ana i ke koʻikoʻi o ka standardization chip liʻiliʻi.

“ʻO nā mea hoʻolele kikowaena kikoʻī, ʻo ia ke ala maikaʻi loa e loaʻa ai ka hana maikaʻi loa no kēlā me kēia kālā no ka watt.No laila, pono loa ia no ka holomua.ʻAʻole hiki iā ʻoe ke hana i nā huahana kikoʻī no kēlā me kēia wahi, no laila ʻo ka mea hiki iā mākou ke hana i kahi kaiaola puʻupuʻu liʻiliʻi - ʻo ia ka hale waihona puke, "wahi a Naffziger.

Ua ʻōlelo ʻo ia i ka Universal Chiplet Interconnect Express (UCIe), kahi maʻamau ākea no ka kamaʻilio Chiplet i puni mai kona hoʻokumu ʻana i ka hoʻomaka ʻana o 2022. Ua lanakila ʻo ia i ke kākoʻo ākea mai nā mea pāʻani ʻoihana nui e like me AMD, Arm, Intel a me Nvidia. e like me nā hōʻailona liʻiliʻi ʻē aʻe.

Mai ka hoʻokuʻu ʻana i ka hanauna mua o nā kaʻina hana Ryzen a me Epyc ma 2017, aia ʻo AMD i ka mua o ka hoʻolālā chip liʻiliʻi.Mai ia manawa, ua ulu ka hale waihona puke o ka hale o Zen o nā ʻāpana liʻiliʻi e hoʻokomo i nā helu helu lehulehu, I/O, a me nā pahu kiʻi kiʻi, hoʻohui a hoʻopili iā lākou i loko o kāna mea kūʻai aku a me nā mea hana kikowaena data.

Hiki ke ʻike ʻia kahi hiʻohiʻona o kēia ala ma AMD's Instinct MI300A APU, i hoʻokuʻu ʻia ma Dekemaba 2023, Hoʻopili ʻia me 13 mau pahu liʻiliʻi liʻiliʻi (ʻehā mau pahu I/O, ʻeono mau pahu GPU, a me ʻekolu mau puʻupuʻu CPU) a me ʻewalu mau pahu hoʻomanaʻo HBM3.

Ua ʻōlelo ʻo Naffziger i ka wā e hiki mai ana, hiki i nā maʻamau e like me UCIe ke ʻae i nā pahu liʻiliʻi i kūkulu ʻia e nā ʻaoʻao ʻekolu e ʻike i ko lākou ala i loko o nā pūʻulu AMD.Ua haʻi ʻo ia i ka pilina kiʻi kiʻi kiʻi - he ʻenehana e hiki ke hōʻoluʻolu i nā bottlenecks bandwidth - me ka hiki ke lawe i nā ʻāpana liʻiliʻi ʻekolu i nā huahana AMD.

Manaʻo ʻo Naffziger ʻaʻole hiki ke hoʻokō ʻia ka ʻenehana me ka ʻole o ka pilina haʻahaʻa.

"ʻO ke kumu āu e koho ai i ka hoʻohui optical no ka mea makemake ʻoe i ka bandwidth nui," wehewehe ʻo ia.No laila pono ʻoe i ka ikehu haʻahaʻa no ka hoʻokō ʻana i kēlā, a ʻo kahi pahu liʻiliʻi i loko o kahi pūʻolo ke ala e loaʻa ai ka haʻahaʻa haʻahaʻa.Ua hoʻohui ʻo ia ua manaʻo ʻo ia ka neʻe ʻana i ka co-packaging optics "e hiki mai ana."

I kēlā hopena, ke hoʻomaka nei kekahi mau mea hoʻomaka kiʻi kiʻi silika i nā huahana hiki ke hana i kēlā.ʻO Ayar Labs, no ka laʻana, ua hoʻomohala i kahi puʻupuʻu photonic kūpono UCIe i hoʻohui ʻia i loko o kahi prototype graphics analytics accelerator Intel i kūkulu ʻia i ka makahiki i hala.

Inā ʻike ʻia nā ʻāpana liʻiliʻi ʻekolu (photonics a i ʻole nā ​​ʻenehana ʻē aʻe) i ko lākou ala i nā huahana AMD.E like me kā mākou i hōʻike mua ai, ʻo ka standardization kekahi o nā pilikia he nui e pono e hoʻokō ʻia e ʻae i nā chips multi-chip heterogeneous.Ua noi mākou iā AMD no ka ʻike hou aku e pili ana i kā lākou hoʻolālā chip liʻiliʻi a e hōʻike iā ʻoe inā loaʻa iā mākou kekahi pane.

Ua hoʻolako mua ʻo AMD i kāna mau ʻāpana liʻiliʻi i nā mea hoʻokūkū hoʻokūkū.ʻO ka ʻāpana Kaby Lake-G o Intel, i hoʻokomo ʻia ma 2017, hoʻohana i ka Chipzilla's 8th-generation core me ka AMD's RX Vega Gpus.Ua ʻike hou ʻia ka ʻāpana ma ka papa NAS o Topton.