I-KEXINT Mellanox MMS4A00 Iyahambisana ne-1.6T OSFP224 2xDR4 DR8 XDR InfiniBand Transceiver Dual MPO-12/APC 1310nm 500m IHS Mod

Incazelo Yomkhiqizo

I-Mellanox MMS4A00 Iyahambisana ne-1.6T OSFP224 2xDR4/DR8 8x200G PAM4 InfiniBand XDR Optical Transceiver Module (SMF, 1310nm, 500m, Dual MPO-12/APC, DOM, IHS/Closed Finned Top)

I-NVIDIA/Mellanox MMS4A00 (980-9IAH1-00XM00) I-OSFP Optical Transceiver ehambisanayo iyimojula ye-OSFP IHS/Closed Finned Top enamachweba amabili enikeza amanethiwekhi e-InfiniBand XDR angu-2x 800Gb/s (1.6T). Iqukethe ukwakheka kwemodi eyodwa, okusekelwe ku-silicon photonics okuneziteshi eziyisishiyagalombili ezifanayo, okusekela ukudluliselwa kwe-2×DR4/DR8. Imbobo ngayinye ye-800G iphela ngesixhumi se-MPO-12/APC esineziteshi ezine. Imojula isebenzisa ukuguqulwa kukagesi kwe-200G-PAM4 futhi inikeza ukuxhumana okuthembekile okufushane okufika kumamitha angu-500 ngaphezu kwemicu eyisishiyagalombili yemodi eyodwa.

Yakhelwe nge-Broadcom 3nm DSP, le transceiver iqinisekisa amazinga amaphutha e-bit aphansi kakhulu kanye nokulethwa kwedatha okuzinzile, okusheshayo ekusetshenzisweni okuphezulu. Idizayini yayo ye-IHS/Closed Finned Top ivumela ukuphathwa okusebenzayo kokushisa kwamaswishi e-Quantum-X800, okusekela kokubili ukucushwa komoya kanye nokuphola koketshezi. Ihambisana kahle nezinhlelo ze-NVIDIA B300 kanye ne-DGX, le module inikeza ukuxhumana kwe-GPU-to-GPU kanye ne-GPU-to-storage okukalayo, okubambezeleka okuphansi, okulungele ukuxhumana kwe-AI, HPC, kanye nezindawo ezinkulu zokuxhumanisa izikhungo zedatha, ivula amandla aphelele ezindwangu zekhompyutha ezisebenza kahle kakhulu.

Qaphela: Ama-transceiver "aphezulu" manje abizwa ngokuthi "i-Integrated HeatSink" (IHS) njengoba ama-cooling fin ehlanganiswe phezulu kwensimbi yesixhumi se-OSFP futhi avame ukusetshenziswa kumaswishi apholiswe ngomoya. Ama-transceiver "aphezulu" manje abizwa ngokuthi "i-Riding Heat Sink" (RHS) njengoba ama-cooling fin ehamba phezu kwekhadi lenethiwekhi noma amakheji angaphakathi ezinhlelo. Lokhu kusetshenziselwa kokubili izinhlelo zokusebenza ezipholiswe ngomoya kanye nezipholiswe uketshezi ze-DGX, ConnectX-7, kanye ne-BlueField-3 DPUS.

Okuvelele Komkhiqizo

I-Broadcom 3nm DSP Chip eyakhelwe ngaphakathi, Ukusetshenziswa kwamandla okuphezulu okungu-25W

Ubuchwepheshe obusekelwe ku-SiPh bamandla aphansi, izindleko kanye nobuningi obuphezulu

Kuqinisekiswe ngo-100%, kuyahambisana nama-NVIDIA Quantum-X800 Switches

Yakhelwe i-DGX GB300/B300 Architectures

I-8x 200G-PAM4 kagesi kuya ku-Dual 4x 200G-PAM4 Optical Parallel

Isebenzisana ne-OSFP224-DR4-800G-FL ye-2x 800G NIC

Ihambisana ne-Hot Pluggable OSFP MSA, IEEE 802.3dj kanye ne-RoHS

Ukuphepha Kwelaser Yekilasi 1 Kuyahambisana Nemithetho kanye Ne-DDM Esekelwayo

Imininingwane

Izixazululo Zokuxhumana

1.6T kuya ku-2x 800G ye-DGX B300

Ikhadi le-Switch-Adapter eliyi-1.6T kuya ku-2x 800G

1.6T kuya ku-1.6T ye-Switch-Switch

Izici

Ukuthuthukisa Ukusebenza Kweqembu Le-B300 AI ngamaModule e-1.6T OSFP224

Ku-architecture ye-NVIDIA B300, i-1.6T OSFP optical module inikeza ukuxhumana kwe-bandwidth ephezulu, i-low-latency phakathi kwama-switches e-Quantum-X800 InfiniBand kanye namaseva e-DGX B300, okuqinisekisa ukuxhumana okuzinzile, okushesha kakhulu, okusheshisa ukuqeqeshwa kwe-AI okusatshalaliswa okukhulu, kanye nokuthuthukisa ukusebenza kahle kweqembu lonke. Funda kabanzi mayelana nesisombululo se-B300 InfiniBand.

Kuhlolwe ngo-100% yi-NVIDIA Environment

Amamojula e-FS 1.6T OSFP DR8 ahlolwa kumadivayisi okuqala e-NVIDIA Quantum-X800 ukuqinisekisa inhlanganisela ephelele yokusebenza komkhiqizo kanye nokuthembeka. Kubandakanya ukuhambisana kwehadiwe (ukuxhuma, ukukhipha i-plug, ukuqala kabusha), ukuhambisana kwesofthiwe (ukuxhumana, amapharamitha). Hlola ukuhambisana neThuluzi lethu Lokuhambisana Ne-Optics-to-Device.

Ubuchwepheshe obusekelwe ku-Broadcom 3nm DSP kanye ne-SiPh bokusebenzisa kahle amandla ngendlela esezingeni eliphezulu

Isebenza nge-chip ye-Broadcom ye-3nm DSP kanye ne-silicon photonics, le transceiver ye-1.6T XDR inikeza ukusetshenziswa kwamandla okuphansi okungu-25W (ukonga amandla okungaphezu kuka-25%) kanye ne-6E-12 BER eqinile (ene-FEC Enabled). Ivumela ukuminyana kwe-port okuphezulu, izidingo zokupholisa ezincishisiwe, kanye nezindleko eziphansi zobunikazi.

I-XDR vs. NDR: Ukuphumelela Kokusebenza Kahle Kakhulu Kwe-AI Cluster
Isixazululo se-XDR esine-transceiver engu-1.6T sithuthukisa ukusebenza kahle kwamandla eqembu ngokusebenzisa i-bandwidth density ephezulu kanye nokwakhiwa kwezixhumanisi okulungiselelwe. Ekulungiselelweni kwama-node angu-512, lesi sixazululo senza kube nomphumela ongenakuqhathaniswa, ukukhuliswa okuthuthukisiwe, izidingo zamandla nesikhala ezincishisiwe kakhulu kanye nokuncishiswa okungu-40.52% kwezindleko eziphelele.

Ukuhlolwa Okuphelele Kukhulisa Ukuthembeka

Ufaneleke ngenqubo eqinile ngemishini ethuthukisiwe ukuqinisekisa ukuthi uthola ama-optics asezingeni eliphezulu futhi athembekile.

Umphumela Wokuhlola Othembekile

Ukuhlolwa kobungcweti kuqinisekisa ukukhanya okuzinzile nokuthembekile okunokuhambisana kwehadiwe nesofthiwe okuqinisekisiwe, okuhlanganisa ukuxhumana, i-BER kanye nemingcele yokusebenza eyinhloko.

Kungani Sikhethe

Iphrofayela Yenkampani

• Unyaka ususunguliwe
  --
• Uhlobo lwebhizinisi
  --
• Izwe / Isifunda
  --
• Imboni Eyinhloko
  --
• Imikhiqizo eyinhloko
  --
• I-Enterprise Lebird
  --
• Inani labasebenzi
  --
• Inani lokukhipha lonyaka
  --
• Thumela Imakethe
  --
• Amakhasimende Ahlanganisiwe
  --