Inkampani i-MetaX namuhla imemezele ukuthi ama-GPU ochungechunge lwayo lwe-MetaX C aqede ukusekelwa kwe-Day 0 kwemodeli ekhiqiza okuqukethwe ngezindlela eziningi i-MiniMax H3 evela kwa-MiniMax. Ngokusebenzisa isitaki sesofthiwe i-MXMACA esizakhele sona kusukela ekuqaleni kuya ekugcineni, abasebenzisi bangathumela ngokushesha i-MiniMax H3 futhi basebenzise amandla ayo aqinile okudala okuqukethwe ngezindlela eziningi.
Inkampani i-MetaX ithe ama-GPU ochungechunge lwe-MetaX C akhelwe phezu kwe-IP eyinhloko ye-GPU ezakhele yona, anokusebenza kahle kwamandla okuphezulu nokusetshenziswa okubanzi, futhi ahambisana ngokunembile nezidingo zamamodeli amakhulu zokucubungula inani elikhulu kakhulu lamapharamitha kanye nokwenza inference enomongo omude. Uchungechunge lwe-MetaX C lusebenzisa i-hardware architecture esebenza kahle kakhulu eyakhelwe ngaphakathi, luhambisana neplathifomu yokubala i-MXMACA, ngaleyo ndlela lwenza kube nokushintshwa okushesha kakhulu kwemodeli i-MiniMax H3.
Isitaki sesofthiwe i-MXMACA siyisisekelo esibalulekile esenza ama-GPU e-MetaX adlule ekubeni “asebenziseke” abe “asebenza kahle futhi kulula ukuwasebenzisa”. Njengesixhumanisi esiyinhloko phakathi kwamandla okubala e-hardware ne-ecosystem yezinhlelo zokusebenza esezingeni eliphezulu, isitaki sesofthiwe i-MXMACA sihlanganisa lonke uchungechunge, kusukela kuma-driver asezingeni eliphansi, ama-interface asezingeni lomsebenzisi, i-compiler ye-MXCC, ukulungiswa okujulile kwama-operator, kuya ekuxhumaniseni nezinhlaka ezijwayelekile zokuqeqesha ne-inference. Sihambisana ngokwemvelo ne-ecosystem ehamba phambili emhlabeni wonke, sisekela i-PyTorch, i-TensorFlow, i-vLLM, i-SGLang nezinye izinhlaka ze-AI ezijwayelekile ezingaphezu kuka-40, futhi sisekela ukusebenza okuzinzile kwamamodeli e-AI angaphezu kuka-500, ngaleyo ndlela sinciphise kakhulu isikhathi esivamile sokulungisa amamodeli.