Kwimeko yokwenza i-intelligence ye-AI ngokukhawuleza, efanelekileyo, kunye nokufumaneka ngokubanzi, i-GPU ziyafumaneka kakhulu. Zifumaneka ekugqibeleni kwe-AI kunye ne-training, kodwa ukhangela inkcazelo lwabo: Yintoni oku kuthetha ngaphandle kwimistimu eyenziwe yi-GPU? Njengoko iinkampani zithanda izisombululo ezininzi ezihlangene ne-hardware emininzi okanye ezininzi, i-CPU ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye z Yintoni i-Rajalakshmi Srinivasaraghavan iye yenza iinkcukacha zayo. I-Researcher kunye ne-engineer enezimali ezininzi kwi-high-performance computing, iye iye iye iye iye iye zithunyelwe ukucacisa ukufikelela kwe-AI kwi-CPUs. Indawo yayo ibandakanya iingcamango zobuchwepheshe kunye ne-community-driven collaboration. "Nge-co-working ne-open-source communities, sinokufumaneka ukuhlaziywa kwe-avant-the-art ezincinciphisa ukusebenza kwi-AI workloads," wathi. "Le nkonzo yandisa i-ecosystem ngokubanzi kwaye iye yenza iqela lethu njengomdlali esisodwa kwi- Enye iziphumo zayo ezininzi ziquka iinkqubo zokusetyenziswa kweCPU. Nangona abaninzi kwimveliso zihlanganisa kwiinkqubo ze-GPU-centric, i-Rajalakshmi ibonise ukuba iinkqubo zokusetyenziswa kwe-CPU zibonisa imiphumela emangalisayo. I-team yayo iye ibonise iipakethe ze-software ezininzi zokusetyenziswa kwe-CPU, okuvelisa iimpumelelo ezininzi zokusebenza. "Sifumana ukunyuka kwe-50% kwimveliso ye-hardware ye-next-generation ngokufanisa kunye nokusetyenziswa kwinkqubo zokusebenza ze-AI ezininzi," ibonise. Le niphumo zibonisa ukuba iinkqubo zokusebenza zokusetyenziswa Ukuphuhliswa kwayo kwiziko zangaphantsi kwe-coding. Yenza ukutshintshwa kwe-Continuous Integration (CI) builds kwiipakethe ezininzi, ukutshintshwa kwe-automating kunye ne-integration. "Ukuphuhliswa oku kwandisa i-reliability kunye nokukhuthaza iinkonzo ukufikelela kwimveliso ye-core," wathi. Ukuphuhliswa kwakhona ngokushesha: iincwadi ze-innovation kunye ne-software optimizations ezincinciphisa iimfuno ze-hardware. Kwi-industry apho i-speed ikakhulukazi i-success, inkqubo yayo yandisa iimveliso ngokukhawuleza kunye ne-intelligent. I-Rajalakshmi inqakraza i-mentorship njenge-part of her work. I-Rajalakshmi iye yandisa iingcali ezintsha, ukutshintshela iinkcukacha ze-technical and strategic needed for CPU optimization. "Ukuvelisa iinkcukacha ezintsha kuboniswe ukuba iingcali zokusebenza ziye kwenziwe kuphela ngexesha elifanelekileyo, kodwa ingxenye ye-trajectory ye-long-term," inqakraza. I-focus yayo kwi-talent yokukhuthaza iindidi ezininzi kunye ne-culture ye-innovation ephilayo, i-asset kwi-competitive space ye-high-performance computing. Umthamo wayo ukufikelela ngexesha elandelayo uye uyakufuneka. Uya kuthatha enye ingxaki ezininzi kwi-industry ngokuphucula kunye nokuqinisekisa ikhodi kwi-simulators ngaphambi kokuzivula i-hardware. Le nqakraza esebenzayo ivumela iindidi ukufumana iingxaki ngokushesha kunye nokuvelisa iinkcukacha ze-software ezisebenza ngokupheleleyo kunye ne-hardware entsha. "Ukuza ukuba i-software stack yethu iyasebenza ngokupheleleyo kunye nokukwazi ukuvelisa ngosuku lokuqala," wathi. Amayunithi ezincinane bakwazi ukufumana oku, kwaye ukunciphisa iingxaki zokusebenza ngexesha elandelayo kwinkqubo ezilandelayo. Ukuphendula kwizifundo kunye neeyunivesithi. Amaphepha efana ne-"Modeling Matrix Engines for Portability and Performance" kunye ne-"A Matrix Math Facility for Power ISA Processors" ibonise inkxaso yayo ekubeni isisombululo se-theory kunye nezicelo ezisebenzayo. Ngokuchasene nezifundo ze-research kunye ne-industry, uqhuba i-dialogue engaphezulu kwi-AI efficiency. Ukukhangisa ngexesha elandelayo, uye bakwazi ukufumana izixazululo ezininzi ezihlangene ne-GPU. Uyakwazi ukufikelela kwi-hardware kunye ne-software. "Nge-tracking ngokugqithisileyo iintlobo ze-industry kunye nokukhangisa iintlobo, sinikezela iinkonzo ezihlangeneyo kwi-hardware kunye ne-software," uye wathi. Kwakhona, umdla elandelayo we-AI iya kufuneka i-efficiency eyenziwe ngalinye kwinqanaba, ukunika i-platforms eziqhelekileyo ze-optimization iinkonzo ezininzi. Ukukhuthaza ukufikelela kwi-CPU kunye nezinye i-non-GPU architectures, ayikho inkqubo yokubhalisa; yindlela yokwakha i-AI ecosystems ezininzi ezinzima, ezinzima, kunye nokufumaneka. Njengoko ke, "Iziqonga lokuphumelela kwihlabathi kwihlabathi ye-AI ibekwe ukwakha izisombululo ezintsha kunye ne-hardware kunye ne-scale kunye ne-demand." Kwi-GPU yehlabathi, i-Rajalakshmi ibonise ukuba i-innovation epheleleyo ibonelela kwiindawo ezidlulileyo. I-Rajalakshmi ibonise i-CPU yayo, ibonise iimfuneko ezintsha zokusetyenziswa kwe-AI kwaye ibonise ngexesha elandelayo apho i-AI ibonakalisa kakhulu, efanelekileyo, kunye nokugcina. Le nqaku lithunyelwe njenge-release ka-Kashvi Pandey phantsi kwe-HackerNoon's Business Blogging Program. Le nqaku lithunyelwe njenge-release ka-Kashvi Pandey phantsi kwe-HackerNoon's Business Blogging Program.
