Me ya faru
Lec ta ba da rahoton cewa Qualcomm ya buɗe rukunin sarrafa jijiya na Hexagon na gaba a ranar 10 ga Satumba don ci gaba da aikin AI mai aiki akan na'urorin hannu. Ƙirar da aka bayar da rahoton tana ƙara Mai Haɓakawa na Element don ayyukan masu canji, yana haɓaka ƙwaƙwalwar ajiya da kashi 50%, kuma yana goyan bayan nau'ikan ƙwararrun masana tare da sigogin biliyan 30, tare da kusan biliyan 3 masu aiki ga kowane alamar da aka samar. Dangane da The Lec, NPU tana goyan bayan tsarin bayanan FP16, INT2, INT4, INT8 da FP8. Qualcomm yayi iƙirarin INT4 prefill aikin na iya zama har zuwa 50% cikin sauri kuma martanin zai iya farawa a cikin daƙiƙa 1.5, kodayake Lec ba ta ba da rahoton gwaji mai zaman kansa na waɗannan alkaluman ba. An yi niyyar NPU don yin aiki tare da Snapdragon's Oryon CPU da Adreno GPU don tafiyar matakai masu yawa da sarrafa AI masu alaƙa da zane. Rahoton ya ce Qualcomm zai ba da ƙarin cikakkun bayanai a taron koli na Snapdragon a Amurka daga 22 zuwa 24 ga Satumba. Majiyar ba ta gano takamaiman wayar salula ba, kwanan wata da aka ƙaddamar, wadatar tallace-tallace, farashi ko tabbatar da sakamakon tambari na ɓangare na uku.
Lec ya ba da rahoton cewa Qualcomm ya bayyana Hexagon NPU na gaba a ranar 10 ga Satumba, yana sanya shi azaman kayan aikin hannu don tsarin AI waɗanda ke fassara mahallin, tunani a cikin aikace-aikace da yin ayyuka ga masu amfani. Kayan aikin da aka ruwaito yana ƙara Mai Haɓakawa na Element tare da scalar data kasance, vector da kari na matrix, tare da manufar haɓaka ƙididdiga-samfurin taswira yayin sarrafa amfani da wutar lantarki.
Rahoton ya ce ƙarfin ƙwaƙwalwar ajiya da aka raba shine 50% sama da na ƙirar da ta gabata. Dalilin da Qualcomm ya bayyana shi ne cewa kiyaye yanayin ƙirar, kunnawa da matsakaicin bayanan tensor kusa da na'ura na iya rage canja wuri zuwa DRAM na waje, mai yuwuwar haɓaka riƙe mahallin da canza aiki. Rahoton baya bayar da kwatance mai zaman kansa tare da aiwatar da Hexagon na baya.
An bayar da rahoton cewa, sabon Hexagon yana goyan bayan nau'ikan ƙwararrun ƙwararru tare da sigogin har zuwa biliyan 30, yayin da ke kunna sigogi kusan biliyan 3 a kowace alama. Har ila yau, Lec ya ba da rahoton NAND flash-to-memory management da caching da aka yi niyya don loda abubuwan ƙwararrun ƙwararrun da suka dace kawai a cikin DRAM kuma suna riƙe abubuwan da aka saba amfani da su a cikin cache.
An ba da rahoton NPU tana goyan bayan FP16, INT2, INT4, INT8 da FP8 daidaitattun tsarin. Lec yana ba da da'awar har zuwa 50% cikin sauri INT4 prefill da amsa yana farawa a cikin daƙiƙa 1.5 zuwa Qualcomm. Babu gwaje-gwaje masu zaman kansu, wadatar na'ura, farashi ko takamaiman samfuran mabukaci da aka bayar.
Me ya sa yake da mahimmanci
Idan rahoton da aka bayar na Qualcomm ya isa na'urorin mabukaci kamar yadda aka bayyana, zai iya sa tsarin AI mafi girma da sanin mahallin ya fi dacewa don aiki a cikin gida akan wayoyi. Tsayawa ƙarin bayanan ƙira kusa da na'ura mai sarrafawa na iya rage zirga-zirgar ƙwaƙwalwar ajiya da latency, yayin da cakuda-na-ƙwararrun gine-gine na iya ba da damar yin amfani da ƙwarewa na musamman ba tare da kunna kowane siga ga kowane ɗawainiya ba. Wannan yana da mahimmanci don keɓantawa, amsawa da kuma amfani da layi, amma fa'idar fa'idar ta kasance ba a tantance ba har sai an sami na'urori, samfura da gwaje-gwaje masu zaman kansu.
Sanarwar ta yi niyya ga matsananci na tsakiya a cikin wayar hannu ta AI: gudanar da samfura masu iya aiki a cikin ƙayyadaddun iyaka akan bandwidth ƙwaƙwalwar ajiya, amfani da baturi da latency. Babban wurin tafkin ƙwaƙwalwar ajiya da zaɓin kunna cakuda-na-ƙwararrun kayan aikin na iya rage adadin bayanan da aka motsa yayin ƙaddamar da gida, idan tsarin gine-ginen Qualcomm ya yi kamar yadda aka bayyana.
Sarrafa gida na iya samun fa'idodi masu amfani saboda ana iya sarrafa wasu mu'amala ta waya ba tare da aika kowace bayanai zuwa sabis na nesa ba. Koyaya, tushen baya kafa garantin keɓantawa, aikin layi, haɓaka baturi ko aiki a cikin aikace-aikace na gaske. Waɗancan sakamakon sun dogara da software, matsawa samfurin, iyakoki na zafi da aiwatar da wayar hannu.
Haɗin kai da aka ruwaito tare da CPU da GPU yana nuna Qualcomm yana kula da AI mai aiki azaman aikin dandamali maimakon keɓantaccen fasalin haɓakawa. Muhimmancinsa zai dogara ne akan ko masu haɓakawa za su iya samun dama ga kayan aikin yadda ya kamata da kuma ko masu yin waya suna jigilar samfura da aikace-aikacen da suke amfani da shi.
Ingantacciyar hanyar sadarwa: Yadda A zahiri yake Aiki
Bincika fasahar da ke bayan wannan ci gaban ta hanyar mu'amala.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Abin kallo na gaba
Shaida mai ma'ana ta gaba za ta kasance bayyanannun taron koli na Qualcomm's Snapdragon da na'urori na ƙarshe ta amfani da NPU. Duba don dandamalin guntu mai suna, samfura masu tallafi, kayan aikin haɓakawa, ainihin ma'aunin kan na'ura, ci gaba da amfani da wutar lantarki da kuma ko lokutan amsa da'awar sun shafi kammala ayyukan aiki na wakili maimakon ƙayyadaddun zanga-zangar. Ba a rubuta samuwa, farashi da rarraba yanki a cikin rahoton.
Qualcomm ya ce za a bayyana ƙarin cikakkun bayanai a taron koli na Snapdragon daga Satumba 22 zuwa 24. Waɗancan bayanan na iya fayyace takamaiman dandamali na Snapdragon, ƙayyadaddun lokacin turawa, wuraren aiki masu tallafi da samun damar haɓakawa.
Gwaji mai zaman kansa yakamata ya tabbatar da ingantaccen aikin INT4 da aka ruwaito, da'awar farawa-dakika 1.5, amfani da wutar lantarki da ci gaba mai dorewa a ƙarƙashin manyan ayyuka masu yawa. Madogararsa ba ta bayar da irin wannan gwajin ba.
Har yanzu ba a san waɗanne wayoyin hannu za su yi amfani da NPU ba, lokacin da za su isa abokan ciniki, abin da za su kashe ko kuma za a kunna duk ƙarfin da aka bayar yayin ƙaddamarwa.