Kwenzekeni
I-Forbes ibika ukuthi i-SemiAnalysis ishicilele i-AgentX, ibhentshimakhi yomthombo ovulekile edlala kabusha izikhathi zomenzeli wekhodi ongaziwa esikhundleni sokusebenzisa ukwaziswa okungaguquki. Ekucushweni okumboziwe, i-SemiAnalysis ithole ihadiwe ye-Nvidia inenzuzo ephindwe kahlanu yokusebenzisa kahle izindleko kune-AMD endaweni ethile yokusebenza, enezikhala ezinkulu kwezinye izitaki zokuphakela ezivulekile. Umthombo onikeziwe awuqinisekisi ngokuzimela ibhentshimakhi noma ukhiqize kabusha imiphumela yawo.
I-Forbes ibika ukuthi i-SemiAnalysis ishicilele i-AgentX ngo-Agasti 24 njengebhentshimakhi yokulandela umkhondo yokuchazwa kwesitayela sokukhiqiza. Esikhundleni sokuthumela ukwaziswa okufanayo kanye nobude bokuphumayo, i-AgentX idlala kabusha ukwakheka kwamaseshini omenzeli wekhodi wangempela, okuhlanganisa imilando emide, ukujika okuningi kanye nokumisa isikhashana ngenkathi umenzeli esebenzisa amathuluzi. I-Forbes ithi i-SemiAnalysis yakha ikhophasi yamaseshini angaphezu kuka-8,000 kanye namathokheni ayizigidi eziyizinkulungwane ezingu-610 kusukela ekubanjweni kwezicelo ze-Claude Code kanye ne-Codex ezibandakanya abasebenzi bayo. I-subset yenguqulo esesidlangalaleni engu-1.0 iqukethe izikhathi ezingu-393 Claude Code futhi ikhishwa ngaphansi kwe-Apache 2.0. Umthombo awunikezi ukuqinisekiswa okuzimele kwalezo zibalo.
I-Forbes ibika ukuthi ukulandelelwa komphakathi kuguqulwa kube amabhulokhi e-hashi ahlanganisiwe weseshini angama-64 amathokheni, alondoloza ubudlelwano besiqalo kuyilapho kususwa ukwaziswa kwangempela kanye nekhodi. I-SemiAnalysis ibike ubude bokufaka obuphakathi kwamathokheni we-142,000, ukukhishwa okuphakathi kwamathokheni angu-444 kanye negebe eliphakathi kwe-3.84-yesibili phakathi kokujika. IForbes iphinde ithi izikhathi ezingu-175 kwezingama-393 ziveze okungenani isikhungo esisodwa esincane. Lezo zici zihloselwe ukuhlola ukuthi ingabe amasistimu wokuphakela ayasigcina futhi asebenzise kabusha isimo senqolobane yenani elingukhiye kuwo wonke amajika, ahambise iseshini kusisebenzi esiphethe leso simo futhi agweme ukucutshungulwa kabusha okungadingekile komongo ophindaphindiwe.
Kwelinye iphuzu lokusebenza okubikiwe, i-Forbes ithi i-SemiAnalysis ilinganise i-Nvidia hardware njengoba ibiza kakhulu ngokuphindwe kahlanu kune-AMD lapho isebenzisa i-GLM 5.3 ngokusebenzisa isitaki sokusebenzisa esivulekile se-SGLang kumathokheni okukhipha angu-150 ngomzuzwana ngomsebenzisi ngamunye. Lesi sihloko sithi ukuqhathanisa kwe-B200 ne-MI355X ye-AMD kubonise inzuzo ye-Nvidia engaba ngu-57% kumathokheni angu-108 ngomzuzwana ngomsebenzisi ngamunye kanye nama-247% kumathokheni angu-141 ngomzuzwana ngomsebenzisi ngamunye. Ku-Qwen 3.5 nge-SGlang, i-Forbes ibika ukuthi i-Nvidia yayingaphambili izikhathi ezingaphezu kuka-20 ngamathokheni angu-90 ngesekhondi ngayinye ngomsebenzisi ngamunye. Lena imiphumela eqondene nokucushwa, hhayi isilinganiso sezingxenyekazi zekhompuyutha zomhlaba wonke.
Imininingwane yomthombo: forbes.com ↗
Kungani kubalulekile
Umbiko uphakamisa ukuthi ukusebenza kwe-AI-inference kwama-ejenti kuncike kakhulu ku-software, ukusetshenziswa kabusha kwenqolobane, umzila kanye nokusetshenziswa kokuqukethwe okude—hhayi nje ukucaciswa kwesisheshisi. Lokho kungase kuthinte ukuthengwa kwengqalasizinda, intengo yamafu kanye nokuncintisana kwe-AMD ne-Nvidia. Okutholiwe kukhawulelwe kuma-hardware ahloliwe, amamodeli, izinguqulo zesofthiwe nokulandelelwa komthwalo womsebenzi.
Ukubaluleka okungokoqobo ukuthi umsebenzi omningi we-ejenti ungachitha isikhathi esiningi kakhulu ukucubungula noma ukusebenzisa kabusha umongo kunokukhiqiza izimpendulo zokugcina. I-Forbes ibika okokufaka kwe-AgentX okumaphakathi kwamathokheni angu-142,000 ngokumelene namathokheni okukhiphayo angu-444 kuphela, iphethini engafani namabhentshimakhi amaningi avamile. Uma ithrafikhi efanayo ivamile emkhiqizweni, amandla okuphinda asebenzise umongo ogcinwe kunqolobane kanye nokuphenduka komzila ngendlela ephumelelayo kungaba nomthelela ezindlekweni nokusabela kwabasebenzeli bekhodi, abasizi bocwaningo namanye amasistimu okusebenzisa amathuluzi. Umthombo onikeziwe awuqinisekisi ukuthi iminonjana imele kangakanani emakethe ebanzi.
I-Forbes ibeka ingxenye enkulu yenzuzo ebikiwe ye-Nvidia kusendlalelo sesofthiwe. I-athikili ichaza ukuphathwa kwenqolobane, umzila, ukuhlela, izikhwebu ezikhethekile kanye namathokheni akhuphukayo owazi umngcele kusitaki se-Nvidia's TensorRT-LLM. I-Forbes ithi ukukhushulwa kwamathokheni kufane nemiphumela yamathokheni agcwele kuzo zonke izinguquko ezihloliwe eziyi-1,087 kumkhondo owodwa we-Qwen 3.5 ngenkathi kunciphisa isikhathi sokucubungula isilinganiso ukujika ngakunye ukusuka ku-185.1 millisecond kuya ku-11.3 millisecond. Uma kunembile, ukuthuthukiswa okunjalo kungenza isisheshisi esidala noma esiqhathanisekayo siqhudelane ngokunciphisa umsebenzi ophindaphindiwe. Baphinde basho ukuthi amazinga ebhentshimakhi angashintsha ngokushesha lapho enza izinguquko zesoftware.
Umbiko ubalulekile emqhudelwaneni ngoba ubekela inselele umbono wokuthi umphakeli wesibili wokusheshisa uzonciphisa ngokuzenzakalelayo isikhundla se-Nvidia ngenani lezingxenyekazi zekhompuyutha kuphela. I-Forbes ithi injini ye-ATOM ye-AMD ishaye uhlelo lwe-GB300 NVL72 olusebenzisa i-vLLM engxenyeni eyodwa ye-Kimi K3 latency curve, nokuthi i-MI355X ye-AMD ifane ne-B200 ku-DeepSeek V4 ne-SGlang ngaphambi kokuthi kamuva i-Nvidia kanye nokwenza ngcono kwe-Inferact kuhlanganiswe. Lokho kuguqulwa kukhombisa amandla we-AMD nokubaluleka kokwamukelwa kwesoftware. I-Forbes iphinde ibike ukuthi i-ATOM inokusetshenziswa okulinganiselwe kokukhiqiza nokuthi i-AMD backend ayizange ibalwe kwezinye izindlela ezihambisana nomxholo we-vLLM, kodwa lezo zimo zesofthiwe zingashintsha.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
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?
Ongakubuka ngokulandelayo
Okulandelayo okubalulekile ukuthi ukulungiselelwa kwe-ATOM ye-AMD nokusekelwa kokusebenza kokuqukethwe okude kwamukelwa kabanzi kusofthiwe evamile njenge-vLLM ne-SGlang. Abathengi kufanele futhi bahlole amanani entengo yesiqalo senqolobane yokukhiqiza, ukusetshenziswa kwememori yokusingatha, umzila weseshini kanye nokubambezeleka kwezinga lokuphenduka. I-SemiAnalysis kulindeleke, ngokusho kweForbes, ukuthi ibuyekeze ibhentshimakhi phakathi nenyanga.
Ukulandelela okubaluleke kakhulu ukuthi igebe elibikiwe liyaqhubeka yini ngemva kokubuyekezwa kwesofthiwe. I-Forbes ithi i-SemiAnalysis ihlele isibuyekezo se-AgentX kungakapheli inyanga, futhi i-athikili iphawula ukuthi ushintsho olukhuphuka ngo-Agasti 21 lwashintsha ukuqhathanisa kwe-DeepSeek V4. Izinguqulo zesikhathi esizayo kufanele zibonise ukuthi amandla e-ATOM ye-AMD afinyelela izitaki zokuphakela ezivame ukusetshenziswa, noma ngabe i-vLLM ne-SGLang zengeza ukwesekwa okwengeziwe kwe-AMD okuvuthiwe kokufana kokuqukethwe okude nokulayishwayo, nokuthi ingabe ukuhola kwe-Nvidia kusala ngemuva kokusetshenziswa kokulungiswa okuqhathaniswayo nhlangothi zombili.
Abathengi bengqalasizinda kufanele bacele abahlinzeki ubufakazi obuqondene nomthwalo wokusebenza kunokuncika ekubalweni kwezibalo zamathokheni ngesekhondi ngalinye. I-Forbes incoma ukuthi kuhlolwe amanani aqhubekayo esiqalo senqolobane ehambisanayo ekukhiqizeni, inani lememori yosokhaya esetshenziselwa ukubuyisela inkumbulo yesisheshisi, nokuthi ingabe umzila weseshini ugcina ingxoxo nomsebenzi ophethe isiqalo saso esigcinwe kunqolobane. I-athikili ibika ukuthi i-SemiAnalysis ithole izinga eliphezulu lokushaywa kwenkumbulo yomkhawulokudonsa ongu-91% ekucushweni okukodwa kwe-B300 kanye no-73% ekucushweni kwe-B200 ngaphansi kwamazinga emali ehlukene, nokusetshenziswa kabusha kwememori yokusingatha. Lezo zilinganiso aziqhathanisi ngokuqondile ngokwanele ukuze kumiswe isimiso esivamile, kodwa zibonisa ukuthi yimiphi imininingwane yokusebenza engathinta izindleko nokubambezeleka.
Ibhentshimark futhi ishiya izinto ezimbalwa ezingaziwa. I-Forbes ithi i-AgentX ithatha indawo yokuqukethwe komkhondo ongaziwa ngamathokheni okwenziwa, okungahlanekezela ukwamukelwa kokuqagela kokuqagela, nokuthi i-SemiAnalysis isebenzise ubude bokwamukela obuthathwe ku-SPEED-Bench esikhundleni sokukukala bukhoma. Ukulandelela kuvela kumahhanisi ekhodi anomongo osindayo, kuyilapho ihhanisi elithambile likhiqiza ukusabalalisa okokufaka okuhlukile. Idizayini evala iluphu ye-AgentX ingase futhi iguqule ingxube yomthwalo womsebenzi njengoba amasistimu asheshayo eqedela izicelo eziningi. Ama-TPU e-Google awekho, futhi kamuva i-Nvidia Rubin hardware kanye ne-MI455X ye-AMD angaphandle kokuqhathanisa okuchaziwe. Izimpendulo zomthengisi, ukuphindaphinda okuzimele kanye nemiphumela kumthwalo wokusebenza wama-ejenti angafaki amakhodi akunikeziwe emthonjeni.