Izinyathelo Zakamuva ze-AMD: Ubambiswano lweCerebras, iMicrosoft Helios, iRyzen 7700X3D, kanye ne-MI300X Benchmarks

Ikhaya » Software » Izinyathelo Zakamuva ze-AMD: Ubambiswano lweCerebras, iMicrosoft Helios, iRyzen 7700X3D, kanye ne-MI300X Benchmarks

Ama-Advanced Micro Devices abelokhu eba yizihloko zezindaba emikhakheni eminingi muva nje, kusukela ekubambisaneni okusha kwe-AI kuya ekuthuthukisweni kwehadiwe yabathengi kanye nesofthiwe. Imizamo yenkampani ihlanganisa ingqalasizinda yesikhungo sedatha, amaprosesa emidlalo, kanye nokusheshiswa kwe-GPU, okubonisa ukusunduza okubanzi kokuqinisa isikhundla sayo endaweni yokubala.

Emasontweni ambalwa edlule kuphela, i-AMD imemezele ukusebenzisana kwezobuchwepheshe ne-Cerebras Systems, yaqinisekisa ukuthunyelwa okukhulu kwepulatifomu yayo ye-Helios rack-scale ne-Microsoft, yabona ukubuyekezwa okuhle kweprosesa yayo ye-Ryzen 7 7700X3D, futhi yathola ukuthuthukiswa kokusebenza kwama-MI300X GPU ayo kusukela kuphrojekthi ye-Unsloth. Lezi zinguquko ziqokomisa umfutho wenkampani kuzo zonke izingxenye zemakethe ezahlukene.

I-AMD kanye ne-Cerebras Zijoyina Amandla Okuthola I-Ultra-Low-Latency AI Inference

I-AMD kanye ne-Cerebras Systems baveze ikhambi lokubona i-AI elingahlanganisiwe elihlanganisa izinhlelo ze-AMD Helios rackscale ne-Cerebras Wafer-Scale Engine. Njengoba kumenyezelwe ku-Advancing AI 2026, lobu budlelwano buhlose ukuletha ukubambezeleka okuphansi kakhulu kwezinhlelo zokusebenza ze-AI ezithuthukisiwe ngenkathi kuthuthukiswa ukugeleza nokusebenza kahle. Ukusethwa okuhlangene kuzobona i-AMD Helios isingatha ukucutshungulwa okusheshayo kwe-high-throughput kanye namafasitela amakhulu omongo, kuyilapho i-Cerebras WSE isheshisa ukukhiqizwa kwamathokheni anamandla e-memory-bandwidth ngokulibaziseka okuncane. Ngokusho kwezinkampani, ipulatifomu ehlanganisiwe ingafinyelela amathokheni aphezulu afinyelela ku-5x ngomzuzwana nge-watt uma kuqhathaniswa nokucushwa kwe-Cerebras kuphela. I-Cerebras ihlela ukufaka i-AMD Helios ezikhungweni zayo zedatha, futhi ikhambi kulindeleke ukuthi litholakale nge-Cerebras Cloud engxenyeni yesibili ka-2026. I-CEO ye-AMD uDkt. Lisa Su waphawula ukuthi ukubheka i-AI kuba ithuba elikhulu lengqalasizinda elidinga izindlela eziguquguqukayo, kuyilapho i-CEO ye-Cerebras u-Andrew Feldman egcizelela isidingo esikhulayo sokubheka okusheshayo kwe-ultra-fast.

I-athikili ehlobene:

Intuthuko Yamuva Yesofthiwe kanye Nezindlela Zobuchwepheshe Ezisafufusa

I-Microsoft Yamukela i-AMD Helios ukuze inikeze i-Azure AI Inference

Kokunye ukunqoba okubalulekile, iMicrosoft ikhethe ukufaka amaseva e-AMD e-Helios rack-scale ezikhungweni zayo zedatha ze-Azure. Lezi zinhlelo zizosetshenziselwa ukunika amandla ukuqagela kwe-AI yemodeli ye-frontier model ye-Microsoft, amakhasimende ayo e-AI, kanye nezinsizakalo ze-Azure AI. Ngaphezu kwalokho, iMicrosoft izokwandisa ukwamukelwa kwayo kwamaprosesa eseva ye-AMD Epyc kanye nama-chip enethiwekhi ye-Pensando ukuthuthukisa ukuxhumana kwe-Azure. I-AMD ilindele ukuqala ukuthumela i-Helios ku-Microsoft nakwamanye amakhasimende engxenyeni yesibili ka-2026. Lesi sivumelwano silandela isivumelwano esifanayo ne-Meta Platforms, sibeka i-AMD endaweni yokubamba isabelo esikhulu semakethe yeseva ye-AI ekhula ngokushesha, okulindelwe ukuthi ifinyelele ku-$838 billion ngo-2030. Lezi zindaba zibe negalelo ekukhuphukeni kwentengo yesitoko ye-AMD ngo-144% kulo nyaka, kanye nabahlaziyi abalindele ukukhula okunamandla kwemali engenayo kumakota azayo.

I-Ryzen 7 7700X3D: I-CPU Yemidlalo Ebalulekile Ngaphansi Kokugqanyiswa

Ngasohlangothini lwabathengi, i-Ryzen 7 7700X3D idunyiswe njenge- CPU yokudlala enenani elihle kakhulu ekuhlolweni okwenziwe ngesandla kwakamuva. Umbuyekezi wakha i-PC encane esebenzisa i-processor, ehlanganiswe ne-PowerColor RX 9070 XT Hellhound, i-16GB ye-DDR5-4800 RAM, kanye ne-2TB Lexar NM790 SSD. Uhlelo lusetshenziswe ukuhlola i-SteamOS ngokumelene ne-CachyOS, kanti umbuyekezi waphawula ukuthi i-7700X3D inikeza ukusebenza okuhle kakhulu kwemidlalo ngentengo encintisanayo. Ngenkathi isipiliyoni sokudlala se-Linux sasivame ukuba sihle, umbuyekezi wavuma ukuthi imisebenzi yobuchwepheshe efana ne-Photoshop ne-Premiere isadinga i-Windows ukuze isebenze kahle kakhulu. Ukusebenza kwe-processor ezihlokweni zemidlalo eziyi-10 kwalinganiswa ngaphambilini, futhi ukusetshenziswa kwangempela kwaqinisekisa isiphakamiso sayo senani eliqinile.

I-athikili ehlobene:

Indlela Yokufaka Nokusebenzisa Amamodeli E-AI Asendaweni Ku-PC Yakho Nge-Ollama

Ama-Benchmark e-AMD MI300X GPU Abonisa Inzuzo Yokusebenza Nge-Unsloth

Ukulungiswa kwesofthiwe kwama-GPU e-MI300X e-AMD nakho kuye kwathuthuka. Iphrojekthi ye-Unsloth, eyaziwa ngokuthuthukisa ukuqeqeshwa kanye nokucabanga ngehadiwe ehlukahlukene, ilethe ukulungiswa kwayo kuma-GPU e-AMD. Emisebenzini yokufunda yokuqinisa isebenzisa i-Llama-3.2-3B GRPO ene-vLLM, i-Unsloth ithole isikhathi sesinyathelo esizinzile samasekhondi angu-1.75 uma kuqhathaniswa namasekhondi angu-2.01 okusetha okujwayelekile kwe-TRL + FA2 + vLLM. I-Unsloth iphinde isekele ukwabelana ngesisindo ne-vLLM ukuze igweme ukuchitha inkumbulo, futhi ibhekana nezinkinga zokunemba ku-LoRA/QLoRA ngokusebenzisa indlela ye-vLLM ye-LoRA ngqo, ivimbele ukukhukhuleka kwesisindo okubangelwa ukuzungeza kwe-bfloat16. Lokhu kuthuthukiswa kwenza ama-GPU e-AMD ancintisane kakhulu ngokulungiswa kahle kanye nemithwalo yemisebenzi yokufunda yokuqinisa.

Kusukela ekubambisaneni kwezikhungo zedatha neCerebras kanye neMicrosoft kuya ekubuyekezweni okunamandla kwe-CPU yabathengi kanye nokuthuthukiswa kwesofthiwe ye-GPU, i-AMD ikhombisa ikhono layo lokuncintisana kulo lonke ububanzi bekhompyutha. Ukugxila kwenkampani ekuqondeni okuhlukanisiwe, izinhlelo ze-rack-scale, kanye nokwenza ngcono isofthiwe okuqhubekayo kuyibeka kahle ezidingweni ezikhulayo ze-AI kanye nekhompyutha esebenza kahle kakhulu. Njengoba kwethulwa imikhiqizo eminingi kanye nokusebenzisana okuhlelelwe ingxenye yesibili ka-2026, umfutho we-AMD awubonisi zimpawu zokwehla kwejubane.