I-Inferx X1 Edget Accelerator yenzelwe ukucubungula imithwalo ye-Real-Time Winloads edinga ukusekelwa kwe-bandwidth ephezulu yamamodeli wokufunda ajulile asebenza ngosayizi omncane we-batch ngesikhathi sangempela. Imithwalo ejwayelekile yemisebenzi inamanethiwekhi ajulile anamamephu amaningi wesici kanye nezinhlobo eziningi zokusebenza, zichaza inkampani. Bangase futhi babe nemigomo yokunemba okudingayo okudinga ukusetshenziswa kwemikhawulo exubile, kufaka phakathi int8, int16 ne-BF16. I-accelerator ivumela ukuhlanganiswa phakathi kwezendlalelo futhi yenzelwe futhi usayizi ophansi we-latch batch (b = 1 Ukuhlobisa ukucubungula) okuvame ukudingwa yile misebenzi.
I-accelerator isekela izakhiwo ze-X86 nengalo kanye nokukhethwa kwe-OS. Ixhasa izinhlobo zekhamera, i-Ir, i-Ultrasonic ne-Sensor Sensor yokufaka kanye ne-Ethernet, i-USB kanye ne-Wi-Fi comms.
I-X1 Desnamic Tensor Persor processor i-ARRAY yenzelwe ukusekela amamodeli akhona futhi azayo we-AI / ML izibuyekezo. I-accelerator yokwakha ivumela ukusekelwa kokucubungula izinhlobo eziningi zedatha kufaka phakathi amakhamera we-resolution high.
Ngokungeziwe kumayunithi we-processor array array kanye ne-12mbyte ye-on-chip sram, ukwakhiwa kwe-X1 kufaka phakathi ukuxhumeka ku-DRAM ye-LPDDR4 yangaphandle yesisindo semodeli, ukucushwa kanye nokugcinwa kwangaphakathi. Kukhona ne-PC3 / 4 PCIE yokuxhumana kwiprosesa yokusingathwa.
Le nkampani iphinde inikeze i-inferx edge yokuthuthukisa isoftware yokuthuthukisa isoftware nge-compider compioner ne-runtime software. I-Model Compizer Convents Models evezwe ku-tensorflow lite noma i-TorschScript futhi ihlanganise ukusebenza ngqo kwi-X1 accelerator. I-Inferx Trussitime ilawula ukwenziwa kwemodeli futhi i-X1 icubungula umusi wedatha ukukhiqiza imiphumela yokuvumayo.