I-NVIDIA Jetson Nano: konke mayelana nebhodi lokuthuthukisa

NVIDIA Jetson Nano Yibhodi lentuthuko ekhethekile. Ingabukeka njengeyakho ngezindlela eziningi Ukusajingijolo Pi, noma I-Arduino, kepha yenzelwe uhlobo oluthile lwephrojekthi. Futhi njengalawa amanye amabhodi entuthuko, ibuye ibe nentengo ephansi futhi isayizi encane uma iqhathaniswa neminye imishini.

Ngokuqondile, uJetson Nano weNVIDIA ubhekise ngqo ekuthuthukisweni kwe- ubuhlakani bokufakelwa kanye namaphrojekthi wamanethiwekhi we-neural network. Indlela eshibhile yokuqala kulo mhlaba, funda ukuthi lezi zinhlelo ezihlakaniphile zisebenza kanjani, futhi wakhe ubuningi bamaphrojekthi ongawacabanga ...

Uyini uJetson Nano?

NVIDIA Jetson Nano yibhodi lezentuthuko, i-SBC yokwakha amaphrojekthi amaningi ngokususelwa kumanethiwekhi we-neural, ukufunda okujulile ne-AI. Ngayo ungakha amaphrojekthi ahlukahlukene kakhulu, kusuka kuzinhlelo zokusebenza ezincane ezihlakaniphile ze-IoT, kuya kumarobhothi ayinkimbinkimbi ngokwengeziwe, amasistimu wokubuka okufakelwayo nokuqashelwa kwento, amadivayisi asabela ngokukhalipha ngokuhlola uchungechunge lwamapharamitha wezinzwa, izimoto ezincane ezizimele, njll.

Kepha konke kunepuleti lobukhulu obuncane, futhi ngentengo impela okungabizi kuqhathaniswa nezinye izinhlelo zobuchwepheshe ezinezici ezifanayo.

Futhi uma uzibuza kungani kufanele ube neyodwa kulawa mabhodi we-NVIDIA Jetson Nano, kufanele ugcine ukuthi amabhodi azokuvumela ukuthi wakhe amaphrojekthi amaningi ngenkathi ufunda ngobuchwepheshe obukhulayo. Kunezinkampani eziningi ngokwengeziwe ezinentshisekelo kubantu abanolwazi lokufunda ngomshini, i-AI, ukufunda okujulile, neminye imikhakha efanayo, ngoba ubuchwepheshe besikhathi esizayo.

Izici zobuchwepheshe

Ukunikezwa kwe-NVIDIA Jetson Nano izici ezihlaba umxhwele ngempela ngosayizi nentengo yayo. Alidluli ngo- € 100, nangamasentimitha ambalwa ngosayizi. Ngaphandle kwalokhu, ingakhula ifike kuma-GFLOP wokusebenza angama-472, anele ukusebenzisa ama-algorithms amaningi we-AI ngokushesha okukhulu futhi acubungule amanethiwekhi wokufakelwa amaningi ngasikhathi sinye.

Futhi ayihehi kuphela lezo zibalo, kepha futhi nokusetshenziswa kwayo okuphansi. Leli bhodi lingaba ne- ukusetshenziswa okuphakathi kuka-5 no-10W. Uma kuqhathaniswa nezinhlelo ezifanayo ngokuqinisekile liphansi, ngakho-ke ubhekene nesistimu esebenza kahle kakhulu. Akuhlangene nakancane neminye imishini esebenzisa amakhulukhulu noma izinkulungwane zama-watts ...

Ngeminye imininingwane, ungabona lokhu uhlu lwemininingwane ephelele:

I-NVIDIA Maxwell GPU enezinhlamvu ze-128 CUDA
I-ARM Cortex-A57 QuadCore CPU
I-RAM 4GB LPDDR4
Isitoreji se-flash esingu-16GB eMMC 5.1
Ukuxhumana:
- Isixhumi se-12-way yekhamera (3 x 4 noma 4 x 2) MIPI CSI-2 DPHY 1.1 (18 Gbps)
- Inethiwekhi yeGigabit Ethernet (RJ-45)
- I-HDMI 2.0 noma i-DP 1.2 display connection | i-eDP 1.4 | I-DSI (1 x 2) 2 ngasikhathi sinye
- Amachweba 1/2/4 PCIE, 1 USB 3.0, 3 USB 2.0
- I-O / O eyengeziwe: 1 SDIO / 2 SPI / 4 I2C / 2 I2S / GPIO
- Isixhumi se-pin esingu-260
Usayizi: 69,6mm x 45mm
Ukusetshenziswa: 5-10w
I-Linux OS nge ikhithi yokuthuthukisa

Imikhiqizo Yomndeni yeNVIDIA Jetson

I-NVIDIA ineziningi zalezi imikhiqizo yokuthuthukiswa kwe-AI ngamanethiwekhi we-neuroanal okufakelwa. Eminye imikhiqizo evelele kakhulu yile:

UJetson Xavier NX: yi-SOM, okungukuthi, iSystem On Module, noma uhlelo oluphelele oluhlanganiswe kwimodyuli eyodwa. Ngaphandle kokubukeka kwayo nosayizi wayo, inikeza amandla ajwayelekile wokusebenzisa amandla amakhulu, anama-TOP angama-21, okungukuthi, ama-21 Tera Operations ngomzuzwana. Lokho kwanele ukusebenzisa amanethiwekhi wokufakelwa amaningi ngokushelela nangesikhathi esisodwa.
UJetson AGX Xavier: enye imodyuli enamandla kakhulu ngokuya ngobuningi bekhompyutha kanye nokusebenza kahle futhi okuze ngemuva kukaJetson Nano, okuvumela ukwenziwa kwezizukulwane ezintsha zemishini ehlakaniphile.
IJetson TX2: enye indlela eya kuJetson Nano, nasemndenini ofanayo. Igqama ngejubane layo elikhulu nokusebenza kahle kwamandla. Idizayinelwe ikakhulukazi izinhlelo zokusebenza ze-AI ezishumekiwe, lapho usayizi nokusetshenziswa kunendaba. Kulokhu, kusekelwe ekwakhiweni kwe-NVIDIA Pascal, enikwe amandla yi-8GB ye-RAM kanye ne-bandwidth efinyelela ku-59,7GB / s.

Thenga i-NVIDIA Jetson Nano

Uma uzimisele ukuqala kumenzi noma emhlabeni we-DIY ngamaphrojekthi enethiwekhi wokufakelwa, unga thenga leli bhodi le-NVIDIA Jetson Nano ezitolo ezikhethekile noma kumapulatifomu afana ne-Amazon, lapho athengiswa ngokwahlukana noma ngamakhithi entuthuko ukuze uqalise ngokushesha ngakho konke okudingayo:

Njengamanje kwethulwe ibhodi le-NVIDIA Jetson Nano ngentengo eyehlisiwe ye- cishe u- $ 59 futhi bangeze naku-WiFi. Izindaba ezinhle, okuwukuphela kwento ukuthi banciphise inkumbulo enkulu yaba yi-2GB. Uma uyifuna kuzofanele ulinde, ngoba manje iku- presale labalingani ...

Ezinye izindlela ze-NVIDIA Jetson Nano

Uma unesifiso sokufunda ngomshini, i-AI kanye namanethiwekhi we-neural wokufakelwa, kufanele wazi okunye ezinye izindlela ze-NVIDIA Jetson Nano, ngoba akuyona ipuleti kuphela yalezi zinhloso. Ungathola amanye ama-SBC aklanyelwe ngqo la maphrojekthi afana nalokhu okulandelayo:

I-Google Coral

IGoogle isungule ibheji, I-Google Coral, kanye nezinye izesekeli namamojula adingekayo ukudala amaphrojekthi we-AI. Phakathi kwezindatshana zale pulatifomu onazo:

IGoogle Coral inezinye izici zobuchwepheshe ekhazimulayo, njenge:

I-NXP i.MX 8M CPU ene-Quad Core Cortex-A53 ne-Cortex-M4F
I-GC7000 Lite Graphics GPU,
Isiprosesa seGoogle Edge TPU esifika ku-4 TOPS noma 2 TOPS / w.
Kufaka i-1GB LPDDR4 RAM
Ukugcina kuze kufinyelele ku-8GB eMMC flash nokuthi kungenzeka ukuthi kunwetshwe kusetshenziswa amakhadi e-MicroSD.
Inokuxhumeka kwe-WiFi, i-USB, i-Bluetooth, i-Ethernet, i-audio jack, i-HDMI, i-MIPI-DSI, namandla ngaphezu kwe-USB-C 5v.

I-Khadas VIM3

I-Khadas VM3 Ngenye indlela yamaphrojekthi wakho we-AI, yize ingenazo ezinye zezici ezinkulu, yibhodi elinesizotha elingaba ithuba elihle lokuqala:

I-CPU A311D x4 Cortex-A73 2.2Ghz ne-x2 Cortex-A53 ku-1.8Ghz.
Nge-NPU ku-5 TOPS
Kuze kufike ku-4GB ye-RAM
16-32GB eMMC Samsung
IMIPI-DIS, HDMI, WiFi, Ethernet, microSD, USB, PCIe ukuxhumana, njll.

I-HiSilicon HiKey 970 (Huawei)

I-HiSilicon yinkampani engaphansi Huawei lokho kukhiqiza amachips. Yebo, ngaphansi kwalo mkhiqizo uzothola enye indlela yokuthuthukisa amaphrojekthi wenethiwekhi ye-neural njenge Sawubona 970, iyahambisana ne-Huawei SDK. Ngaphezu kwalokho, inezici ezithile ezithakazelisayo:

I-ARM Kirin ene-Cortex A73 QuadCore + Cortex-A53 QuaCore
IMali G72 MP12 GPU
Ama-NPU azinikezele
I-6GB ye-LPDDR4
Inkumbulo ye-flash engu-64GB
I-WiFi, i-microSD, i-HDMI, i-USB, ukuxhumana kwe-PCIe, njll.
UEFI

I-Sophon BM1880 (i-ARM e-hybrid + RISC-V)

ISophon BM1880 Yibhodi ehlukile elenziwe yiSophon.ia. Uma uthatha isinqumo sokuyithenga, uzothola ezinye izici ezinjenge:

2x Cortex-A53 CPU ku-1.5Ghz + RISC-V ku-1Ghz
1 TPUs @ INT8 sibonga iprosesa yeTensor
4GB LPDDR4
32GB eMMC flash
Ukuxhumana i-Ethernet, i-WiFi, i-USB, i-MicroSD, iJack, njll.

I-Intel Neural Stick

Enye iphrojekthi efana naleyo eyedlule yile I-Intel Neural Stick. I-Version 2 isiyatholakala manje, futhi okucacile kulokhu ukuthi yinduku ye-USB ongayixhuma kahle kwi-PC ukuqala amaphrojekthi akho, yize inokusebenza okuncane kakhulu kunamabhodi wangaphambilini. Futhi, uma udinga amandla amaningi, ungawasebenzisa amaningi kuhabhu le-USB ukwengeza amandla ...

Si bethenga le Nduku yeNeural, ibiza cishe u- € 100, futhi iyahambisana neLinux neWindows. Ngaphezu kwalokho, ivumela ukusebenza ne-OpenVINO njengethuluzi lokuthuthukisa.

I-Rockchip RK3399Pro

I-Rockchip unayo le khithi yokuthuthukisa yokufunda ejulile esheshayo esebenza ngokuqinile okwakha ngayo amaphrojekthi ajabulisa kakhulu futhi ahlukahlukene. Ixhasa iTensorFlow Caffe kuze kufike ku-3 TOPS, kanye nezinhlelo zokusebenza ze-Android ne-GNU / Linux.

Uma ufuna ukuyithenga, unayo nge izinhlobo ezahlukahlukene (i-odolwe kusuka entweni ephansi kuya phezulu):

Hardwarelibre