NVIDIA Jetson Nano: konke malunga nebhodi yophuhliso

I-NVIDIA Jetson Nano Ibhodi yophuhliso ekhethekileyo. Ingakhangeleka njengeyakho ngeendlela ezininzi Raspberry Pi, okanye Arduino, Kodwa yenzelwe ngokukodwa uhlobo oluthile lweprojekthi. Kwaye njengazo ezinye iibhodi zophuhliso, ikwaxabiso eliphantsi kwaye lisayizi encinci xa kuthelekiswa nezinye izixhobo.

Ngokukodwa, i-NVIDIA's Jetson Nano ijolise ngokukodwa kuphuhliso lwe ubukrelekrele bobugcisa kunye neeprojekthi zenethiwekhi ezingezizo ezokwenziwa. Indlela engabizi kakhulu ukuqala kweli hlabathi, funda ukuba zisebenza njani ezi nkqubo zikrelekrele, kwaye wenze ubuninzi beeprojekthi onokucinga ngazo ...

Yintoni uJetson Nano?

I-NVIDIA Jetson Nano Yibhodi yophuhliso, i-SBC yokwenza iiprojekthi ezininzi ngokusekwe kuthungelwano lwe-neural, ukufunda okunzulu kunye ne-AI. Ngayo unokwenza iiprojekthi ezahlukeneyo kakhulu, ukusuka kwizicelo ezincinci ezikrelekrele ze-IoT, ukuya kwiirobhothi ezintsonkothileyo, iinkqubo zombono ezingezizo kunye nokuqondwa kwento, izixhobo ezisabela ngobukrelekrele ngokuvavanya uthotho lweeparameter zenzwa, izithuthi ezizimeleyo ezincinci, njl.

Kodwa zonke zinepleyiti yesayizi encinci, kwaye ngexabiso kakhulu ngexabiso eliphantsi xa kuthelekiswa nezinye iinkqubo zobuchwephesha ezineempawu ezifanayo.

Kwaye ukuba uyazibuza kutheni kufuneka unayo kwezi bhodi ze-NVIDIA Jetson Nano, kuya kufuneka ugcine engqondweni ukuba ezi bhodi ziya kukuvumela ukuba wenze iiprojekthi ezininzi ngelixa ufunda ngetekhnoloji ekhulayo. Ziyanda iinkampani ezinomdla kubantu abanolwazi lokufunda ngomatshini, i-AI, ukufunda okunzulu, kunye nolunye uqeqesho olufanayo, kuba itekhnoloji yexesha elizayo.

Iimpawu zobugcisa

Izibonelelo zeNVIDIA Jetson Nano Iimpawu ezinomdla ngokwenene ngokobungakanani bayo nexabiso layo. Ayidluli kwi- € 100, kwaye ngeesentimitha ezimbalwa ngobukhulu. Ngaphandle koku, inokukhula ukuya kuthi ga kwi-472 yokusebenza kwe-GFLOPs, eyoneleyo yokuqhuba ii-algorithms ezininzi ze-AI ngokukhawuleza kwaye zenze ii-network ezininzi ze-neural ngaxeshanye.

Kwaye ayichukumisi kuphela kula manani, kodwa nakwindlela esetyenziswa ngayo. Le bhodi inokuba nefayile ye- Ukusetyenziswa okuphakathi kwe-5 kunye ne-10W. Xa kuthelekiswa neenkqubo ezifanayo ngokuqinisekileyo kusezantsi, ke ujongene nenkqubo efanelekileyo kakhulu. Akunanto yakwenza nabanye oomatshini abatya amakhulu okanye amawaka eewat ...

Ngolwazi oluthe kratya, uyakubona oku uluhlu olupheleleyo lweenkcukacha:

I-NVIDIA Maxwell GPU ene-128 CUDA cores
I-ARM Cortex-A57 QuadCore CPU
I-RAM 4GB LPDDR4
16GB eMMC 5.1 yokugcina ugcino
Ukuxhuma:
- Isinxibelelanisi sekhamera ye-12-ndlela (3 x 4 okanye 4 x 2) MIPI CSI-2 DPHY 1.1 (18 Gbps)
- Inethiwekhi yeGigabit Ethernet (RJ-45)
- HDMI 2.0 okanye DP 1.2 bonisa uqhagamshelo | eDP 1.4 | I-DSI (1 x 2) 2 ngaxeshanye
- Izibuko i-1/2/4 PCIE, i-1 i-USB 3.0, i-3 i-USB 2.0
- I-O eyongezelelweyo: 1 SDIO / 2 SPI / 4 I2C / 2 I2S / GPIO
- Ikhonkco engama-260
Ubukhulu: 69,6mm x 45mm
Ukusetyenziswa: 5-10w
I-Linux OS nge Ikhithi yophuhliso

Iimveliso zeNVIDIA Jetson zoSapho

I-NVIDIA ineqela lezi zinto iimveliso zophuhliso lwe-AI ngothungelwano lwe-neuroanal. Ezinye zeemveliso ezibalaseleyo zezi:

UJetson Xavier NX: yi-SOM, oko kukuthi, iNkqubo yeModyuli, okanye inkqubo epheleleyo edityaniswe kwimodyuli enye. Ngaphandle kokubonakala kwayo kunye nobukhulu bayo, ibonelela ngamagunya okuxhaphaza amakhulu, ukuya kuthi ga kwi-21 TOPs, Oko kukuthi, ukuSebenza kweTera 21 ngomzuzwana. Kwanele ukuqhuba iinethiwekhi ezininzi ezingezizo ezenzekelayo kakuhle nangokukodwa.
UJetson AGX Xavier: enye imodyuli enamandla kakhulu ngokobungakanani bekhompyuter kunye nokusebenza kakuhle kwaye oku kuza emva kweJetson Nano, ukuvumela ukwenziwa kwezizukulwana ezitsha zoomatshini abakrelekrele.
IJetson TX2: enye indlela eya kuJetson Nano, nakusapho olunye. Imele isantya sayo esikhulu kunye nokusebenza kwamandla. Eyona nto yenzelwe usetyenziso lwe-AI olungisiweyo, apho ubungakanani kunye nokusetyenziswa komcimbi. Kule meko, isekwe kuyilo lwe-NVIDIA Pascal, inikwe amandla yi-8GB ye-RAM kunye ne-bandwidth efikelela kwi-59,7GB / s.

Thenga iNVIDIA Jetson Nano

Ukuba uzimisele ukuqala kumenzi okanye kwihlabathi le-DIY ngeeprojekthi zenethiwekhi ezingezizo ezenziweyo, unako thenga le bhodi yeNVIDIA Jetson Nano kwiivenkile ezikhethekileyo okanye kumaqonga anjengeAmazon, apho athengiswa ngokwahlukeneyo okanye ngeekiti zophuhliso ukuze uqalise ngokukhawuleza ngayo yonke into oyifunayo:

Okwangoku ibhodi ye-NVIDIA Jetson Nano isungulwe ngexabiso eliphantsi malunga ne- $ 59 kananjalo bongeze iWiFi. Iindaba ezimnandi, inye kuphela into yokuba bayinciphisile imemori ephambili ukuya kwi-2GB. Ukuba uyayifuna kuya kufuneka ulinde, kuba okwangoku kuphela presale yamaqabane ...

Enye indlela eya kwi-NVIDIA Jetson Nano

Ukuba unomdla wokufunda ngomatshini, i-AI kunye neenethiwekhi ezenziweyo ezingezizo, kuya kufuneka uyazi ezinye ezinye iindlela zeNVIDIA Jetson Nano, kuba ayisiyiyo ipleyiti kuphela yezi njongo. Unokufumana ezinye ii-SBCs eziyilelwe ngokukodwa ezi projekthi zilandelayo:

Ikhorali kaGoogle

UGoogle wenze ibheji, Ikhorali kaGoogle, kunye nezinye izixhobo kunye neemodyuli ezifunekayo ukwenza iiprojekthi ze-AI. Phakathi kwamanqaku eli qonga unayo:

UGoogle Coral unazo ezinye Iimpawu zobugcisa ezibengezelayo, njenge:

I-NXP i.MX 8M CPU ene-Quad Core Cortex-A53 kunye neCortex-M4F
IGC7000 Lite Graphics GPU,
Iprosesa kaGoogle Edge TPU ukuya kuthi ga kwi-4 TOPS okanye kwi-2 TOPS / w.
Kubandakanya i-1GB LPDDR4 RAM
Ukugcina ukuya kuthi ga kwi-8GB eMMC flash kunye nokubanakho kokuyandisa usebenzisa amakhadi e-MicroSD.
Inonxibelelwano lweWiFi, i-USB, iBluetooth, iEthernet, ijack yeaudiyo, iHDMI, iMIPI-DSI, kunye nokuhanjiswa kwamandla kwe-USB-C 5v.

IKhadas VIM3

IKhadas VM3 Ngenye indlela yeeprojekthi zakho ze-AI, nangona ingenazo ezinye zeempawu ezinkulu, yibhodi ethobekileyo enokuthi ibe ithuba elihle lokuqala:

CPU A311D x4 Cortex-A73 2.2Ghz kunye x2 Cortex-A53 kwi 1.8Ghz.
Nge-NPU kwi-5 TOPS
Ukuya kwi-4GB ye-RAM
16-32GB eMMC Samsung
I-MIPI-DIS, i-HDMI, i-WiFi, i-Ethernet, i-microSD, i-USB, ii-PCIe zonxibelelwano, njl.

I-HiSilicon HiKey 970 (iHuawei)

I-HiSilicon yinkampani ephantsi Huawei evelisa iichips. Ewe, phantsi kolu phawu uya kufumana enye indlela yokuphuhlisa iiprojekthi zenethiwekhi ezinje Molweni 970, Iyahambelana neHuawei SDK. Ukongeza, inezinto ezithile ezinomdla:

I-ARM Kirin eneCortex A73 QuadCore + Cortex-A53 QuaCore
IMali G72 MP12 GPU
Ii-NPU ezinikezelweyo
I-6GB ye-LPDDR4
Imemori ye-64GB yememori
I-WiFi, i-microSD, i-HDMI, i-USB, ii-PCIe zonxibelelwano, njl.
UEFI

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

ISophon BM1880 Yenye ibhodi ephuhliswe nguSophon.ia. Ukuba uthatha isigqibo sokuyithenga, uya kufumana amanqaku anje ngala:

2x Cortex-A53 CPU kwi1.5Ghz + RISC-V kwi1Ghz
1 TPUs @ INT8 enkosi kumqhubekekisi weTensor
4GB LPDDR4
Ifowuni ye32GB eMMC
Unxibelelwano lwe-Ethernet, iWiFi, i-USB, i-MicroSD, uJack, njl.

Intsingiselo ye-Intel Neural

Enye iprojekthi efana neyangaphambili yile Intonga ye-Intel Neural. Inguqulelo yesi-2 ngoku iyafumaneka, kwaye into ebonakalayo kule meko kukuba yintonga ye-USB onokuthi uqhagamshele ngokulula kwiPC ukuqala iiprojekthi zakho, nangona inokungafani kakhulu kuneebhodi zangaphambili. Kwakhona, ukuba ufuna amandla angaphezulu, ungazisebenzisa ezininzi kwiziko le-USB ukongeza amandla ...

Si ukuthenga le Ndonga yeNeuralIxabiso lentengo malunga ne- € 100, kwaye iyahambelana neLinux kunye neWindows. Ukongeza, ivumela ukusebenza ne-OpenVINO njengesixhobo sophuhliso.

I-Rockchip RK3399Pro

I-Rockchip unayo le nto inamandla yokukhawulezisa izixhobo zophuhliso lokufunda ezinokwenza iiprojekthi ezinomdla kunye nezahlukeneyo. Ixhasa iTensorFlow Caffe ukuya kuthi ga kwi-3 TOPS, kunye neenkqubo zokusebenza ze-Android kunye ne-GNU / Linux.

Ukuba ufuna ukuyithenga, unayo kwi Iinguqulelo ezahlukeneyo (iyalelwe ukusuka kwelona xabiso lisezantsi ukuya kwelona liphezulu):

Hardwarelibre

I-NVIDIA Jetson Nano: konke malunga nebhodi yophuhliso