I-CUDA Cores ichazwe
I-CUDA, isichazamazwi sokuthi i-Compute Architect Device Architecture, i-teknoloji eyenziwe ngu-Nvidia osheshisa izinqubo zokubalwa kwe-GPU.
Nge-CUDA, abacwaningi nabathuthukisi be-software bangathumela ikhodi C, C ++, nekhodi ye-Fortran ngqo kwi-GPU ngaphandle kokusebenzisa ikhodi yomhlangano. Lokhu kuvumela ukuba basebenzise i-computing efanayo lapho izinkulungwane zemisebenzi, noma izintambo, zenziwa ngesikhathi esisodwa.
Imininingwane ngeCUDA Cores
Kungenzeka ukuthi uke wabona igama elithi CUDA uma uthenga ngekhadi levidiyo le- Nvidia. Uma ubheka ukupakishwa kwekhadi elinjalo noma ufunde ukubuyekezwa kwekhadi levidiyo, uzovame ukubona isibalo senombolo ye-CUDA cores.
Ama-CUDA ama-processors afanayo afana neprosesa kwikhompyutha, okungenzeka kube yinkambinkimbi yombili noma yesikhulumi. I-Nvidia GPUs, noma kunjalo, ingaba nezinhlamvu eziningana eziningana. Lezi zinhlamvu zithwale imisebenzi ehlukahlukene evumela ukuthi inani lezinhlamvu liqondane ngokuqondile nesivinini namandla weGPU.
Njengoba i-CUDA cores inesibopho sokubhekana nayo yonke idatha ehamba ngeGPU, ama-cores aphatha izinto ezifana nehluzo emidlalweni yevidiyo ezimweni ezifana nokuthi izinhlamvu nezindawo zilayishaphi.
Izicelo zingakhiwa ukuze zizuze ukusebenza okwenyuka kwama-CUDA. Ungabona uhlu lwalezi zicelo ekhasini lezinhlelo zokusebenza ze-Nvidia's GPU.
Izinhlamvu ze-CUDA zifana namaphrosesa we-AMD's Stream; bavele bathiwa ngokuhlukile. Noma kunjalo, awukwazi ukulinganisa i-300 CUDA Nvidia GPU ene-300 ye-Stream Processor AMD GPU.
Ukukhetha iKhadi Levidiyo Nge-CUDA
Inombolo ephakeme yezinhlamvu ze-CUDA ngokuvamile kusho ukuthi ikhadi levidyo linikeza ukusebenza okusheshayo jikelele. Noma kunjalo, inani lezinhlamvu ze-CUDA lingenye yezinto eziningana okufanele uzicabange lapho ukhetha ikhadi levidyo .
I-Nvidia inikeza ububanzi bamakhadi abukeka njengembalwa njengezingcingo ezingu-8 ze-CUDA, njengeGeForce G100, kuma-C7A angu-5 760 ama-CUDA ku-GeForce GTX TITAN Z.
Amakhadi e-Graphics ayenama-Tesla, Fermi, iKepler, Maxwell, noma ama-Pascal amakhemikhali asekela i-CUDA.