Wani sabon salo na gama-gari manufar tsara API NVIDIA CUDA 10.2, kusan watanni goma bayan sigar 10.1. Wannan laburaren ya haɗa da ƙarin cikakken API don gudanar da ƙwaƙwalwar ajiya ta kama-da-wane akan katin zane, tare da ƙarin daidaitattun ayyuka don rabon ƙwaƙwalwar ajiya da jeri na adreshin ƙwaƙwalwa.
Cuda dandamali ne mai ƙididdigar lissafi wanda Nvidia ya ƙirƙira wanda za'a iya amfani dashi don haɓaka haɓaka ta amfani da ikon sashin sarrafa kayan zane (GPU) a cikin tsarin ku. Cuda shine tsarin software barin masu haɓaka software don samun damar tsarin koyarwar kama-da-wane na GPU kuma zuwa ga abubuwan daidaitaccen lissafi, don aiwatar da ginshiƙan lissafi.
CUDA gwada amfani da fa'idodin GPUs akan CPUs babban dalili ta amfani da daidaituwa da aka bayar ta maɓallansa da yawa, wanda ke ba da izinin ƙaddamar da adadi mai yawa na zaren lokaci ɗaya.
Sabili da haka, idan an tsara aikace-aikace ta amfani da zaren da yawa waɗanda ke aiwatar da ayyuka masu zaman kansu (wanda shine abin da GPUs ke yi yayin sarrafa zane-zane, aikinsu na al'ada), GPU zai iya ba da babban aiki.
Menene sabo a Nvidia CUDA 10.2?
Wannan sigar tana cike da dakunan karatu waɗanda ke ba da sabbin ayyuka., gyaran kura-kurai, da ingantaccen aiki don yanayin mahalli daya da daya na GPU.
A cikin wannan sigar an kara sabon layin hulda da tsarin aiki a ainihin lokacin (RTOS) don NVIDIA DRIVE OSda ake kira Hadin gwiwar Sadarwar Sadarwa ta NVIDIA.
Akwai manyan hanyoyin sadarwa guda biyu: NvSciBuf don musayar cikakkun wuraren ƙwaƙwalwar ajiya da NvSciSync don aiki tare. Waɗannan fasalulluka suna cikin samfoti.
A matakin matakan dandamali, CUDA 10.2 shine sabon salo wanda zai kasance don macOS, Bugu da kari, RHEL 6 ba za a sake samun tallafi ba kamar yadda RHEL 2010 ba za a sake samun tallafi kwata-kwata a sigar CUDA ta gaba ba (kamar dai kamfanonin Microsoft C ++ compilers 2013 zuwa XNUMX).
Baya ga wannan Nvidia kuma yana shirya ƙaramin nau'i akan ayyukan da ake dasu. Yanzu nvJPEG wani ɗakin karatu ne daban, ayyukan NPP Compression Primitives masu dacewa suna gab da ɓacewa.
Daga sauran canje-canjen da suka fito daga tallan, Mayila mu iya gano cewa haɓakawa da haɓakawa sun inganta don al'amuran amfani masu zuwa:
- Multi-GPU ba tare da 2 canza ikon ba
- R2C da Z2D canje-canje marasa kyau
- 2D canzawa tare da ƙananan girma da adadi mai yawa.
Idan kanaso samun karin bayani game da wannan sabon sigar na CUDA, zaku iya tuntuba mahada mai zuwa.
Yadda ake girka Nvidia CUDA akan Ubuntu da abubuwan da suka dace?
Domin shigar da CUDA akan tsarin, ya zama dole mu sami direbobin Nvidia shigar. Idan har yanzu ba ku da su, kuna iya tuntuɓar su labarin mai zuwa.
Yanzu a matsayin mataki na farko dole ne mu sauke rubutun shigarwa na CUDA, wanda zamu iya samu daga tashar ta hanyar buga wannan umarnin:
wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda_10.2.89_440.33.01_linux.run
Anyi wannan yanzu dole ne mu ba da izinin aiwatarwa ga rubutun tare da:
sudo chmod +x cuda_10.2.89_440.33.01_linux.run
Zamu sanya wasu kunshin da suka dace.
sudo apt-get install gcc-6 g++-6 linux-headers-$(uname -r) -y sudo apt-get install freeglut3 freeglut3-dev libxi-dev libxmu-dev
Kuma yanzu zamu gudanar da rubutun tare da:
sudo sh cuda_10.2.89_440.33.01_linux.run
Yayin aikin shigarwa za a yi mana wasu tambayoyi abin da za mu amsa, asali zai tambaye mu idan muka yarda da yanayin amfani, idan muna son canza kundin adireshin, a tsakanin sauran abubuwa.
Ina ya kamata su kula shine lokacin da aka tambaye su idan kuna son shigar da direbobin Nvidia inda zasu amsa a'a tunda dole ne a girka su.
Bayan ka gama shigarwa, kawai zasu saita canjin yanayin su a cikin fayil ɗin da za mu ƙirƙira a cikin hanyar mai zuwa /etc/profile.d/cuda.sh.
sudo nano /etc/profile.d/cuda.sh
Kuma a cikin za mu sanya waɗannan abubuwan masu zuwa:
export PATH=$PATH:/usr/local/cuda/bin export CUDADIR=/usr/local/cuda
Sun kuma ƙirƙiri fayil ɗin:
sudo nano /etc/ld.so.conf.d/cuda.conf
Kuma mun ƙara layin:
/usr/local/cuda/lib64
Kuma a ƙarshe za mu zartar:
export PATH=/usr/local/cuda-10.2/bin:/usr/local/cuda-10.2/NsightCompute-2019.1${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64\
${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
sudo ldconfig