I-NVIDIA ipapashe ukukhululwa kokuqala okuzinzileyo kwesebe elitsha lomqhubi we-NVIDIA 470.57.02. Umqhubi uyafumaneka kwiLinux (ARM, x86_64), FreeBSD (x86_64) kunye neSolaris (x86_64).
Iinguqulelo eziphambili:
- Inkxaso eyongeziweyo yee-GPU ezintsha: i-GeForce RTX 3070 Ti, i-GeForce RTX 3080 Ti, T4G, A100 80GB PCIe, A16, PG506-243, PG506-242, CMP 90HX, CMP 70HX, A100-506PG207, A100-506PG217, A50-XNUMXPGXNUMX I-CMP XNUMXHX.
- Inkxaso yokuqala eyongeziweyo ye-OpenGL kunye ne-Vulkan hardware acceleration ye-X11 izicelo ezisebenza kwindawo ze-Wayland zisebenzisa icandelo le-Xwayland DDX. Ngokujonga iimvavanyo, xa usebenzisa isebe lomqhubi we-NVIDIA 470, ukusebenza kwe-OpenGL kunye ne-Vulkan kwizicelo ze-X eziqaliswe kusetyenziswa i-XWayland iphantse yafana nokusebenza phantsi kwe-server ye-X eqhelekileyo.
- Ukukwazi ukusebenzisa iteknoloji ye-NVIDIA NGX kwiWayini kunye nephakheji yeProton, ephuhliswe yiValve yokusebenzisa imidlalo yeWindows kwiLinux, iphunyeziwe. Kubandakanya iWayini kunye neProton, ngoku ungaqhuba imidlalo exhasa itekhnoloji ye-DLSS, ekuvumela ukuba usebenzise i-Tensor cores yamakhadi evidiyo e-NVIDIA ukukala umfanekiso obonakalayo usebenzisa iindlela zokufunda ngomatshini ukunyusa isisombululo ngaphandle kokulahlekelwa ngumgangatho.
Ukusebenzisa ukusebenza kwe-NGX kwiinkqubo zeWindows eziqaliswe kusetyenziswa iWayini, ilayibrari ye-nvngx.dll ibandakanyiwe. KwiWayini kunye nokukhutshwa okuzinzileyo kweProton, inkxaso ye-NGX ayikaphunyezwa, kodwa utshintsho lokuxhasa lo msebenzi sele luqalisile ukubandakanywa kwisebe leProton Experimental.
- Imida isusiwe kwinani lemixholo ye-OpenGL ekwanjalo, enqandwa kuphela ngobungakanani bememori ekhoyo.
- Inkxaso eyongeziweyo yetekhnoloji ye-PRIME yokukhuphela imisebenzi yonikezelo kwezinye ii-GPU (i-PRIME Display Offload) kuhlengahlengiso apho umthombo kunye nee-GPU ekujoliswe kuzo zicutshungulwa ngumqhubi we-NVIDIA, kunye naxa i-GPU yomthombo iqhutywe ngumqhubi we-AMDGPU.
- Inkxaso eyongeziweyo yezandiso ezintsha zeVulkan: VK_EXT_global_priority (VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT, ivumela ukusetyenziswa kokuchaswa okungafaniyo kwi-SteamVR), i-VK_EXT_global_priority_query, VK_EXT_provoking_vertex, VK_EXT_dynamic_state_extended_EXT_dynamic VK_ EXT_vertex_input_dynamic_state, VK_EXT_ycbcr_2plane_2_fomati, VK_NV_inherited_viewport_scissor.
- Ukusebenzisa iipropati zehlabathi ze-Vulkan ngaphandle kwe-VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT ngoku kufuna ufikelelo lweengcambu okanye amalungelo e-CAP_SYS_NICE.
- Yongeza imodyuli entsha ye-kernel nvidia-peermem.ko evumela ukuba i-RDMA isetyenziswe ukufikelela ngokuthe ngqo kwimemori ye-NVIDIA GPU ngezixhobo zomntu wesithathu ezifana neMellanox InfiniBand HCA (I-Host Channel Adapters) ngaphandle kokukopisha idatha kwimemori yenkqubo.
- Ngokungagqibekanga, ukuqaliswa kwe-SLI kunikwe amandla xa usebenzisa ii-GPU ezinenani elahlukileyo lememori yevidiyo.
- Izicwangciso ze-nvidia kunye ne-NV-CONTROL zibonelela ngezixhobo zokulawula ezipholileyo ngokungagqibekanga kwiibhodi ezixhasa ulawulo olupholileyo lwesoftware.
- I-firmware ye-gsp.bin ibandakanyiwe, esetyenziselwa ukuhambisa ukuqaliswa kunye nolawulo lwe-GPU ukuya kwicala le-GPU System Processor (GSP) chip.
Ngelo xesha, kwiNkomfa yabaPhuhlisi beMidlalo, i-NVIDIA ibhengeze ikhowudi yomthombo ovulekileyo we-RTXMU (i-RTX Memory Utility) i-SDK toolkit phantsi kwelayisenisi ye-MIT, evumela ukusetyenziswa kokudibanisa kunye nokuhanjiswa kwe-BLAS (izakhiwo ezisezantsi zokukhawuleza) ukunciphisa kakhulu ukusetyenziswa kwememori yevidiyo. I-Compaction yenza kube lula ukunciphisa usetyenziso lwememori ye-BLAS iyonke nge-50%, kwaye ukuhanjiswa kuphucula ukusebenza kakuhle kogcino lwe-buffer ngokudibanisa izithinteli ezininzi ezincinci kumaphepha angama-64 KB okanye i-4 MB ngobukhulu.
I-NVIDIA ikwavule ikhowudi yethala leencwadi le-NVRHI (NVIDIA Rendering Hardware Interface) kunye nesakhelo seDonut phantsi kwelayisenisi ye-MIT. I-NVRHI ngumaleko ongabonakaliyo osebenza ngaphezulu kwee-API zegraphics ezahlukeneyo (Direct3D 11, Direct3D 12, Vulkan 1.2) kwiWindows kunye neLinux. I-Donut ibonelela ngeseti yezinto ezakhelwe kwangaphambili kunye nezigaba zonikezelo lweprototyping yexesha lokwenyani iinkqubo zonikezelo.
Ukongeza, i-NVIDIA ibonelele ngenkxaso ye-Linux kunye noyilo lwe-ARM kwi-SDK: i-DLSS (i-Deep Learning Super Sampling, ukulinganisa umfanekiso ongokoqobo usebenzisa iindlela zokufunda ngomatshini), i-RTXDI (RTX Direct Illumination, ukukhanya okuguquguqukayo), i-RTXGI (RTX Global Illumination, ukuzonwabisa ukukhanya okukhanyayo ), NRD (NVIDIA Optix AI-Acceleration Denoiser, usebenzisa umatshini wokufunda ukukhawulezisa unikezelo lomfanekiso oluyinyani).
umthombo: opennet.ru