{"id":120066,"date":"2024-10-07T01:09:14","date_gmt":"2024-10-06T23:09:14","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/proekt-zluda-prodolzhit-razvitie-v-forme-universalnoj-realizaczii-cuda"},"modified":"2024-10-07T01:09:14","modified_gmt":"2024-10-06T23:09:14","slug":"proekt-zluda-prodolzhit-razvitie-v-forme-universalnoj-realizaczii-cuda","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sq\/blog\/novosti-interneta\/proekt-zluda-prodolzhit-razvitie-v-forme-universalnoj-realizaczii-cuda","title":{"rendered":"Projekti ZLUDA do t\u00eb vazhdoj\u00eb zhvillimin si nj\u00eb realizim universal i CUDA","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Andrzej Janik prezantoi planin e zhvillimit t\u00eb m\u00ebtejsh\u00ebm t\u00eb projektit ZLUDA, q\u00eb zhvillon nj\u00eb implementim t\u00eb hapur t\u00eb teknologjis\u00eb CUDA. Nd\u00ebrsa versioni i par\u00eb i ZLUDA kishte si q\u00ebllim krijimin e nj\u00eb implementimi CUDA p\u00ebr GPU-t\u00eb Intel, dhe versioni i dyt\u00eb p\u00ebr GPU-t\u00eb AMD, versioni i tret\u00eb i paraqitur tani do t\u00eb p\u00ebrqendrohet n\u00eb zhvillimin e nj\u00eb implementimi universl t\u00eb CUDA p\u00ebr \u00e7do GPU, p\u00ebrve\u00e7 GPU-ve NVIDIA. Q\u00ebllimi i projektit ZLUDA \u00ebsht\u00eb t\u00eb ofroj\u00eb n\u00eb sisteme me GPU, t\u00eb ndryshme nga NVIDIA, mund\u00ebsin\u00eb e ekzekutimit t\u00eb aplikacioneve CUDA t\u00eb pa modifikuara me nj\u00eb performanc\u00eb t\u00eb ngjashme me at\u00eb t\u00eb aplikacioneve q\u00eb ekzekutohen pa nd\u00ebrmjet\u00ebs. Kodi i projektit vazhdon t\u00eb shp\u00ebrndahet n\u00ebn licencat MIT dhe Apache 2.0. <\/p>\n
Fillimi i projektit ZLUDA ishte zhvilluar p\u00ebr GPU Intel, por n\u00eb vitin 2021 kjo kompani e konsideroi ofrimin e mund\u00ebsis\u00eb p\u00ebr t\u00eb ekzekutuar aplikacione CUDA n\u00eb GPU Intel si t\u00eb pap\u00ebrgatitur p\u00ebr biznesin dhe nuk vazhdoi t\u00eb mb\u00ebshtes\u00eb zhvillimin e iniciativ\u00ebs. N\u00eb fillim t\u00eb vitit 2022, zhvilluesi u largua nga Intel dhe n\u00ebnshkroi nj\u00eb kontrat\u00eb me AMD p\u00ebr zhvillimin e nj\u00eb shtrese p\u00ebr pajtueshm\u00ebrin\u00eb me CUDA p\u00ebr GPU-n\u00eb e k\u00ebsaj kompanie. Pas dy viteve, kompania AMD humbi interesin p\u00ebr mund\u00ebsin\u00eb e ekzekutimit t\u00eb aplikacioneve CUDA n\u00eb GPU AMD. <\/p>\n
N\u00eb p\u00ebrputhje me kushtet e kontrat\u00ebs dhe pas marrjes s\u00eb lejes p\u00ebr publikimin nga p\u00ebrfaq\u00ebsuesi i AMD gjat\u00eb nj\u00eb komunikimi me email, Andrzej hapi kodin e punimeve, t\u00eb krijuara gjat\u00eb pun\u00ebs n\u00eb AMD dhe q\u00eb lejojn\u00eb ekzekutimin e aplikacioneve CUDA n\u00eb struktur\u00ebn ROCm dhe runtime HIP (Heterogeneous-computing Interface for Portability) t\u00eb zhvilluar nga kompania AMD. Gjasht\u00eb muaj pas publikimit, avokat\u00ebt e AMD kontaktuan Andrzejn dhe i b\u00ebn\u00eb t\u00eb ditur se leja e dh\u00ebn\u00eb m\u00eb par\u00eb gjat\u00eb komunikimit nuk kishte forc\u00eb ligjore. N\u00eb p\u00ebrfundim, Andrzej u detyrua t\u00eb hiqte nga qarkullimi kodin ZLUDA, t\u00eb krijuar gjat\u00eb pun\u00ebs n\u00eb AMD. <\/p>\n
Varianti i ri i ZLUDA do t\u00eb bazohet n\u00eb kodin ekzistues para fillimit t\u00eb pun\u00ebs s\u00eb Andzhejt me AMD, nuk do t\u00eb jet\u00eb i lidhur me ndonj\u00eb GPU t\u00eb caktuar dhe do t\u00eb p\u00ebrqendrohet n\u00eb ekzekutimin e aplikacioneve q\u00eb p\u00ebrdorin CUDA p\u00ebr t\u00eb p\u00ebrshpejtuar detyrat e lidhura me m\u00ebsimin makin\u00eb dhe inteligjenc\u00ebn artificiale (m\u00eb par\u00eb ZLUDA ishte kryesisht i orientuar ndaj aplikacioneve p\u00ebr krijimin e p\u00ebrmbajtjes, si Arnold Render, Blender dhe 3DF Zephyr). Nd\u00ebr t\u00eb tjera, planifikohet t\u00eb ofrohet mund\u00ebsia e pun\u00ebs s\u00eb kornizave Llama.cpp, PyTorch dhe TensorFlow mbi ZLUDA me p\u00ebrfshirjen e optimizimeve CUDA q\u00eb aplikohen p\u00ebr GPU-t\u00eb NVIDIA. <\/p>\n
N\u00eb fillim projekti do t\u00eb p\u00ebrqendrohet n\u00eb pun\u00ebn me GPU-t\u00eb AMD, por pastaj do t\u00eb adaptohet p\u00ebr GPU-t\u00eb Intel. Implementimi p\u00ebr GPU-t\u00eb AMD do t\u00eb krijohet nga e para, do t\u00eb mb\u00ebshtes\u00eb GPU-t\u00eb, t\u00eb pakt\u00ebn mbi arkitektur\u00ebn e setit t\u00eb komandave RDNA1, dhe do t\u00eb p\u00ebrdor\u00eb stack-un ROCm 6.1+ n\u00eb vend t\u00eb stack-ut ROCm 5, i cili u p\u00ebrdor n\u00eb implementimin e m\u00ebparsh\u00ebm p\u00ebr AMD. Arrihet pariteti i funksionalitetit me implementimin e kaluar, t\u00eb cilin Andzhej ishte i detyruar ta hiqte nga akses i hapur, do t\u00eb planifikohet t\u00eb arrihet n\u00eb tremujorin e tret\u00eb t\u00eb vitit 2025. \n Burimi: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"