GPU grafička kartica objašnjenje: od piksela do AlexNeta

Dokumentarac prati kako je čip projektiran za renderiranje igara, s tisuće jezgri za paralelno bojanje piksela, postao temelj dubokog učenja. Pratite evoluciju od fiksnog pipeline-a 1999. do programabilnih shadera 2001., CUDA SDK-a 2007. i AlexNeta 2012. koji je pogrešku klasifikacije smanjio s 26,2 % na 15,3 %. Saznajte i gdje GPU danas nailazi na granice: memorijska propusnost raste stotinu puta sporije od računalne snage. Poglavlja 00:00 Ploča s rezultatima 02:11 Silicij: fizički predmet 05:37 Tisuće jezgri: arhitektura za propusnost 09:30 Kako GPU postaje opće računalo: Brook i CUDA 12:33 Dvije kartice, milijun i dvjesto tisuća slika: AlexNet 2012. 16:48 Matrica u hardware-u: Tensor jezgre i Volta 20:30 GPU naspram TPU-a: fleksibilnost naspram specijalizacije 25:03 Zid memorije, trošak snage i otvorena pitanja 28:20 Isti čip, drukčiji zadatak Izvori - ImageNet / image-net.org — ILSVRC 2012 Classification Task Results: https://image-net.org/challenges/LSVRC/2012/results.html - ImageNet / Stanford University — ImageNet Large Scale Visual Recognition Challenge 2012: https://image-net.org/challenges/LSVRC/2012/ - NeurIPS / Curran Associates — ImageNet Classification with Deep Convolutional Neural Networks (NeurIPS 2012 proceedings page): https://proceedings.neurips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html - NeurIPS / Curran Associates — ImageNet Classification with Deep Convolutional Neural Networks — NeurIPS proceedings: https://proceedings.neurips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html - NeurIPS / Curran Associates — ImageNet Classification with Deep Convolutional Neural Networks: https://proceedings.neurips.cc/paper_files/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf - technical.city — NVIDIA GeForce GTX 580 specifications: https://technical.city/en/gpu/GeForce-GTX-580 - Grokipedia / Wikipedia — GeForce 256 — Grokipedia (mirror of Wikipedia content): https://grokipedia.com/page/GeForce_256 - PC Tech Vault — GeForce 256 DDR GPU Specifications — PC Tech Vault: https://pctechvault.com/gpu/nvidia-geforce-256-ddr/ - Grokipedia / Wikipedia — GeForce 256 — Grokipedia (mirrors Wikipedia): https://grokipedia.com/page/GeForce_256 - NVIDIA Developer Blog — NVIDIA Hopper Architecture In-Depth: https://developer.nvidia.com/blog/nvidia-hopper-architecture-in-depth/ - NVIDIA Newsroom (official press release) — NVIDIA Announces Hopper Architecture, the Next Generation of Accelerated Computing: https://nvidianews.nvidia.com/news/nvidia-announces-hopper-architecture-the-next-generation-of-accelerated-computing - Nanowerk — Transistors explained, and what a 2 nm process node really means: https://www.nanowerk.com/transistors-explained.php - TechPowerUp — NVIDIA Hopper Whitepaper Reveals Key Specs of Monstrous Compute Processor: https://www.techpowerup.com/294503/nvidia-hopper-whitepaper-reveals-key-specs-of-monstrous-compute-processor - Wikipedia — Fabless manufacturing — Wikipedia: https://en.wikipedia.org/wiki/Fabless_manufacturing - Design-Reuse (reproducing NVIDIA/TSMC joint press release) — NVIDIA and TSMC Ship One-Billionth GeForce Graphics Processor: https://www.design-reuse.com/news/202519606-nvidia-and-tsmc-ship-one-billionth-geforce-graphics-processor/ - HEXUS (reproducing press release quotes) — NVIDIA and TSMC Ship One-Billionth GeForce Graphics Processor — HEXUS: https://hexus.net/tech/news/graphics/28468-nvidia-tsmc-ship-billionth-geforce-gpu/ - ThunderCompute — NVIDIA GPU Architecture Explained – ThunderCompute: https://www.thundercompute.com/blog/nvidia-gpu-architecture-explained - NVIDIA (official documentation) — Programming Model – CUDA Programming Guide (NVIDIA official documentation): https://docs.nvidia.com/cuda/cuda-programming-guide/01-introduction/programming-model.html - NVIDIA (official documentation) — Writing SIMT Kernels – CUDA Programming Guide (NVIDIA official documentation): https://docs.nvidia.com/cuda/cuda-programming-guide/02-basics/writing-cuda-kernels.html - Yosef Kreinin (personal technical blog, widely cited in GPU literature) — SIMD ‹ SIMT ‹ SMT: parallelism in NVIDIA GPUs – Yosef Kreinin: https://www.yosefk.com/blog/simd-simt-smt-parallelism-in-nvidia-gpus.html - Jason McKesson (paroj.github.io) — Vertices and Registers: GeForce 256 – gltut (OpenGL tutorial history): https://paroj.github.io/gltut/History%20GeForce.html - Microsoft (official press release) — Microsoft Announces Release of DirectX 8.0: https://news.microsoft.com/source/2000/11/09/microsoft-announces-release-of-directx-8-0/ - reference.org — GeForce 3 series – reference.org: https://reference.org/facts/GeForce_3_series/3dJXIwRV - … Napomene - Naracija: sintetski (AI) glas. Glazba: originalna, sintetizirana za ovaj video. - Sheme su pojednostavljeni prikazi. #dokumentarac #GPU #dubokoucenje

Pogledaj na YouTubeu