[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"trend-detail-cuda-programming-en":3},{"keyword":4,"slug":5,"score":6,"change":7,"peakScore":6,"history":8,"relatedTrends":9,"relatedArticles":33},"CUDA programming","cuda-programming",500,0,[6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6],[10,14,18,22,26,29],{"keyword":11,"slug":12,"score":13},"OpenAI","openai",10000,{"keyword":15,"slug":16,"score":17},"Anthropic","anthropic",8100,{"keyword":19,"slug":20,"score":21},"Claude","claude",4029,{"keyword":23,"slug":24,"score":25},"Docker","docker",2129,{"keyword":27,"slug":28,"score":25},"Web3","web3",{"keyword":30,"slug":31,"score":32},"Cloudflare","cloudflare",1857,[34,43,51,60,68],{"id":35,"slug":36,"title":37,"summary":38,"source":39,"category":40,"cover_image":41,"image_url":41,"created_at":42},"72bb2eee-5f4c-457f-acb0-1e9017666cd9","scale-turns-cuda-code-into-portable-gpu-builds-en","SCALE turns CUDA code into portable GPU builds","I break down Spectral Compute’s SCALE compiler and show how it makes CUDA code portable across GPU vendors.","techplanet.today","tools","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784210604408-mbgn.png","2026-07-16T14:02:55.909899+00:00",{"id":44,"slug":45,"title":46,"summary":47,"source":48,"category":40,"cover_image":49,"image_url":49,"created_at":50},"9f973836-4d14-4435-b3b7-fb180e57b5fc","cuda-architecture-sms-cores-memory-en","CUDA Architecture Explained: SMs, Cores, Memory","CUDA GPUs split work across SMs, thousands of cores, and layered memory. Here’s why that design beats CPUs on parallel tasks.","oboe.com","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775197314080-mnf9.png","2026-04-03T06:21:38.505008+00:00",{"id":52,"slug":53,"title":54,"summary":55,"source":56,"category":57,"cover_image":58,"image_url":58,"created_at":59},"68bfa04a-94c4-4c8a-921c-61e93ab207aa","cuda-cp-async-ampere-hbm-latency-en","cp.async on Ampere: Hide HBM Latency on A100","Ampere’s cp.async moves data without stalling warps, cutting HBM waits from 450–600 cycles into overlapped compute on A100.","softwarefrontier.substack.com","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775167612143-4qvu.png","2026-04-02T22:06:36.521272+00:00",{"id":61,"slug":62,"title":63,"summary":64,"source":65,"category":40,"cover_image":66,"image_url":66,"created_at":67},"e05a606a-88b9-45cd-8c3e-7ad0b30b7b5d","cuda-in-2025-why-gpus-still-win-en","CUDA in 2025: Why GPUs Still Win","CUDA powers NVIDIA GPUs across AI, science, and simulation, with up to 10x weather-model speedups and deep learning gains in the thousands.","grokipedia.com","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775149432831-x799.png","2026-04-02T17:03:38.270396+00:00",{"id":69,"slug":70,"title":71,"summary":72,"source":73,"category":40,"cover_image":74,"image_url":74,"created_at":75},"5dda57f2-dfb7-4970-98ec-2e6ad298dd8c","cuda-asinf-accuracy-no-performance-hit-en","CUDA asinf() Gets More Accurate Without Slowing Down","A developer tuned asinf() for CUDA 12.8 and kept the 26-instruction baseline while improving accuracy, a rare win for GPU math.","forums.developer.nvidia.com","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775142952141-rcb7.png","2026-04-02T15:15:33.15066+00:00"]