http://forum.lowyat.net/ Mon, 12 Oct 2026 21:44:54 +0800 FeedCreator 1.7.2 http://forum.lowyat.net/topic/5579326 Hi sifus and brothers,<br /><br />Looking for advice to build a PC mainly for local Machine Learning / AI model training. <br /><br />My specific workload involves training computer vision models on video datasets (unpacking video frames and training frame-by-frame / sequence models).<br /><br />Need your advice on the component selection:<br /><br />1. GPU: RTX 5060 Ti 16GB vs RTX 5070 12GB<br />- The 5060 Ti has weaker compute and lower memory bandwidth, but comes with 16GB VRAM.<br />- The 5070 has significantly faster compute cores, but is capped at 12GB VRAM.<br />Since video data blows up memory rapidly (batch size x frame count x resolution), is the extra 4GB VRAM strictly necessary to avoid Out-Of-Memory (OOM) crashes, or is the 5070 speed uplift worth dealing with smaller batch sizes?<br /><br />2. RAM: 32GB vs 64GB DDR5?<br />With PyTorch DataLoader extracting and staging uncompressed video frames across multiprocessing workers, will 32GB easily bottleneck, or should I just get 64GB (2x32GB) from the start?<br /><br />3. Storage: 1TB vs 2TB NVMe?<br />Factoring in Ubuntu/Windows, CUDA/PyTorch caches, checkpoints, and storing raw/unpacked frame datasets, can 1TB survive or is a 2TB Gen4 SSD mandatory?<br /><br />Target budget: Around RM [Put your RM budget here, e.g. RM 5,000 - RM 7,000]<br /><br />Appreciate any advice or feedback from anyone running local ML/CV workloads or builders here. Thanks in advance&#33;<br /> racwhee01 Hardware Tue, 06 Oct 2026 20:31:59 +0800