A discussion on X about potential shifts in HBM (High Bandwidth Memory) demand for AI systems. The analysis suggests that rather than continuously increasing HBM capacity per GPU, future designs may use smaller HBM stacks combined with interconnect technologies like NVLink to distribute memory across larger clusters. This could pressure HBM wafer consumption by 2028 as new production capacity from SK Hynix, Micron, and Samsung comes online, though long-term contracts and pricing may protect memory manufacturers' margins.
Citi projects memory undersupply through 2031 driven by AI demand for HBM and DRAM, with HBM bit demand rising 62% in 2027 and 69% in 2028. Morgan Stanley estimates Nvidia will consume 37.3% of total HBM demand, followed by Google at 36% and AMD at 12.1%. Industry analysts debate whether a 2028 memory downcycle could occur, citing factors like HBM despec and new fab capacity.
AMD Helios is a GPU platform for AI with 432 GB of HBM4 memory and 40 PFLOPs of FP4 compute. This post presents an educational ladder of BF16 GEMM kernel implementations that progressively optimize for Helios's architectural features, including partitioned LDS, simplified cache hierarchy, and 32-thread waves.
Memory chip shortages are forcing smartphone and laptop manufacturers to redesign products, advance-order components, and test for counterfeit chips, with memory costs reaching 60% of device bill of materials for budget phones. Companies like Fairphone, Jolla, and Framework are adopting strategies including modular designs, batch testing, and salvaged components to secure adequate supply amid expectations the shortage will persist until 2030.
A graphics programmer at 505 Games explains the complexity of texture memory layouts when converting textures between gaming platforms. The article covers block compression, texel ordering, mipmaps, and texture tiles, demonstrating how platform-specific memory organization differs despite using the same hierarchy of building blocks.
Citi projects AI infrastructure demand will create memory and storage shortages through 2031, with HBM bit demand rising 62-69% annually while supply grows only 19-22%, leaving deficits of 8-10%. Memory manufacturers like Micron, SK Hynix, and Samsung are positioned to benefit from sustained pricing power and higher margins.
Citi forecasts severe memory shortages extending through 2031 driven by AI infrastructure demand. HBM bit demand is projected to surge 62% in 2027 and 69% in 2028, while DRAM and NAND supply will fall short of demand by 8.7-9.7% and 5.5-6.1% respectively, creating sustained pricing pressure across memory markets.