#verarubin — Public Fediverse posts

The financial architecture of next-generation AI infrastructure has fundamentally inverted. NVIDIA's upcoming liquid-cooled Vera Rubin NVL72 rack is projected to command an estimated $7.8 million—effectively doubling procurement costs over a single product cycle.

While public attention remains fixed on the premium price of GPU silicon, bottom-up supply chain tracking reveals that the most aggressive capital accumulation is occurring within the memory subsystem. Aggregate allocations for HBM4 and system LPDDR5X have experienced a staggering 435% cost explosion, now devouring nearly 30% of the total rack bill of materials.

https://www.buysellram.com/blog/why-nvidias-vera-rubin-racks-cost-double-and-memory-is-the-surprise/

#DataCenter #AIInfrastructure #NVIDIA #VeraRubin #Semiconductors #HardwareEconomics #ITAD #TechProcurement #ServerRAM #EnterpriseStorage #tech

#Intel Announces "#CrescentIsland" #Inference-Optimized #Xe3P #GPU With 160GB vRAM, but it's not shipping until at least H2'2026
Given the timing this will be going up against the AMD Instinct #MI450 series and NVIDIA #VeraRubin.
This new graphics card is built on next-gen Xe3P Celestial micro-architecture. Xe3P will be optimized around performance-per-Watt and Crescent Island will feature 160GB of LPDDR5x memory to allow plenty of space for #largelanguagemodels (#LLM).
https://www.phoronix.com/review/intel-crescent-island