Reply to thread

Message: [QUOTE="Queen, post: 89881, member: 27"] NVIDIA apparently told its board partners to just sit on a bunch of finished RTX 50 SUPER cards instead of shipping them, and yeah, that's as wild as it sounds. At least one partner already has retail-ready units sitting in a warehouse somewhere, fully baked and ready to go, but there is no new release window from Nvidia yet. The holdup traces back to the new 3GB GDDR7 chips needed for the memory bump. NVIDIA can slot these in without touching the PCB or the bus width, just swap the old 2GB modules out, except each new chip runs $60 to $70 versus roughly $20 before. Capacity only goes up by half, yet the price tag more than triples, and multiply that by six or eight chips per card, and margins take a real hit. Three SUPER variants are apparently locked and loaded. The 5080 SUPER and 5070 Ti SUPER both land at 24GB, the 5070 SUPER bumps to 18GB, and board power sits around 415W, 350W, and 275W, respectively per Seasonic's calculator leak. The budget 5050 with 9GB seems to have quietly vanished from the lineup entirely, likely because three of those pricey chips on a cheap card just don't pencil out, so that one sounds shelved or dead. Meanwhile, Nvidia dusted off the old RTX 3060 12GB as a stopgap, with Palit and Gainward rolling out fresh batches near the original price. Blame the broader memory crunch, since Samsung, SK hynix, and Micron are all pouring capacity into AI-focused HBM instead. Jensen himself flagged GDDR7 shortages as a bottleneck. CES came and went with no SUPER cards, then late 2026 got floated, and now early 2027 is on the table. [/QUOTE]

Name