Tech startup proposes a novel way to tackle massive LLMs using the fastest memory available to mankind
SMRTR summary
A new AI compute platform called Corsair, developed by startup d-Matrix, offers impressive performance for large language model inference. The PCIe card features 9.6 PFLOPs of FP4 compute power and 2GB of SRAM-based memory. It uses LPDDR5 instead of expensive HBM memory, with up to 256GB per card. D-Matrix claims Corsair delivers 10x better interactive performance, 3x energy efficiency, and 3x cost-performance compared to GPUs like Nvidia's H100. Mass production is set to begin in Q2 2025.
SMRTR provides this summary for quick context. The original article belongs to TechRadar.
Read the original article