d-Matrix Revolutionizes AI Inference with Lightning-Fast Architecture at OCP Summit

At the Open Compute Project Summit, d-Matrix showcased a groundbreaking live inference pipeline, achieving a 50.6% reduction in time per output token, demonstrating significant performance with its Corsair XPUs and NVIDIA H100 GPUs. Their Attention-FFN Disaggregation technique enhances efficiency for specific models, although at the cost of a proprietary software stack. The innovative SquadRack system can handle massive models, pushing the boundaries of AI technology.