Parasail is tackling the latency and cost challenges of AI model deployment by splitting workloads between high-performance GPUs and purpose-built accelerators. The company uses NVIDIA hardware for compute-intensive prefill tasks while offloading latency-sensitive decoding to d-Matrix's Corsair platform. This model allows Parasail to maximize the utility of its current data center infrastructure without waiting for new facility construction.
In section Releases
Parasail Pairs NVIDIA GPUs with d-Matrix to Boost Inference Speeds
By integrating d-Matrix Corsair accelerators with its existing NVIDIA Hopper and Blackwell GPU fleets, inference provider Parasail is aiming for a tenfold increase in token generation speeds. This heterogeneous compute strategy seeks to optimize inference economics by assigning specific processing tasks to the hardware best suited for them.
At the core of this efficiency is the Corsair chip’s Digital In-Memory Compute architecture, which integrates processing and memory on a single silicon die. By eliminating the energy and time costs associated with data movement between separate chips, the system delivers improved performance and energy efficiency. Parasail’s proprietary kernel optimization technology manages this routing, dynamically directing workloads to the optimal hardware in real time across its network of over 40 global data centers.
Comments (0)
No comments yet. Be the first!