SEAL is proven to deliver a 5x throughput improvement on industry-standard recommendation model benchmarks under real-world traffic
SEAL unlocks the compute power of your infrastructure by offloading memory-bound operations such as embedding table lookups
SEAL is the only OCP M.2 accelerator which can be programmed to hardware-accelerate multiple memory-intensive workloads
SEAL supports recommendation model co-location and sharding with zero performance degradation, and is complementary to next-gen AI inference accelerators
Hard latency constraints are a necessary feature of data center deployments. SEAL delivers highly deterministic latency-bounded throughput
SEAL accelerates all number formats, bit depths and quantization schemes. Compatible with all major deep learning frameworks
