Mindbeam adds Nemotron 3.5 Super VL support to inference engine

Mindbeam announced support for Nvidia's Nemotron 3.5 Super VL in its Litespark Inference Engine and a version of the model adapted to generate CUDA code, which runs computations on Nvidia GPUs. The company said the engine supports tuning for throughput, latency and hardware utilization. Mindbeam reported that its adapted model passed 82.1% of KernelBench Level 2 problems, compared with 22.8% before adaptation. Its inference performance claims come from a warm-start benchmark using two Nvidia H200 GPUs.