What's the biggest bottleneck you hit - GPU compute, memory bandwidth, or network latency for asset streaming? Curious how it compares to native WebGPU.