QubGPU Benchmarks — Share Your Performance Results

⚡ QubGPU — Neural Datagram Protocol is designed for extreme throughput on modern GPU clusters. This thread is the community benchmark board.

When sharing results, please include:

• GPU model and VRAM
• Driver version and CUDA/ROCm version
• Batch size and precision (FP32 / BF16 / INT8)
• Throughput (tokens/sec or TFLOPS)
• Latency P50 / P95 / P99
• Any custom kernel patches applied

Comparing implementations? Use the standardised qubgpu-bench CLI tool included in the SDK — it ensures reproducible results across environments.