RTX 5080 + RTX 3090 ACHIEVES 80 TOK/S ON QWEN

The RTX 5080 and RTX 3090 pairing delivers 80 tokens per second on the 27-billion parameter Qwen 3.6 model with 8-bit quantization. This throughput level makes the setup viable for practical LLM deployments requiring sustained generation speeds. The dual-card approach leverages NVIDIA's latest RTX 5080 architecture alongside the established RTX 3090, balancing performance and cost efficiency. Q8 quantization reduces model size while maintaining reasonable accuracy, a critical factor for fitting large models across consumer-grade GPUs. The benchmark has gained traction in developer communities, with 117 points and 43 comments on Hacker News indicating strong interest in multi-GPU inference configurations. Such setups address the gap between consumer hardware capabilities and production-scale LLM serving requirements. The achievement highlights ongoing optimization efforts in the open-source ML community, where combining older and newer generations of GPUs offers practical alternatives to enterprise-class hardware for inference workloads.