AMD Adds Day 0 Qwen3.8 Support, Radeon AI PRO R9700 Hits 51.8 Tokens per Second

AMD has demonstrated rapid support for emerging open-source models by achieving day-zero compatibility with Qwen3.8-27B on their Radeon AI PRO R9700 GPUs. The reported throughput of 51.8 tokens per second represents solid performance for local inference, comparable to high-end consumer GPU offerings and validating AMD's ROCm software stack maturity for production workloads.

This development is significant for practitioners seeking GPU vendor diversity in their local deployments. AMD's proactive optimization of new models reduces the traditional lag where NVIDIA GPUs received priority support, democratizing access to performance-optimized inference paths. For organizations with AMD infrastructure or those building hardware-agnostic deployment pipelines, day-zero model support reduces friction in adoption cycles.

The performance metrics suggest RDNA2+ architecture is viable for serious local inference work. As AMD continues closing the software maturity gap with CUDA, organizations can now confidently plan multi-GPU inference clusters with ROCm as a primary target, reducing vendor lock-in and creating competitive pressure that benefits the entire local LLM ecosystem.

Read the full article on Google News.

Source: Google News · Relevance: 9/10