在1400美元的AMD GPU上运行Gemma-4-31B模型，256K上下文——实测与补丁

摘要

开发者成功在售价约1400美元的AMD GPU（Radeon RX 7900 XTX）上运行Google的Gemma-4-31B模型，支持256K上下文长度。通过TurboQuant和RDNA4优化补丁，实现了可用的推理性能。该实验展示了AMD消费级GPU在大型语言模型推理中的潜力，降低了AI模型部署的硬件门槛。

来源链接

• https://github.com/KaiFelixBennett/gemma4-turboquant-rdna4
• https://www.tomshardware.com/tech-industry/cyber-security/amd-denies-researcher-a-usd10-000-bug-bounty-after-fixing-critical-auto-updater-vulnerability-security-flaw-took-124-days-to-patch
• https://www.tomshardware.com/pc-components/motherboards/various-vendors-add-amd-expo-ultra-low-latency-to-600-series-motherboards-in-latest-bios-updates-tech-tightens-memory-subtimings-on-compatible-kits-boosting-fps-by-up-to-4-percent
• https://www.tomshardware.com/pc-components/gpus/radeon-rx-9070-xt-finally-appears-in-steam-hardware-survey-rdna-4-flagship-surprisingly-lands-just-behind-rtx-5080