[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f18z77ybp2129r":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":12,"bodyZh":19,"bodyEn":20,"sourceName":21,"sourceUrl":22,"updatedAt":23},"llamacpp-windows-rdna-vulkan-fa-crash-2026-11-16","RUNTIME","2026-11-15T23:08:59.991Z","Windows上的AMD Vulkan新回归报告：7800 XT的绕行参数，9070 XT未必能用","Windows AMD Vulkan regression reports: an RX 7800 XT workaround does not transfer to RX 9070 XT","llama.cpp社区出现两份Windows专有驱动下的Flash Attention崩溃报告。两个型号的绕行结果不同，尚不能归结为所有AMD卡的统一问题。","Two community reports describe Flash Attention crashes with the proprietary AMD Vulkan driver on Windows. Workarounds differ between the GPUs; this is not evidence of a universal AMD failure.",[],[13,16],{"id":14,"name":15},16,"AMD Radeon RX 9070 XT",{"id":17,"name":18},31,"AMD Radeon RX 7800 XT","11月15日UTC，llama.cpp issue #32353新增一份Windows Vulkan崩溃报告，随后另一位用户补充RDNA4上的类似现象。它们是社区自报，不是本站复现，也还不能证明所有AMD显卡都有同一根因。短提示能出字，不等于实际输入长度下稳定可用。\n\n首份报告使用RX 7800 XT 16GB、Windows 11 Pro 10.0.26300和AMD专有驱动amdvlk64.dll 32.0.32015.16001；引擎是自编译b12421、提交75fd7f9ec。测试权重为Gemma4-26B-A4B的Uncensored-HauhauCS-Balanced-IQ4_XS衍生版本，不能当成原版Gemma4的记录。作者称短提示可运行，较长输入在prefill阶段触发驱动模块异常；旧b12370在其测试中正常。\n\n关键差异在评论：RX 7800 XT作者强制GGML_VK_FORCE_FA_FLIP=kvsep后能继续运行，但RX 9070 XT用户用官方b12425包、gpt-oss-20b MXFP4测试时，direct与kvsep均崩溃，关闭FA才通过。两份报告使用不同权重和构建，因此不能把其中一人的绕行参数写成通用修复，更不能拿原文速度数值拼成跨卡比较。\n\n排障先保存当前可用构建，再记录模型完整文件名、驱动、启动命令和触发输入长度；同一台机器一次只改一个设置。若关闭FA测试，KV类型也必须确认后端支持，并记录随之发生的配置变化。暂时能绕过不等于漏洞已被修复，报告将异常指向驱动模块也不等于已完成代码二分定位。\n\n本次只发布兼容性提醒，不把崩溃当正常跑分，不给没有测过的卡下结论。来源及两位报告者的具体命令、日志与差异均见原帖和评论：https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F32353","On 15 November UTC, llama.cpp issue #32353 reported a Windows Vulkan crash, followed by a second user's RDNA4 report. These are community observations, not AraoAI reproductions or proof of one root cause across all AMD GPUs. A successful short prompt does not establish stability at realistic input lengths.\n\nThe first setup was an RX 7800 XT 16GB, Windows 11 Pro 10.0.26300, and the proprietary AMD driver amdvlk64.dll 32.0.32015.16001. It used self-built b12421, commit 75fd7f9ec, and the Gemma4-26B-A4B Uncensored-HauhauCS-Balanced IQ4_XS derivative—not the original Gemma4 weights. The author reported that longer prompts crashed during prefill while b12370 worked in that test.\n\nThe important distinction appears in the comments. Forcing GGML_VK_FORCE_FA_FLIP=kvsep worked for the RX 7800 XT reporter. A second reporter, using an RX 9070 XT, official b12425 binaries and gpt-oss-20b MXFP4, saw both direct and kvsep paths crash; disabling FA worked there. Different models and builds prevent treating either workaround as a universal fix or combining their speeds into a GPU comparison.\n\nKeep a known-working build before testing an update. Record the full weight filename, driver, launch command and triggering input length; change one setting at a time on the same machine. When testing FA disabled, verify that the KV types are supported and document any additional configuration changes. A workaround is not a confirmed upstream fix, and a fault in the driver module is not a completed source-code bisection.\n\nThis is a compatibility notice, not a clean benchmark or a claim about untested cards. Original commands, logs and both reports: https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F32353","llama.cpp GitHub issue及窗口内评论","https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F32353","2026-11-15T23:08:59.992Z"]