[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f5p0awgue1262":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":12,"bodyZh":13,"bodyEn":14,"sourceName":15,"sourceUrl":16,"updatedAt":17},"llamacpp-b12443-nvidia-vulkan-fp32-fa-shmem-2026-11-17","RUNTIME","2026-11-16T23:05:58.618Z","llama.cpp b12443修复NVIDIA Vulkan共享内存溢出：不是显存装不下模型","llama.cpp b12443 fixes NVIDIA Vulkan FP32 Flash Attention shared-memory overflow","11月16日合入的修复在共享内存需求超出设备上限时关闭staging。它针对特定Vulkan Flash Attention路径，不等于所有NVIDIA显卡提速，也不是扩大显存容量。","Merged on November 16, the fix disables shared-memory staging when its requirement exceeds device capacity. It targets a particular Vulkan Flash Attention path, not model VRAM capacity or a universal NVIDIA speedup.",[],[],"llama.cpp在11月16日合入PR #32244，并于当天发布的b12443中列入一项NVIDIA Vulkan修复：当Flash Attention所需的共享内存超过设备可用上限时，禁用shmem staging。PR明确指向不具备该路径所需FP16能力的NVIDIA设备。这是已合入的上游代码变更，不是本站真机复现，也没有可据此发布的通用提速百分比。\n\n先分清两种“内存”。模型权重、KV cache通常关乎显存容量；此处的shared memory则是GPU工作组内使用的局部资源。两者不是一个容量池。因而，一张卡即使有足够显存容纳权重，特定计算内核仍可能触及共享内存限制。看到溢出报错就换更小量化，未必是在修真正的瓶颈。\n\n这次修法也不是给设备增加资源，而是在需求超过上限时不走相应staging路径。它说明该配置需要另一条执行路径，不能推导为“所有Flash Attention都失效”，更不能延伸为CUDA后端同样受影响。PR正文没有给出跨卡测试表，因此本站不补写受益型号清单。\n\n对正在排错的用户，较稳妥的做法是保留原构建，记录完整GPU型号、驱动、Vulkan能力及失败日志，再用b12443或包含该修复的后续构建复测。模型文件、量化、上下文长度和启动参数保持不变，先确认失败是否消失，再讨论速度差异；不能把一次启动成功写成长期稳定。\n\n截至本次采集，原PR没有公开评论中的独立复现。本站据发布记录确认“修复已进入构建”，不承诺每种旧卡都已验证，也不将这条工程修复当成新的硬件跑分。\n\n来源：https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fpull\u002F32244\n发布：https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Freleases\u002Ftag\u002Fb12443","On November 16, llama.cpp merged PR #32244 and included it in release b12443. The change disables shared-memory staging when Flash Attention would require more shared memory than the device provides. The PR specifically mentions NVIDIA devices lacking FP16 capability for this path. This is an upstream code change, not a hardware reproduction by AraoAI; it supplies no general speedup percentage.\n\nShared memory here is not the same resource as VRAM used for model weights and KV cache. It is local storage used by GPU workgroups. A model can fit in VRAM while a particular kernel still exceeds its shared-memory limit. Choosing a smaller quantization therefore does not necessarily address this failure.\n\nThe patch does not increase device capacity. It avoids the staging path when that path needs too much shared memory. This should not be read as evidence that all Flash Attention is broken, or that CUDA has the same problem. The PR provides no cross-GPU results, so we do not invent a list of validated cards.\n\nFor troubleshooting, keep the old build and record the GPU, driver, Vulkan capabilities and failure log. Retest with b12443 or a later build containing the fix while holding the model, quantization, context and launch settings constant. First establish whether the failure disappears; only then compare performance. A successful startup is not proof of long-run stability.\n\nAt collection time the PR had no public independent reproduction in its comments. The release verifies that the fix shipped, not that every older card has been tested.\n\nSource: https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fpull\u002F32244\nRelease: https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Freleases\u002Ftag\u002Fb12443","llama.cpp GitHub PR与发布记录","https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fpull\u002F32244","2026-11-16T23:05:58.619Z"]