[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fldy5xq13ec6a":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":15,"bodyZh":22,"bodyEn":23,"sourceName":24,"sourceUrl":25,"updatedAt":26},"qengine-flash-next-graph-oom-margin-2026-10-30","HARDWARE","2026-10-29T23:09:41.874Z","3090混合双卡社区排错：给CUDA graph留余量，比把显存塞满更重要","Community mixed-3090 report: CUDA graph headroom matters more than filling VRAM","10月29日新增的 QEngine 报告定位到 CUDA graph 显存分配失败。作者将专家缓存旁预留量从256改为1024MB后完成整轮测试，同时付出速度代价；仅适用于该报告条件，本站未复现。","An October 29 QEngine report identifies a CUDA graph allocation failure. Increasing the expert-cache margin from 256 to 1024MB allowed a full run, at a speed cost. This is one community report, not an AraoAI reproduction.",[12],{"id":13,"name":14},11,"Qwen3.8-Flash-Next",[16,19],{"id":17,"name":18},2,"NVIDIA RTX 3090",{"id":20,"name":21},9,"NVIDIA RTX 3090 Ti","10月29日18:37 UTC，club-3090 #1846 新增复测评论：RTX 3090与3090 Ti组合运行 Qwen3.8-Flash-Next，QEngine 在 CUDA graph 实例化时显存耗尽。本文只采用当日新评论，不把原帖前一天的跑分当新测量。\n\n作者固定同一镜像digest及 `FLASH_COMM=ce FLASH_MAX_SEQS=4 FLASH_KV_ARCHIVE_GB=0`。日志先出现 `graph instantiate out of memory, other graphs dropped`，随后在 `flash_seq.cu:298` 报内存分配失败并终止。它与 `FLASH_COMM=auto` 下的非有限数故障不是同一问题；换通信模式，并不能保证消除后续显存分配风险。\n\n关键对照是 `FLASH_EC_MARGIN_MB`。作者把专家缓存旁每卡预留量从默认256改为1024MB后，重新创建容器并完成包含两种思考设置与持续运行检查的整轮测试，自报期间重启次数为0。该次同机A\u002FB的叙事任务decode从141.8变为127.8 tok\u002Fs，10K预填充从1159变为1096 tok\u002Fs。两项属于不同测试阶段，不拼成一条标准跑分；它们说明稳定性余量有代价，不是免费提速。\n\n不能反推“1024MB适合所有机器”。这只有一位报告者在一套混合双卡配置上的结果；维护者另见同类预警而未崩溃，也说明预警不等于必然退出。报告中的静态空闲显存还受启动时KV池和专家缓存分配影响，不能只因数字小就判运行失败。相关PR #1856 当次核验尚未合并，且只改说明，不改引擎默认值。\n\n排错时保存镜像digest、全部环境变量、致命日志与容器重启计数。自动重启可能掩盖中途崩溃，“现在能请求”不等于整轮测完。QEngine当前许可为 PolyForm Noncommercial 1.0.0，有非商业使用限制，部署前需核对许可。本站未独立复现，不把本例扩成硬件购买建议。\n\n来源：[当日复测评论](https:\u002F\u002Fgithub.com\u002Fnoonghunna\u002Fclub-3090\u002Fissues\u002F1846#issuecomment-6505231322)；[说明PR](https:\u002F\u002Fgithub.com\u002Fnoonghunna\u002Fclub-3090\u002Fpull\u002F1856)；[引擎许可](https:\u002F\u002Fgithub.com\u002Fbpx03\u002FQEngine\u002Fblob\u002Fmain\u002FLICENSE)。","At 18:37 UTC on October 29, a new comment on club-3090 issue #1846 supplied a fatal log and a same-day rerun. A mixed RTX 3090 and RTX 3090 Ti setup serving Qwen3.8-Flash-Next through QEngine exhausted GPU memory while instantiating CUDA graphs. The issue body contains the previous day's measurements; this article uses the new comment within the current window.\n\nThe author kept the image digest and `FLASH_COMM=ce FLASH_MAX_SEQS=4 FLASH_KV_ARCHIVE_GB=0` fixed. The engine first logged `graph instantiate out of memory, other graphs dropped`, then reported an allocation failure at `flash_seq.cu:298` and aborted. This is distinct from the non-finite-value failure under `FLASH_COMM=auto`. Changing communication mode does not guarantee enough memory for later graph allocations.\n\nThe controlled setting was `FLASH_EC_MARGIN_MB`, the per-card margin beside the expert cache. After increasing it from the default 256 to 1024MB and recreating the container, the author reports a complete matrix, both thinking settings and a soak run with zero restarts. Narrative decode changed from 141.8 to 127.8 tok\u002Fs; 10K prefill changed from 1159 to 1096 tok\u002Fs. These are different test phases, not a single paired benchmark record. They demonstrate a speed cost for headroom, not a free optimization.\n\nThe result does not establish 1024MB as a universal setting. It is one reporter on one mixed-card rig. A maintainer saw the same warning without a crash, so the warning alone is not proof that failure is inevitable. Static free VRAM also reflects boot-time KV-pool and expert-cache allocations. PR #1856 was unmerged when checked and changes documentation only, not engine defaults.\n\nKeep the exact image digest, environment, fatal logs and restart counter. Automatic restart can conceal a failed evaluation: a responsive endpoint afterward does not prove that the full run completed. QEngine currently uses PolyForm Noncommercial 1.0.0, with noncommercial restrictions that must be checked before deployment. AraoAI has not reproduced this report and does not present it as purchasing advice.\n\nSources: [same-day rerun](https:\u002F\u002Fgithub.com\u002Fnoonghunna\u002Fclub-3090\u002Fissues\u002F1846#issuecomment-6505231322); [documentation PR](https:\u002F\u002Fgithub.com\u002Fnoonghunna\u002Fclub-3090\u002Fpull\u002F1856); [engine license](https:\u002F\u002Fgithub.com\u002Fbpx03\u002FQEngine\u002Fblob\u002Fmain\u002FLICENSE).","club-3090 independent community report","https:\u002F\u002Fgithub.com\u002Fnoonghunna\u002Fclub-3090\u002Fissues\u002F1846#issuecomment-6505231322","2026-10-29T23:09:41.875Z"]