프레임당 제출된 꼭짓점이 너무 많음: 복잡한 모델 또는 모델 수가 많으면 대역폭을 더 많이 차지하고 음영 처리하는 데 더 오래 걸릴 수 있습니다.
꼭짓점 크기 문제는 평균 바이트 / 꼭짓점 트랙을 통해 진단될 수도 있으며 이 트랙은 32바이트 또는 꼭짓점을 초과하지 않는 것이 좋습니다.
그림 2: 평균 값이 31.3바이트인 단일 프레임의 평균 꼭짓점 크기
발생할 수 있는 문제를 진단하는 가장 좋은 방법은 프레임 프로필 트레이스를 사용하여 버텍스 형식을 분석하는 것입니다.
이 페이지에 나와 있는 콘텐츠와 코드 샘플에는 콘텐츠 라이선스에서 설명하는 라이선스가 적용됩니다. 자바 및 OpenJDK는 Oracle 및 Oracle 계열사의 상표 또는 등록 상표입니다.
최종 업데이트: 2025-07-27(UTC)
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["필요한 정보가 없음","missingTheInformationINeed","thumb-down"],["너무 복잡함/단계 수가 너무 많음","tooComplicatedTooManySteps","thumb-down"],["오래됨","outOfDate","thumb-down"],["번역 문제","translationIssue","thumb-down"],["샘플/코드 문제","samplesCodeIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-07-27(UTC)"],[],[],null,["# Analyze vertex memory bandwidth usage\n\nThe memory bandwidth of vertex data can be a potential bottleneck for your game's GPU performance. There are some counters in an AGI system profile that can help diagnose vertex memory bandwidth issues.\n\nQualcomm Adreno counters\n------------------------\n\nOn devices with Qualcomm Adreno GPUs, some notable counters include:\n\n| Counter | Description |\n|----------------------|---------------------------------------------------------------------|\n| Vertex Memory Read | Bandwidth of vertex data read from external memory. |\n| Average Bytes/Vertex | Average size of vertex data, in bytes. |\n| % Vertex Fetch Stall | Percentage of clock cycles where the GPU is blocked on vertex data. |\n\nARM Mali counters (WIP)\n-----------------------\n\nOn devices with ARM Mali GPUs, some notable counters include:\n\n| Counter | Description |\n|--------------------------------------------|----------------------------------------------------------------------------------------------|\n| Load/store read beats from external memory | Data beats read from external memory by the load/store unit, averaged over the shader cores. |\n| Load/store read beats from L2 cache | Data beats read from the L2 cache by the load/store unit, averaged over the shader cores. |\n| \\[More\\] | |\n\nTo calculate the overall bandwidth from average read beats, the counter value is multiplied by the bus width (typically 16 bytes) and by the total number of shader cores. \\[More\\]\n\nCounter analysis\n----------------\n\nTo measure the behavior of these counters, you can measure the average and peak\nbandwidth over the course of a single GPU frame, which can be delineated with a\ncontiguous block of **GPU Utilization**.\n**Figure 1:**Vertex memory read bandwidth for a single frame, with an average value of 327 MBps and a peak value of 1.16 GBps\n\nWe recommend a peak vertex memory read bandwidth of no higher than 1.5 GBps, and an average bandwidth no higher than 500 MBps. Higher values are indicators of one of a few common issues:\n\n- **Vertex size is too big**: Vertices may have large vertex attributes or a large number of vertex attributes, affecting vertex shading time at large.\n- **Vertex attribute streams are not split**: Vertex attributes are interleaved into a single buffer, reducing cache efficiency.\n- **Too many vertices submitted per frame**: Complex models and/or a large number of models may take up greater bandwidth and take longer to shade.\n\nVertex size issues may also be diagnosed through the **Average Bytes / Vertex**\ntrack, which we recommend to be no higher than 32 bytes or verteces.\n**Figure 2:**Average vertex size for a single frame, with an average value of 31.3 bytes\n\nThe best way to diagnose which of these problems you may be facing is by taking a frame profile trace to [analyze vertex formats](/agi/frame-trace/vertex-formats)."]]
