你好，我在使用沐曦的N260和NVIDIA的RTX3090使用av库进行硬件解码的时候发现，3090的显存增加了300MB作用，而N260的显存才增加了50MB左右，这是为什么
2025年8月25日
5
-
-
尊敬的开发者您好，麻烦您给出操作系统，系统内核，mx-smi信息，docker info信息，镜像名称，容器启动命令，程序代码，程序运行命令
-
RTX 3090
Ubuntu18.04
docker info
Client: Debug Mode: false Server: Containers: 48 Running: 43 Paused: 0 Stopped: 5 Images: 66 Server Version: 19.03.12 Storage Driver: overlay2 Backing Filesystem: extfs Supports d_type: true Native Overlay Diff: true Logging Driver: json-file Cgroup Driver: systemd Plugins: Volume: local Network: bridge host ipvlan macvlan null overlay Log: awslogs fluentd gcplogs gelf journald json-file local logentries splunk syslog Swarm: inactive Runtimes: nvidia runc Default Runtime: nvidia Init Binary: docker-init containerd version: 7f7fdf5fed64eb6a7caf99b3e12efcf9d60e311c runc version: v1.1.14-0-g2c9f560 init version: fec3683 Security Options: apparmor seccomp Profile: default Kernel Version: 5.15.0-139-generic Operating System: Ubuntu 20.04.6 LTS OSType: linux Architecture: x86_64 CPUs: 48 Total Memory: 251.5GiB Name: leatop-225 ID: EPP5:RNZH:YKB7:EBCF:AQGF:3UTC:5PJ7:PR4K:YES4:I42B:TK3A:K6V5 Docker Root Dir: /var/lib/docker Debug Mode: false Registry: https://index.docker.io/v1/ Labels: Experimental: false Insecure Registries: harbor.dubhe.com 127.0.0.0/8 Registry Mirrors: https://hub.geekery.cn/ https://ghcr.geekery.cn/ https://hub.uuuadc.top/ https://docker.anyhub.us.kg/ https://hub-mirror.c.163.com/ https://registry.aliyuncs.com/ https://docker.mirrors.ustc.edu.cn/ https://6kx4zyno.mirror.aliyuncs.com/ https://mirrors.tuna.tsinghua.edu.cn/ Live Restore Enabled: true
镜像使用的是自己的
ffmpeg的版本为4.3.2
PyAV的版本为14.4.0
程序代码为：
def hw_decode_only_for_monitoring(input_file): """ 强制使用 h264_cuvid 进行硬件解码，但不进行任何后续处理。 脚本会持续运行以便于使用 nvidia-smi 进行监控。 """ container = None try: options = { 'rtsp_transport': 'tcp', 'stimeout': '50000-000', # 5秒超时 } container = av.open(input_file, options=options) in_stream = container.streams.video[0] logging.info(f"输入流信息: {in_stream.width}x{in_stream.height}, @ {in_stream.average_rate} fps") # --- 1. 强制创建 h264_cuvid 解码器上下文 (您的原始逻辑) --- decoder_ctx = av.CodecContext.create("h264_cuvid", "r") if in_stream.codec_context.extradata: decoder_ctx.extradata = in_stream.codec_context.extradata logging.info("已成功创建并配置 h264_cuvid 硬件解码器。") # --- 2. 纯解码循环 --- FRAME_LIMIT = 1000 # 增加帧数限制，以便有足够的时间进行观察 processed_count = 0 started = False logging.info("="*60) logging.info("解码已开始。请现在打开另一个终端并运行 'nvidia-smi'。") logging.info("推荐使用持续监控命令: 'watch -n 1 nvidia-smi'") logging.info("="*60) for packet in container.demux(video=0): if packet.dts is None: continue if not started and packet.is_keyframe: logging.info("接收到第一个关键帧，解码正在全速进行...") started = True if not started: continue if processed_count >= FRAME_LIMIT: logging.info(f"已达到 {FRAME_LIMIT} 帧的处理上限。") break for frame in decoder_ctx.decode(packet): # 我们只解码，不对解码后的帧做任何事 # The 'frame' object exists in GPU VRAM at this point. processed_count += 1 # 为了不在控制台刷屏太快，可以只在特定间隔打印 if processed_count % 100 == 0: print(f"已解码 {processed_count} / {FRAME_LIMIT} 帧...") if processed_count >= FRAME_LIMIT: break logging.info("解码循环结束。") # 在程序退出前暂停，给你最后一次查看nvidia-smi的机会 input("按回车键退出程序...") except Exception as e: logging.error(f'程序发生严重错误: {e}', exc_info=True) finally: # --- 3. 释放资源 --- if container: container.close() logging.info("输入流容器已关闭。") if __name__ == "__main__": main() -
尊敬的开发者您好，麻烦您给出沐曦卡操作系统，系统内核，mx-smi信息，docker info信息，镜像名称，容器启动命令，程序代码，程序运行命令
-
N260
Ubuntu22.04
mx-smi的信息见下图
docker info
Client: Docker Engine - Community Version: 27.4.0 Context: default Debug Mode: false Plugins: buildx: Docker Buildx (Docker Inc.) Version: v0.19.2 Path: /usr/libexec/docker/cli-plugins/docker-buildx compose: Docker Compose (Docker Inc.) Version: v2.31.0 Path: /usr/libexec/docker/cli-plugins/docker-compose Server: Containers: 32 Running: 12 Paused: 0 Stopped: 20 Images: 59 Server Version: 27.4.0 Storage Driver: overlay2 Backing Filesystem: extfs Supports d_type: true Using metacopy: false Native Overlay Diff: true userxattr: false Logging Driver: json-file Cgroup Driver: systemd Cgroup Version: 2 Plugins: Volume: local Network: bridge host ipvlan macvlan null overlay Log: awslogs fluentd gcplogs gelf journald json-file local splunk syslog Swarm: inactive Runtimes: io.containerd.runc.v2 runc Default Runtime: runc Init Binary: docker-init containerd version: 88bf19b2105c8b17560993bee28a01ddc2f97182 runc version: v1.2.2-0-g7cb3632 init version: de40ad0 Security Options: apparmor seccomp Profile: builtin cgroupns Kernel Version: 5.15.0-72-generic Operating System: Ubuntu 22.04.3 LTS OSType: linux Architecture: x86_64 CPUs: 104 Total Memory: 251.5GiB Name: ictrek ID: 15ce4c68-5b5c-4e19-80ea-746f67744c6d Docker Root Dir: /nfs/docker-data Debug Mode: false Experimental: false Insecure Registries: harbor.testplatform.com:8443 127.0.0.0/8 Registry Mirrors: https://82m9ar63.mirror.aliyuncs.com/ https://ccr.ccs.tencentyun.com/ https://docker.gs/ https://harbor.testplatform.com:8443/ https://docker.m.daocloud.io/ https://docker.1ms.run/ https://docker.xuanyuan.me/ Live Restore Enabled: false Default Address Pools: Base: 192.168.0.0/16, Size: 24 WARNING: bridge-nf-call-iptables is disabled WARNING: bridge-nf-call-ip6tables is disabled
镜像为自己定义的
ffmpeg的版本为4.4.2
PyAV的版本为12.2.0
程序代码：
import av import argparse import logging def parase_and_check(): """解析命令行参数并检查环境""" # 移除了 --output 参数，因为我们不再写入文件 parser = argparse.ArgumentParser(description='视频硬件解码工具（使用MXVID）') parser.add_argument('-i', '--input', required=True, help='输入RTSP URL或文件路径') args = parser.parse_args() # 检查 h264_mxvid 解码器是否可用 if 'h264_mxvid' not in av.codecs_available: print("致命错误: h264_mxvid 硬件解码器不可用。请检查驱动和PyAV环境。") exit() else: print("成功检测到 h264_mxvid 解码器！") # 设置日志级别，可以帮助调试 av.logging.set_level(av.logging.INFO) logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s') return args def get_nal_unit_type(packet_data): """从H.264数据包中解析NAL单元类型""" # 确保packet_data是bytes类型 if not isinstance(packet_data, bytes): packet_data = bytes(packet_data) start_code_len = 4 if packet_data.startswith(b"\x00\x00\x00\x01") else 3 # 增加长度检查，防止索引越界 if len(packet_data) <= start_code_len: return -1 # 无效或不完整的数据包 nal_header = packet_data[start_code_len] nal_unit_type = nal_header & 0x1F # 低 5 位是类型 return nal_unit_type def hardware_decode_only(input_file): """ 使用指定的硬件解码器进行解码，但不进行任何编码或写入操作。 """ container = None try: logging.info(f"正在打开输入流: {input_file}") options = { 'rtsp_transport': 'tcp', 'fflags': 'nobuffer', 'flags': 'low_delay', } container = av.open(input_file, options=options) video_stream = container.streams.video[0] logging.info(f"视频流信息: {video_stream.width}x{video_stream.height} @ {video_stream.average_rate} fps") logging.info("正在创建 h264_mxvid 硬件解码器上下文...") codec_ctx = av.CodecContext.create("h264_mxvid", "r") logging.info("硬件解码器创建成功！") frame_count = 0 pps_sps_ready = False # 使用更清晰的变量名 logging.info("开始解码视频流，等待第一个SPS NAL单元...") for demux_packet in container.demux(video_stream): # 您的原始逻辑：等待SPS/PPS以确保解码器可以正确初始化 if not pps_sps_ready: nal_type = get_nal_unit_type(demux_packet) if nal_type == 7: # NAL unit type 7 is SPS logging.info("检测到SPS，开始正式解码！") pps_sps_ready = True # 只有在接收到SPS后才开始送入数据包进行解码 if pps_sps_ready: try: # 将数据包送入解码器 for frame in codec_ctx.decode(demux_packet): # 解码成功，我们只计数，不做其他任何事 frame_count += 1 # 为了不在控制台刷屏太快，每100帧打印一次进度 if frame_count % 100 == 0: print(f"已成功解码 {frame_count} 帧...") except Exception as e: # 捕获解码单帧时可能发生的错误 logging.warning(f"解码一帧时出错，已跳过: {e}") continue # --- 移除了所有与输出、编码、mux相关的代码 --- print("\n------------------------------------") logging.info(f"解码流程完成，共解码 {frame_count} 帧。") return True except Exception as e: logging.error(f"发生严重错误: {e}", exc_info=True) return False finally: # 确保输入容器总是被关闭 if container: container.close() logging.info("输入流已关闭，资源已释放。") def main(): """主程序入口""" args = parase_and_check() if hardware_decode_only(args.input): print("\n任务成功完成！") else: print("\n任务失败。") if __name__ == "__main__": main() -
系统内核为5.15.0-72-generic