|
白嫖了一张quadro T600,满载40W功耗,尝试放进nas里看看能做什么
之前发的一篇用核显跑frigate识别加速的一个在pve lxc中跑frigate+核显加速的详细指南 - 『HomeAssistant』综合讨论区 - 『瀚思彼岸』» 智能家居技术论坛 - Powered by Discuz!
这次换成N卡跑cuda看看效果
这次平台是物理黑群晖,使用矿神源的N卡驱动,版本535,cuda12.2,注意frigate要求的nvidia gpu计算能力版本要>5,型号大概就是9系之后,具体型号支持可以查询CUDA GPUs - Compute Capability | NVIDIA Developer,我这张T600是20系的图灵架构,计算能力版本7.5,符合条件
4路 2K 25fps录制+4路 720P 6fps识别,显存占用500M左右,使用的是cc150 cpu 8c16t,占用基本可以忽略,GPU触发检测时大概占用10~20%,日常功耗多个十来瓦
检测延迟20ms出头,比核显高一些,貌似跑的模型不同,核显跑只识别到过人和车,相同配置n卡跑可以跑出来猫狗
(n卡跑转码也给力,jellyfin里同一部4K H265 10bit的片,加上HDR转换,核显转半天转不出来画面。换NVENC转码速度可以去到100多帧)
docker-compose部署参考:
version: "3.9"
services:
frigate:
container_name: frigate
privileged: true # this may not be necessary for all setups
restart: unless-stopped
image: ghcr.io/blakeblackshear/frigate:stable-tensorrt
deploy: # <------------- Add this section
resources:
reservations:
devices:
- driver: nvidia
# device_ids: ['0'] # this is only needed when using multiple GPUs
#count: 1 # number of GPUs
capabilities: [gpu]
shm_size: "128mb" # update for your cameras based on calculation above
volumes:
- /etc/localtime:/etc/localtime:ro
- /volume1/docker/frigate/config:/config
- /volume4/NVR/frigate:/media/frigate
- type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
target: /tmp/cache
tmpfs:
size: 1000000000
ports:
- "8971:8971"
- "5555:5000" # Internal unauthenticated access. Expose carefully.
- "8554:8554" # RTSP feeds
- "8555:8555/tcp" # WebRTC over tcp
- "8555:8555/udp" # WebRTC over udp
environment:
- NVIDIA_VISIBLE_DEVICES=all # 映射NVIDIA显卡
- NVIDIA_DRIVER_CAPABILITIES=all # 映射NVIDIA显卡
- FRIGATE_RTSP_PASSWORD=password
frigate内检测器的相关配置:
其他配置可参考核显那篇,第一次启动需要构建模型,等待十来分钟的样子
detectors:
tensorrt:
type: tensorrt
device: 0 #This is the default, select the first GPU
detect:
width: 1280
height: 720
fps: 6
enabled: true
model:
path: /config/model_cache/tensorrt/yolov7-320.trt
input_tensor: nchw
input_pixel_format: rgb
width: 320
height: 320
|
评分
-
查看全部评分
|