Runpod•4mo ago

Deploying blender on serverless doesn't utilize GPU

I am using Dockerfile to create a serverless instance with blender. Problem is the telemetry shows my GPU utilization is 0.

Testing same code in local it utilizes the GPU.

How i can be sure that serverless is not utilizing GPU is the time taken on 5090 (serverless) > 4070 (local machine) and how logs gets stuck where my GPU utilization spikes in local machine.

i have attached a telemetry image which shows utilization is 0 when i encounter same sets of logs. compared to spikes seen in local machine.

here is my dockerfile

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    wget \
    xz-utils \
    python3 \
    python3-pip \
    libxrender1 \
    libxi6 \
    libxext6 \
    libx11-6 \
    libgl1-mesa-glx \
    libxkbcommon0 \
    libsm6 \ 
    && rm -rf /var/lib/apt/lists/*

WORKDIR /opt
RUN wget -q https://download.blender.org/release/Blender4.2/blender-4.2.1-linux-x64.tar.xz \
    && tar -xf blender-4.2.1-linux-x64.tar.xz \
    && rm blender-4.2.1-linux-x64.tar.xz \
    && mv blender-4.2.1-linux-x64 blender

ENV PATH="/opt/blender:$PATH"

WORKDIR /app

COPY requirements.txt .
RUN pip3 install --no-cache-dir -r requirements.txt

COPY scripts/ ./scripts/
COPY src/ ./src/

ENV PYTHONPATH="/app/scripts:$PYTHONPATH"

EXPOSE 8000

CMD ["python3", "src/app.py"]

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    wget \
    xz-utils \
    python3 \
    python3-pip \
    libxrender1 \
    libxi6 \
    libxext6 \
    libx11-6 \
    libgl1-mesa-glx \
    libxkbcommon0 \
    libsm6 \ 
    && rm -rf /var/lib/apt/lists/*

WORKDIR /opt
RUN wget -q https://download.blender.org/release/Blender4.2/blender-4.2.1-linux-x64.tar.xz \
    && tar -xf blender-4.2.1-linux-x64.tar.xz \
    && rm blender-4.2.1-linux-x64.tar.xz \
    && mv blender-4.2.1-linux-x64 blender

ENV PATH="/opt/blender:$PATH"

WORKDIR /app

COPY requirements.txt .
RUN pip3 install --no-cache-dir -r requirements.txt

COPY scripts/ ./scripts/
COPY src/ ./src/

ENV PYTHONPATH="/app/scripts:$PYTHONPATH"

EXPOSE 8000

CMD ["python3", "src/app.py"]

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    wget \
    xz-utils \
    python3 \
    python3-pip \
    libxrender1 \
    libxi6 \
    libxext6 \
    libx11-6 \
    libgl1-mesa-glx \
    libxkbcommon0 \
    libsm6 \ 
    && rm -rf /var/lib/apt/lists/*

WORKDIR /opt
RUN wget -q https://download.blender.org/release/Blender4.2/blender-4.2.1-linux-x64.tar.xz \
    && tar -xf blender-4.2.1-linux-x64.tar.xz \
    && rm blender-4.2.1-linux-x64.tar.xz \
    && mv blender-4.2.1-linux-x64 blender

ENV PATH="/opt/blender:$PATH"

WORKDIR /app

COPY requirements.txt .
RUN pip3 install --no-cache-dir -r requirements.txt

COPY scripts/ ./scripts/
COPY src/ ./src/

ENV PYTHONPATH="/app/scripts:$PYTHONPATH"

EXPOSE 8000

CMD ["python3", "src/app.py"]

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    wget \
    xz-utils \
    python3 \
    python3-pip \
    libxrender1 \
    libxi6 \
    libxext6 \
    libx11-6 \
    libgl1-mesa-glx \
    libxkbcommon0 \
    libsm6 \ 
    && rm -rf /var/lib/apt/lists/*

WORKDIR /opt
RUN wget -q https://download.blender.org/release/Blender4.2/blender-4.2.1-linux-x64.tar.xz \
    && tar -xf blender-4.2.1-linux-x64.tar.xz \
    && rm blender-4.2.1-linux-x64.tar.xz \
    && mv blender-4.2.1-linux-x64 blender

ENV PATH="/opt/blender:$PATH"

WORKDIR /app

COPY requirements.txt .
RUN pip3 install --no-cache-dir -r requirements.txt

COPY scripts/ ./scripts/
COPY src/ ./src/

ENV PYTHONPATH="/app/scripts:$PYTHONPATH"

EXPOSE 8000

CMD ["python3", "src/app.py"]

Solution

I have figured out the issue and it had to do more with the blender version. Apparently 4.2.1 does not sit well with using GPU and CPU compared to other later versions. Switching to 4.3.2 made significant improvements in the time. According to web Blender 4.3.2 is the successor to 4.2 LTS, bringing significant usability improvements, a rewritten Grease Pencil engine for better performance, and a more flexible windowing system to Blender

run pod

1st image

local machine

2nd image