
What I love:
Instant, GPU‑ready environment – Out of the box, these VM images include the full ML stack: TensorFlow, PyTorch, scikit‑learn, CUDA, cuDNN, NCCL, NVIDIA drivers, JupyterLab, and more
Wide framework and hardware support – Choose from CPUs or GPUs, and frameworks like TensorFlow Enterprise, PyTorch, HPC images—it’s easy to match your hardware and software needs
Smooth integration with Google Cloud – Seamless links to BigQuery, Vertex AI, Cloud Storage, and other services keep my workflow fluid and efficient
Zero setup time – The time saved from bypassing manual installs of drivers, frameworks, and Jupyter is huge—educes friction and eliminates compatibility headaches Review collected by and hosted on G2.com.
While the Deep Learning VM Image is powerful and saves a lot of setup time, there are some areas for improvement:
Costs can escalate quickly, especially when using GPU instances for long-running jobs or accidentally leaving VMs running.
Customization is sometimes limited; installing very specific package versions or uncommon libraries can break pre-configured dependencies.
Startup times for large VMs (especially with GPUs) can occasionally feel slow compared to some other cloud providers.
Managing quota limits and regional availability of GPUs sometimes interrupts workflows if capacity is limited in a zone.
It’s a great tool for Google Cloud users, but users coming from AWS or Azure might face a learning curve with GCP’s networking and IAM model. Review collected by and hosted on G2.com.