How deeply does Graph Neural Networks in Action cover memory-to-data ratio?

Defines the ratio, gives a four-to-ten-times guideline, and applies it to a concrete graph dataset and hardware decision.

Memory-to-data ratio in Graph Neural Networks in Action

The memory-to-data ratio is the quotient of available memory capacity divided by the size of the dataset being processed.

The memory-to-data ratio divides available memory capacity by the dataset size. It is a practical planning metric for checking whether hardware can hold and process a workload comfortably. section 7.3.2

Use the ratio during scale planning and evaluation:

Estimate the dataset size for each phase of the machine-learning lifecycle. section 7.3.2
Reserve roughly four to ten times the dataset size in memory as a rule of thumb. section 7.3.2
Benchmark representative preprocessing and training workloads, then investigate memory errors or crashes when capacity is insufficient. section 7.3.2
Use techniques such as subgraph sampling when larger graphs exceed available device memory and an upgrade is impractical. section 7.4.2

Sizing memory for a graph workload

A sparse graph dataset containing 10 GB of data suggests reserving approximately 40 GB to 100 GB of memory. If the available GPU cannot handle larger graphs, subgraph sampling can make the workload manageable without immediately obtaining a larger-memory device. section 7.4.2

Table 7.2 Pros and cons of processor choice

Hardware	Recommended Workload	Pros	Cons
CPU	Preprocessing	Suitable for data collection and preprocessing More affordable than GPUs and TPUs	Slower for training due to lack of accelerated parallel processing
GPU	Training	Excellent for training due to parallel processing	More expensive than CPUs Surpassed by TPUs for deep learning tasks
TPU	Preprocessing and training	Faster computation time and cost-effectiveness for deep learning tasks	Requires specific software infrastructure Limited to Google platforms
NPU	Training	Optimized for deep learning and especially good for on-device AI applications, reducing reliance on cloud services	Limited to specific AI workloads, primarily neural network-based tasks
IPU	Training	Especially good for graph-based tasks such as GNNs	Can be more complex to program and optimize compared to NPUs

Compare processor choices alongside memory planning: CPUs suit preprocessing, GPUs suit training, and graph workloads may benefit from specialized processors with different programming and infrastructure trade-offs. section 7.4.2

How Graph Neural Networks in Action covers it

Also explained in

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Graph Neural Networks in Action. The section links point to the parts of the book it draws on.