Cloud TPU offering
Cloud TPU v2
64 GB High Bandwidth Memory (HBM)
Cloud TPU v3
128 GB HBM
Cloud TPU v2 Pod Alpha
4 TB HBM
2-D toroidal mesh network
- Proven Reference Models
- Use proven Google-qualified reference models, optimized for performance, accuracy, and quality for many real-world use cases of object detection, language modeling, sentiment analysis, translation, image classification, and more.
- At their core, Cloud TPUs and Google Cloud's Data and Analytics services are fully integrated with other GCP offerings, providing unified access across the entire service line. Run machine learning workloads on Cloud TPUs and benefit from Google Cloud Platform’s industry-leading storage, networking, and data analytics technologies.
- Connect Cloud TPUs to Custom Machine Types
- You can connect to Cloud TPUs from custom VM types, which can help you optimally balance processor speeds, memory, and high-performance storage resources for your individual workloads.
- Preemptible Cloud TPU
- Save money by using preemptible Cloud TPUs for fault-tolerant machine learning workloads, such as long training runs with checkpointing or batch prediction on large datasets. Preemptible Cloud TPUs are 70% cheaper than on-demand instances, making everything from your first experiments to large-scale hyperparameter searches more affordable than ever before.
“Cloud TPU Pods have transformed our approach to visual shopping by delivering a 10X speedup over our previous infrastructure. We used to spend months training a single image recognition model, whereas now we can train much more accurate models in a few days on Cloud TPU Pods. We’ve also been able to take advantage of the additional memory the TPU pods have, allowing us to process many more images at a time. This rapid turnaround time enables us to iterate faster and deliver improved experiences for both eBay customers and sellers.”— Larry Colagiovanni VP of New Product Development, eBay