The world of artificial intelligence is moving at lightning speed. At Google Cloud, we’re committed to providing best-in-class infrastructure to power your AI and ML workloads. Dataflow is a key component of Google Cloud’s AI stack, allowing you to create batch and streaming pipelines that support a variety of analytics and AI use cases. We’re excited to share a set of modern features that give you more choice, accessibility, and efficiency when it comes to running batch and streaming ML workloads.
More Choices: Performance-optimized hardware
We understand that not all ML workloads are created equal. That’s why we’re expanding our hardware offerings to give you the flexibility to choose the accelerator that best fits your specific needs.
-
New GPU: We’re constantly adding the latest and greatest GPUs to our lineup and recently announced support for: NVIDIA H100 GPU (A3 High and A3 Mega VM with enhanced networking capabilities). This means you can leverage cutting-edge hardware to accelerate your AI inference workloads. Leading enterprises leverage Dataflow’s GPUs to power innovative customer experiences for threat intelligence platform providers. flash point Enhance document translation to media providers spotify Allows for extensive podcast previews.
-
TPU: For large-scale ML tasks, Tensor Processing Units (TPUs) provide a powerful and cost-effective solution. We recently announced support for TPU V5E, V5P, V6EThis enables leading ML builders to efficiently run large-scale, low-latency machine learning inference workloads directly within Dataflow jobs.
Improving accelerator availability
Having access to the hardware you need, when you need it, is key to keeping your ML projects on track. We’ve introduced a new way to consume accelerators, making it easier than ever to get the resources you need.
-
GPU/TPU reservations: Now available for reservation GPUs and TPUs for Dataflow jobsso you can get the resources you need, when you need them. This is important for critical workloads that cannot afford to wait for resources to become available.
-
Flexstart GPU provisioning: For batch jobs with flexible start times, securing GPUs can be a manual and uncertain process due to high industry-wide demand. our new Flexstart provisioning model enabled by Dynamic workload scheduler (DWS) effectively addresses this issue. Instead of failing a job if accelerator resources are unavailable, Dataflow now queues the job and automatically starts the job as soon as the required GPUs are available. This eliminates the need for repeated manual resubmissions, reduces the risk of out-of-stocks, and improves developer productivity.
