Nvidia Tesla V100 GPU is almost equal to 100 CPUs. This gives customers more power to handle computationally demanding applications, like machine learning, analytics, and video processing. One can select as many as eight NVIDIA Tesla V100 GPUs, 96 vCPU and 624GB of system memory in a single VM, receiving up to 1 petaflop of mixed precision hardware acceleration performance.
NVIDIA V100s are available immediately in the following regions: us-west1, us-central1 and europe-west4. Each V100 GPU is priced as low as $2.48 per hour for on-demand VMs and $1.24 per hour for Preemptible VMs. Making Nvidia Tesla V100 available on the compute engine is part of Google’s GPU expansion strategy. Similar to Google GPUs, the V100 is also billed by the second and Sustained Use Discounts apply.
NVIDIA Tesla P100 GPU, on the other hand is a good fit if one wants a balance between price and performance. One can select up to four P100 GPUs, 96 vCPUs and 624GB of memory per virtual machine. The P100 is also now available in europe-west4 (Netherlands) in addition to us-west1, us-central1, us-east1, europe-west1 and asia-east1.
* Maximum vCPU count and system memory limit on the instance might be smaller depending on the zone or the number of GPUs selected.
** GPU prices listed as hourly rate, per GPU attached to a VM that are billed by the second. Pricing for attaching GPUs to preemptible VMs is different from pricing for attaching GPUs to non-preemptible VMs. Prices listed are for US regions. Prices for other regions may be different. Additional Sustained Use Discounts of up to 30% apply to GPU on-demand usage only.
Google Cloud makes managing GPU workloads easy for both VMs and containers by providing,
- Google Compute Engine where customers can use instance templates and managed instance groups to easily create and scale GPU infrastructure.
- NVIDIA V100s and other GPU offerings in Kubernetes Engine, where Cluster Autoscaler helps provide flexibility by automatically creating nodes with GPUs, and scaling them down to zero when they are no longer in use.
- Preemptible GPUs for both Compute Engine managed instance groups and Kubernetes Engine’s Autoscaler optimizes the costs while simplifying infrastructure operations.
- Google announce the largest overhaul of their Cloud Speech-to-Text
- Google’s kaniko – An open-source build tool for Docker Images in Kubernetes, without a root access
- How machine learning as a service is transforming cloud