Home
Blog
Gcore Launches AI Cloud Based on Graphcore IPU in Manassas, US

Gcore Launches AI Cloud Based on Graphcore IPU in Manassas, US

October 27, 2023

2 min read

Gcore Launches AI Cloud Based on Graphcore IPU in Manassas, US

We are pleased to announce the expansion of the Gcore AI Cloud powered by Graphcore Bow IPU to our data center in Manassas, Virginia. Strategically located near Washington DC, Manassas is at the heart of the US’s Eastern Seaboard region, a hub of tech innovation. The AI Cloud provides virtual vPODs with multiple configurations, suitable for running all types of AI/ML workloads—including those that are particularly demanding.

What Is the Basis of the Infrastructure?

The AI cluster is built on the Graphcore Bow Pod₆₄, an IPU system that supports multiple users and various AI workloads across virtual vPODs. The Bow Pod₆₄ contains 94,208 IPU cores and delivers 22.4 petaFLOPS of total compute power. These capabilities allow us to deliver powerful AI computing for training and inference, enabling our customers to develop and deploy their AI/ML models in the cloud and at the edge.

Bow Pod is compatible with all popular ML frameworks, including TensorFlow, PyTorch, and Paddle Paddle, and with cloud-native applications such as Kubernetes, Prometheus, and Grafana.

Manassas Virtual vPOD Configurations

Here are the virtual vPOD configurations we offer at the new Manassas location:

Product	Instance configuration	Number of IPUs	Price per hour
BOW-vPOD₄	60 vCPU / 116GB RAM / 1100GB NVMe (ephemeral) / 100Gbit/s Interconnect	1	$3.68
BOW-vPOD₁₆	120 vCPU / 232GB RAM / 2200GB NVMe (ephemeral) / 100Gbit/s Interconnect	16	$13.70
BOW-vPOD₁₆	240 vCPU / 464GB RAM / 4400GB NVMe (ephemeral) / 100Gbit/s Interconnect	16	$15.80
BOW-vPOD₆₄	240 vCPU / 464GB RAM / 4400GB NVMe (ephemeral) / 100Gbit/s Interconnect	64	$64.30

All configurations are available now via our Cloud platform.

You can also try out vPOD₄ for free for 24 hours. To get the trial, contact our sales team.

What Is a Virtual vPOD?

A virtual vPOD is a flavor in which a Poplar® server—a server that manages Bow IPU servers in a cluster—is deployed on a virtual machine. This is in contrast to a dedicated vPOD, which deploys a Poplar® server on a dedicated bare metal server.

A virtual vPOD provides direct access to a host server and fast connectivity to physical IPU accelerators, offering a cost-effective solution to those who need fast and dynamic access to IPU computing power. You can provision virtual vPODs in just five minutes, compared to fifteen minutes for dedicated physical vPODs.

Benefits of Virtual vPODs

Virtual vPODs let you:

Deploy your development environment seamlessly.
Install and run code at lightning speed.
Enhance your experience with popular ML frameworks like TensorFlow, PyTorch, and Keras.
Leverage ephemeral storage for stateless workloads.
Use external volumes to connect to external block storage and add new data volumes.
Use suspension mode to pause your cluster, useful for unpredictable workloads, resource management, and cost optimization; you can resume the cluster in minutes.

Learn More

Visit our AI IPU product page to explore all the features and benefits of our AI Cloud.

Watch our webinar to learn how IPUs enable you to train AI models for various applications.

Check out our blog to find out how the Gcore AI IPU Cloud can help you with your AI/ML development, training, and inference needs.

Gcore Team

Content Team

Mission Space chooses European sovereignty: why the Luxembourg space startup moved to Gcore

An interview with Alexey Shirobokov, CEO & Founder of Mission Space with Dima Maslennikov, Head of Startups at Gcore, recorded at House of Startups, Luxembourg. At Gcore, we work closely with startups building at the edge of deep t

Introducing GPU VMs on NVIDIA AI infrastructure in Sines (EU): flexible, cost-efficient compute for AI workloads

Some AI jobs require the full power and predictability of dedicated bare metal clusters. Others need something more agile: compute that can be sized up or down quickly, used for a burst of experimentation, powered down when idle, and spun b

Gcore Everywhere AI evolves to full-lifecycle management with Slurm, Jupyter, and token-based inference integrations

AI adoption has a fragmentation problem. Organizations routinely stitch together separate tools for development, training, and serving, each with its own infrastructure, access controls, and operational overhead. The result is a patchwork t

Introducing faster, lower-cost LLM inference with NVIDIA Dynamo

Imagine if you could click a button and suddenly your GPUs increase their throughput by 6x. Or reduce latency by 2x. Or route inference requests seamlessly across different GPU types.That's the experience we're bringing to our inference cus

New AI inference models on Application Catalog: translation, agents, and flagship reasoning

We’ve expanded our AI inference Application Catalog with three new state-of-the-art models, covering massively multilingual translation, efficient agentic workflows, and high-end reasoning. All models are live today via Everywhere Inference

New AI inference models available now on Gcore

We’ve expanded our Application Catalog with a new set of high-performance models across embeddings, text-to-speech, multimodal LLMs, and safety. All models are live today via Everywhere Inference and Everywhere AI, and are ready to deploy i