AI Intelligent Processing Units (IPU)
Infrastructure
Exclusive AI cloud infrastructure to accelerate machine learning.
AI Infrastructure as a Service
We bring together Computer.Com's IPUs and the Computer.Com's Cloud services for building AI IPU infrastructure under unified UI and API for ML acceleration.
Get started quickly, save on computing costs, and seamlessly scale to massive IPU compute on demand and with ease.
Computer.Com's IPU cloud services are now available, with free trials and a range of pricing options enabling innovators everywhere to make new breakthroughs in machine intelligence.
Why have we chosen Graphcore IPUs?
Exclusive solution pack
Computer.Com’s IPU-based AI cloud is a Graphcore Bow IPU-POD scale-out cluster, offering an effortless way to add state of the art machine intelligence compute on demand, without the need for on-premises hardware deployment and AI infrastructure building from scratch.
The IPU is an entirely new kind of massively parallel processor, co-designed from the ground up with the Poplar® SDK, to accelerate machine intelligence. Cloud IPU’s robust performance and low cost make it ideal for machine learning teams looking to iterate quickly and frequently on their solutions.
Cloud IPU virtual vPODs make it easy to use IPU hardware by providing direct access to the BOW host machines. Users have full access to each IPU virtual instance, allowing them to install and run any code they wish in an ultrafast connection with IPU accelerators. This also enables the use of ephemeral storage, execution of custom code in input pipelines, and better integration of Cloud IPUs into research and production workflows.
- Connect external block storage for system and data volumes. Easily attach new data volumes.
- Reduced deployment time: compared with dedicated vPOD, virtual vPODs spawned up to 5 minutes.
- Suspension mode to temporarily pause the cluster and avoid unnecessary charges while also being able to resume quickly.
Massive Performance Leap
World-leading performance for natural language processing, Computer.Com's vision and graph networks
Unique architecture for differentiated results
Low latency inference
Much More Flexible
Designed for training and inference
Support for wide range of ML models
Make new breakthroughs for competitive advantage
Easy to Use
Support from AI experts
Extensive documentation, tutorials and pre-canned models
Popular ML framework support
Suspension mode for Cloud virtual vPODs
Suspension mode provides a cost-effective and resource-efficient solution for temporarily pausing a virtual private cloud environment when it is not in use. By utilizing this feature, customers can effectively reduce expenses while preserving the integrity of the data and configurations.
- Only storage and Floating IP (if active) are charged when a cluster is suspended
- Cluster can be easily reactivated with the same configuration
- The network configuration and cluster data are stored on external block storage, excluding ephemeral storage information. This offers the ability to modify the configuration and expand the cluster as required, providing greater flexibility
Features and advantages
World-class performance for natural language processing
Build, train and deploy ready-to-use ML models via dashboard, API, or Terraform
Dataset management and integration with S3/NFS storage
Version control: Hardware, Code, Dataset
Secure Trusted Cloud platform
Free egress traffic (for public or hybrid solutions)
SLA 99.9% guaranteed uptime
High-skilled technical support 24/7
Made in the EU
AI full lifecycle tools and integrations
ML and AI solutions:
- TensorFlow
- Keras
- PyTorch
- Paddle Paddle
- ONNX
- Hugging Face
Receiving and processing data:
- Storm
- Spark
- Kafka
- PySpark
- MS SQL
- Oracle
- MongoDB
Development tools:
- Visual Studio Code
- PyCharm
- Jupyter
- GitLab
- GitHub
- RStudio
- Xcode
- Airflow
Exploration and visualization tools:
- Seaborn
- Matplotlib
- TensorBoard
Programming languages:
- JavaScript
- R
- Swift
- Python
Data platforms:
- PostgreSQL
- Hadoop
- Spark
- Vertika
Accelerate ML with ready-made AI Infrastructure
With the AI Infrastructure, customers can now easily train and compare models or custom code training, and all your models are stored in one central model repository. These models can now be deployed to the same endpoints on Computer.Com’s AI Infrastructure.
Computer.Com’s IPU-based AI cloud is designed to help businesses across various fields, including finance, healthcare, manufacturing, and scientific research. It is built to support every stage of their AI adoption journey, from building proof of concept to training and deployment.
AI model development
ML models: Face recognition, Object detection
AI training and hyperparameter tuning
ML Model delivery and deployment pipelines
Locations
IPU-POD systems
Ready to order in Luxembourg in June 2022
IPU-POD systems let you break through barriers to unleash entirely new capabilities in machine intelligence with real business impact. Get ready for production with IPU-Pod64 and take advantage of a new approach to operationalize your AI projects.
IPU-Pod64 delivers ultimate flexibility to maximize all available space and power, no matter how it is provisioned. 16 petaFLOPS of AI-compute for both training and inference to develop and deploy on the same powerful system.
IPU-POD systems
Ready to order
IPU-Pod256 is available in Amsterdam. It allows customers to explore AI compute at a supercomputing scale. Designed to accelerate large and demanding machine learning models, IPU-Pod256 gives you the AI resources of a tech giant.
Pricing
Depending on the location, servers have different traffic options*. You can find explicit information below:
| Product | Server Config | IPUs | Quantity | Price | |
|---|---|---|---|---|---|
|
Bow Pod4 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
4 |
|
|
|
|
Bow Pod16 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
16 |
|
|
|
|
Bow Pod64 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
64 |
|
|
|
|
Bow Pod128 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
128 |
|
|
|
|
BowPod256 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
256 |
|
|
|
|
BowPod1024 |
2×7763/ 512GB RAM / 2×450 SATA + 7×1.8Tb nvme / 2x100G |
1024 |
|
|
| Product | Server Config | IPUs | Quantity | Price | |
|---|---|---|---|---|---|
|
BOW-vPOD4 |
60 vCPU / 116GB RAM / 1100GB NVMe (ephemeral) / 100Gbit/s Interconnect |
4 |
|
|
|
|
BOW-vPOD16 |
120 vCPU / 232GB RAM / 2200GB NVMe (ephemeral) / 100Gbit/s Interconnect |
16 |
|
|
|
|
BOW-vPOD16 |
240 vCPU / 464GB RAM / 4400GB NVMe (ephemeral) / 100Gbit/s Interconnect |
16 |
|
|
|
|
BOW-vPOD64 |
240 vCPU / 464GB RAM / 4400GB NVMe (ephemeral) / 100Gbit/s Interconnect |
64 |
|
|
| Product | Server Config | IPUs | Quantity | Price | |
|---|---|---|---|---|---|
|
IPU-POD4 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
4 |
|
$ 2.66 / 1 hour |
|
|
IPU-POD16 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
16 |
|
$ 11.76 / 1 hour |
|
|
IPU-POD64 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
64 |
|
$ 49.21 / 1 hour |
|
|
IPU-POD128 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
128 |
|
$ 98.43 / 1 hour |
|
|
IPU-POD256 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
256 |
|
$ 196.87 / 1 hour |
|
|
IPU-POD1024 |
2×5320 / 384GB RAM / 2x960GB SSD / 2x100G |
1024 |
|
$ 787.51 / 1 hour |
