Cloud Services for GPU Computing

Login


 
 
 
 
Additional Options

Hoopoe provides users with a Compute-on-Demand, cloud solution, for extreme system utilization.

Introduction to Hoopoe™

Hoopoe is a cloud solution and infrastructure for organizations, allows using of GPU hardware for computational intensive tasks, while running on existing systems.

Hoopoe provides users with a cloud solution for GPU computing.
GPU hardware provides a tremendous processing power for computation intensive applications. It is possible to offload arithmetic computations to the GPU and free the CPU for other tasks.
Today GPUs can provide over 1 TFLOPS of computation power, that can be compared to almost 50 CPU cores, along with fast memory access.
Each device has a dedicated internal memory of 4 GB, to store data. This can simply replace an existing 32 bit system.
To read more about GPU architecture, technology overview and application case studies: GPU Architecture Benefits.

Hoopoe Functionality

Not every organization can manage a GPU cluster. By means of specific hardware configurations or required software.
There is a current need for many types of applications to offload computations to a GPU, but most of the server hardware in organizations doesn't support an additional GPU device in the server box.
When it comes to question of having hetrogeneos systems, some with a GPU device and some without, the decision is mostly to prefer and keep the existing systems.

In a world of IT economy, a cloud service is the best solution for every organization for reducing total expenses and cluster maintenance.

Hoopoe provides access to GPU enabled systems with different configurations to match the needs of most demanding applications, utilizing 1 - 4 GPU devices accessible to a single system.

Using Hoopoe, users can run applications that use the GPU hardware, but also on a native and familiar environment with Microsoft Windows or Red-Hat Linux.

In addition to machine access services, Hoopoe provides a unique infrastructure that allows using of multiple GPU devices through a Web Service interface. This way reducing the need and complexity of management and communication between different systems.
Hoopoe takes care for the quality of service and notifies upon completion of the task.
This solution provides a real-time distributing system, built on top a dedicated software to utilize the GPU hardware.

Hoopoe Compute-on-Demand

Hoopoe provides users with a unique feature for Compute-on-Demand based service.
The system was built on top of a real-time, dedicated distributing software, with Hoopoe it is possible to distribute tasks to GPU hardware without overhead.
It is the job of the user to define the layout of the task and distribtion semantics, and supply the necessary data to process (through files).
From this point, Hoopoe takes care of the work, and sends the tasks to available GPUs, taking care of the process, quality of results and many other aspects.
Being a real-time solution, Hoopoe can manage thousands of GPUs in on second.

To read more: Hoopoe Architecture.

Features & Highlights

Hoopoe provides access to the latest NVIDIA hardware, including Tesla cards for extreme performance.

Supporting the following software architectures: Hardware support: Operating systems:
  • Microsoft Windows HPC Server 2008
  • RedHat Linux 5, 32/64 bit

Instance Types

Description

Four classes of plans are provided for organizations to choose from. Meeting low-end applications demands to extreme high-end.

  • Low-end - exposes to the user a system equipped with a single GPU device to serve as a co-processor for general computations.
  • Mid-range - exposes to the user a system equipped with a dual GPU configuration for demanding computations.
  • High-end - exposes to the user a system equipped with a quad GPU configuration, offered in high density for intensive computations or part of multi-core oriented application.

All configurations offer a variety of operating systems to work with.
Systems used within Hoopoe have passed certification to work with GPU hardware and deliver best performance.

Plans

Low-end Mid-range High-end
Operating system - Linux 32/64 bit
- Microsoft HPC Server 2008
- Linux 32/64 bit
- Microsoft HPC Server 2008
- Linux 32/64 bit
- Microsoft HPC Server 2008
GPU Tesla C1060 Tesla S1070 Tesla S1070
GPU# 1 2 4
Total GPU Cores 240 480 960
CPU Intel Xeon X5482 Intel Xeon X5482 Intel Xeon X5482
CPU# 2 2 2
Total CPU Cores 8 8 8
RAM 8 GB 16 GB 32 GB
All rights reserved © 2008-2011. Company for Advanced Supercomputing Solutions Ltd.