HP Released High Accuracy CPU with Capacity of 30 Trillion Calculations

Published:2011/6/13 4:10:00 Author:Amy From:SeekIC

Since May this year, NVIDIA has released a new generation of Tesla M2090 GPU product claiming the fastest high efficiency parallel processors over the world. It is applied in Hewlett-Packard’s new ProLiant SL390 G7 server. The HP SL390 series servers can be embedded eight M2090 at most. Meantime, SL390 G7 product can provide the mixed computing environment for GPU and CPU combination.

In the conference of HP DISCOVER, HP published GPU Starter kit for users and cooperative partners. There is no need to use M2090 GPU coprocessor in server to realize the 10 trillion calculation capability. The GPU Starter Kit is composed of two ProLiant SL6500 server cases and eight ProLiant SL390s G7 2U computation nodes. A server case can contain three GPU. This product has been put into market in April with low-profile.

In addition, according to the technique information, every SL390x G7 server node has two most powerful X5675 processors with dominant frequency 3.06GHz. Eight nodes co-work with peak value performance of double-precision floating-point computing power reaching 1.18 trillion times. Every node has three M2070 GPU coprocessors without fans to enable the peak value performance of double-precision floating-point computing power add up to 12.36 trillion times. GPU and CPU floating-point computing capability in a stander will be 13.54 trillion times.

HP stated that GPU Starter Kits will be preinstalled in nodes with enterprise version Linux operating system. Cluster Manager Program and Linux value kits extended version provide high performance for users. This kind of product is also suitable for CUDA D&R environment and normal operation time. A DL380 is configured in stander as a control node, and a port finiBand exchange and a port 24 Ethernet switche are also added in stander. In general, after HP users start up nodes and the network and the storage are got through, the application software runs.

In case users need more powerful performance, the server of every HP’s node can be added M2090 GPU. This kind of GPU product has higher clock frequency and larger memory bandwidth with 512 cores integrated. Double-precision floating point computing capability can reach 665 billion every sec with whole floating point calculation ability of 24 GPU coprocessor 16 trillion. The related integrated circuit is MC33091.

Reprinted Url Of This Article: http://www.seekic.com/blog/Appliance/2011/06/13/HP_Released_High_Accuracy_CPU_with_Capacity_of_30_Trillion_Calculations.html

Print this Page | Comments | Reading(356)