The first fruit of Hewlett Packard Enterprise’s buy of SGI is set to hit the streets in July with the release of a high performance system - the HPE SGI 8600.
The system is a liquid cooled petascale box assembled on legacy SGI ICE AXA architecture, and is aimed squarely at punters involved with beefy scientific and engineering projects.
The 8600 scales to more than 10,000 nodes without additional switches, and uses integrated switches and hypercube tech that supports arrays of liquid-cooled Nvidia Tesla GPU accelerators, hooked up by NVLink interconnects.
This is one of a flurry of HPC machine teasers the firm went public with this week at ISC in Frankfurt; the others are the HPE Apollo 6000 Gen10 System and the HPE Apollo 10 Series.
The 6000 Gen 10 is an air-cooled, HPC platform giving up to 300 teraflops per rack, and uses ‘silicon of trust’ tech to boost application licensing efficiency, reduce latency, up IOPs muscle and lower power consumption and cooling. At least that is what it said on the tin.
Chemical biz BASF has piloted the system to digitise its chemical research, cutting down on computer simulation and modelling times from months to days, HPE told us.
The Apollo 10 Series is for more price-conscious HPC customers - it is all relative - built for entry level Deep Learning and AI apps that are supposed to be easier to manage and deploy.
The sx40 System is a 1U dual socket Intel Xeo Gen10 server with support for up to 4 Nvidia Tesla SXM2 GPUs with NVLink. The pc40 System is a 1U dual socket Intel Xeon Gen10 server that supports up to 4 PCIe GPU cards.
The systems come with an updated Performance Software Suite that aids management, optimisation and monitoring of the HPC gear.
More details on the spec and prices will be fleshed out in public next month. ®