microsoft hgx-1

The data centre infrastructure market is seeing significant changes, with big-name suppliers of servers and other equipment being challenged by so-called “white-box” suppliers. 

White box suppliers is a term to indicate that the companies which produced the equipment are not as well known as the big names, such as Cisco, Microsoft, Dell, IBM and market leader in servers HPE.

But inevitably, some of these supposedly anonymous white box suppliers will become well known eventually, and to a great extent some of them are already famous.

Foxconn, the main contract manufacturer for Apple, making its iPhones, iPads and other devices, is also in the server-making business.

In fact, it makes servers for Dell, HP and others, as well as for itself, branding them Foxconn.

It may not be a household name exactly, but Foxconn is quite well known, although its original name – Hon Hai Precision is probably not quite as widely used.

Whatever its name, the company has recently announced that it is doubling its manufacturing capacity in China, according to Nikkei.

Growing demand in China and increased orders from overseas are said to be the reason for the additional capacity.

Increasingly large compute workloads created by big data and artificial intelligence applications are said to be the main reasons for needing more servers.

Meanwhile, another of the so-called white box companies – Super Micro – has launched a new petabyte scale server which it calls a “1U SuperServer”.

The system uses new all-flash non-volatile memory express, or NVMe, and just a bunch of flash, or JBOF, and supports 32 hot-swap NVMe SSDs.

Super Micro says the 1U all-NVMe Storage Servers and JBOF disaggregate storage into shared pools that are rapidly becoming the preferred hardware infrastructure for demanding Big Data analytics applications such as autonomous driving and real-time financial fraud detection.

Up to 12 hosts can be directly connected to the 1U pooled NVMe storage.  Alternatively, for customers who want to deploy an NVMe over Fabric (NVMeoF) solution, hundreds of hosts can be connected to the pooled high-performance NVMe storage over Ethernet, Infiniband or Omnipath.

Super Micro says its servers and solutions “help maximize high-performance storage resource utilization” and reduce the datacenter footprint resulting in lower total cost of ownership.

Charles Liang, president and CEO of Super Micro, says: “With Supermicro 3.0, our enterprise customers benefit from the industry’s broadest selection of first-to-market server and storage systems with global reach, premium quality, RAS security, rack scale management and global services, and our new all-flash 32 hot-swap drives in a high-density 1U system design is the latest example of how Supermicro continues to lead the way for NVMe technology.

“With more than triple the all-flash storage density of previous 1U solutions, this Supermicro system will take us to Petabyte scale in a single 1U system in the near future.

“This new JBOF supports flexible configurations with up to twelve hosts or head nodes and extremely high data transfer throughput up to 64GB per second.”

Super Micro has also recently released a 6U SuperBlade disaggregated server.

And while neither Microsoft nor Facebook can be considered “anonymous” or white box suppliers, the two builders of hyperscale data centres has unveiled a new type of high-powered server for data centres which uses ope standards under the Open Compute Project.

As explained in Top500.org, the supercomputing website, the accelerator uses eight Nvidia Tesla P100 graphics processing units in one chassis.

Microsoft has produced a version it calls HGX-1, and Facebook has produced another it calls Big Basin.

Like the Super Micro systems, the idea of the Microsoft-Facebook server is to disaggregate the GPUs and central processing units in cloud environments, allowing users to decide how to mix and match according to their requirements.

This is expected to be the requirements for those building and using applications which require machine learning.

Nvidia has been investing a lot of resources into developing GPUs with artificial intelligence capabilities, and machine learning – a branch of AI – is said to be one of the P100’s major areas of expertise.

Top 500 says an eight-GPU configuration of these devices shown by Microsoft and Facebook “will yield over 80 teraflops at single precision and over 160 teraflops at half precision”.

LEAVE A REPLY

Please enter your comment!
Please enter your name here