Mellanox is now Nvidia
Nvidia Networking Thanks to Mellanox
Nvidia is getting bigger. This is all thanks to its acquisition originally announced in March 2019 with Mellanox Technologies Ltd. This merger is designed to provide Nvidia with better networking capabilities. All being achieved through Mellanox's communication hardware.
This will be administered with the integration of Nvidia’s artificial intelligence (AI) solutions. This will process and monitor data in real-time on Mellanox’s communication hardware. Through acquiring Mellanox, Nvidia networking will establish better performance from networked devices.
Not only that but better resource allocation, and lower running costs. Nvidia will now control a company that makes network and internet solutions as we know them.
Who are Mellanox?
Mellanox Technologies Ltd. was originally founded in 1999 by a group of Israeli executives that left Intel and Galileo Technology. The business was originally designed to offer network communications equipment. It then became a multinational best known for its InfiniBand and Ethernet technology.
It serviced all aspects of computer connectivity. Specializing in everything from high-performance computing to network hardware such as switches, data centers, and cloud computing solutions. In its later years, it also branched out in data storage optimization due to its need for better data communication.
Mellanox; A Brief History
In its relatively brief history, Mellanox produced many disruptive offerings including ethernet as we know it and storage area network (SAN) devices. It did this through regular acquisitions of smaller companies.
This enabled them to combine separate intellectual properties. It's initial public offering (IPO) on the NASDAQ occurred in February 2007 and generated the company $102 million. This in turn pushed its overall value to $500 million.
In 2010 Oracle Corporation became a major shareholder in the company by purchasing 10% of its stock. This was to help secure Oracle's use of the InfiniBand technology in its Exalogic and Exadata offerings.
In 2011 Mellanox acquired Voltaire Ltd. for £218 million, a key provider of network switches that were later utilized in Mellanox offerings. 2013 saw Mellanox acquire XLoom Communications Ltd assets. This included chip-scale packaging along with some of its key personnel.
Kotura Inc. and IPtronics were also acquired in the same year. These companies were developers of phonic technologies and photonic interconnect technologies respectively. Mellanox acquired Integrity Project in 2014 that produced software connectivity and EZchip Semiconductor in 2016.
Merging with Nvidia
In March 2019 Nvidia stated that it was interested in buying Mellanox for $6.9 billion. Microsoft, Intel, and Xilinx were also interested in acquiring the company.
In November 2017 Starboard Value, a hedge fund founded in 2002 purchased around 10% of Mellanox.
Soon after in January 2018, the hedge fund criticized the research and development expenditure of Mellanox. The following day the company released news that it would cease its research element of the company. From this, 100 jobs were lost in the US. At the time 2900, mostly Israeli personnel, were employed by Mellanox.
Starboard Value notified shareholders that they should elect new board members while nominating 9 including Jeffery Smith CEO of Starboard Value.
Shareholders in May 2018 approved an election and by June three board members stepped down. Two were replaced by Starboard Value nominees and one was agreed upon by both parties. Soon after these changes in 2019 Mellanox was acquired by Nvidia.
Nvidia’s Vision After the Merge
While $7 billion is a large expense to absorb by Nvidia, it is not expected to get buyers' remorse any time soon. Nvidia networking hopes to combine its computing expertise and Mellanox's high-performance networking technology to create higher performant solutions through AI integration. The question is what are the key assets that will make or break the profitability of the acquisition?
Fundamentally AI enables large repetitive tasks to be completed through an automated process. This makes it exceptionally useful for data storage and network balancing. At present, there are already storage solutions based on the latest non-volatile memory NVM protocol.
This utilizes this technology to preempt the loading of data. These systems have allowed supercomputers to work efficiently with big data and are currently being adopted in the research sector.
NVM Protocol Enabling AI Management Under the Hood
The NVM protocol allows NVMe interfaces to achieve bidirectionality of data in any given instance over a wide bandwidth. Data does not need to be formatted when entering or exiting a storage solution.
This is in contrast to SATA or ATA and as such the whole process of data, communications is much faster. Companies such as Microsoft, Facebook, and Google have all stated that while SATA and ATA were stepping stones to the NVMe protocol there is no intent to develop another.
This is due to it meeting decades-old requirements that were never fully satisfied by SATA or ATA from SCSI and other predecessors.
What is Artificial Intelligence?
Artificial intelligence is based on a statistically weighted matrix. In this matrix data such as an image is added to one side of the matrix, data filters are applied and algorithms run at each node until the data reaches the other side of the matrix.
From decisions made at each stage, it is possible to determine for instance if the image was either a sheep or a dog. The caveat is that the matrix is only as good as the filters used for data assessment. Also the optimization of the weight used for each node.
This optimization process is called deep learning but in reality, is just a dataset usually in the form of a database parsed into the matrix in an iterative process. The process during the learning stage before roll out or during the use of the matrix occurs until a stated error threshold is reached.
The Rise of Artificial Intelligence
Originally when AI was being developed a few years ago it was graphics cards that were found to have the best processors for AI R&D when compared to high-level CPUs. As such AI was cultivated on OpenGL and other readable languages.
This transitioned to container-based programming. Here code could be quickly created in environments that could be quickly added to hardware with supporting AI SDKs. During this period the creation of AI-centric processors, both surface mount and discrete was achieved.
This technology now enables AI-assisted driving to occur. This includes many supporting AI processors in parallel and series workflows to handle more complex operations. Sensor data is pushed through a CPU to the AI processors and results are passed back to the CPU for task delegation.
Putting the Pieces Together
What this all means is that Nvidia has the ability to produce AI-centric hardware that can support network and storage optimization. This is now along with the network hardware to enable solutions to be produced all through Nvidia.
Imagine every switch, data center, cloud solution, and storage solution all preempting a user's needs. Just as or before a user logs into their computer, the data center could find big data that is of potential interest.
This data has been stored on parallel NVMe devices and then in a parallel feed to the user's terminal in sections when required; preempting this data lineup and queuing it ready. The network would become more efficient from not needing to push unneeded data and more diverse data. Alternately the same data could be accessed and worked on by a team all at once.
Conclusion
The purchase of Mellanox by Nvidia will allow them to build network hardware that supports Nvidia AI processors, thus creating more performant networks through better AI integration.
This will improve how users handle big data and data storage management. Load balancing, backup, and recovery will enable fewer overheads to be required for larger data usage.