The rumors are confirmed at long last and Chinese PC and server maker Lenovo Group is buying IBM’s x86 server business for $2.3 billion. What does this mean for the HPC community?
IBM sells x86 clusters into the broad HPC community and also peddles BlueGene/Q massively parallel systems based on its PowerPC processors as well as clusters based on its enterprise-grade Power Systems machines. Big Blue accounts for about a third of the machines on the TOP500 list ranking of supercomputers, and many of them are based on its System x rack servers, iDataPlex hyperscale machines, or BladeCenter blade servers. Many customers in the HPC space were no doubt looking at the NextScale “vanity-free” designs that came out last September from IBM, which has higher density and lower costs than the iDataPlex machines.
Now, all of these product lines are being sold off to Lenovo, which will continue to sell and support them. In fact, the arrangement between the two parties calls for IBM to support the products with its existing teams “for an extended period of time,” according to Adalio Sanchez, general manager of the System x Division, who spoke on a conference call with journalists to go over the deal. (We have gone through the terms of the complex deal over at EnterpriseTech, if you want all of the nitty gritty details.)
With so many government agencies using IBM’s x86 servers in their various incarnations, there are possible national security implications to the sale of the System x Division to Lenovo. Moreover, some HPC centers that have a long tradition of buying systems from IBM may give pause. The same holds true for IBM’s largest enterprise customers, where IBM tends to sell lots of x86 iron to wrap around its Power Systems and System z machines.
Rumors about a deal between IBM and Lenovo have been circulating since last April, so it is no surprise to any academic supercomputing center or national lab that this was possible. But today, it has happened, and now people are forced to think through the ramifications.
“First of all, we are committed to a seamless transition with our clients,” Sanchez said on the call when asked by HPCwire what affect the deal would have on HPC and enterprise customers. “I have had the opportunity to speak to a number of clients this morning in the spaces you describe, and once they understand what we are trying to do, I think they are very, very excited about the opportunities here. To your point, there may be some concerns about sensitivities here. Having said that, you know that the bulk of today’s x86 servers in the marketplace all originate from China. And so I think the pragmatics of the situation will overcome any of these concerns.”
Under the terms of the deal, Lenovo is licensing and reselling IBM’s Global Parallel File System as well as the Platform Computing grid and cluster management software, which are key ingredients to HPC clusters sold by IBM whether they are based on x86 or Power processors. The deal also gives IBM the right to resell the x86 server line from Lenovo, and thus Big Blue will be able to build clusters just like it always has.
“At the end of the day, I see a lot of great alignment here, quite frankly,” said Peter Hortensius, who is president of the Think Business Group at Lenovo and a senior vice president in the overall organization. This is the part of Lenovo that already sells x86 servers and storage arrays. “Today, when IBM sells those solutions, it’s the IBM hardware coupled with the IBM software. In the future, regardless of who ends up selling it, it is the IBM software and this business’ x86 hardware. So there is no reason why we shouldn’t be able to make these very complementary and continue exactly, in terms of value to the customer, as we are today.”
It remains to be seen how HPC shops large and small who use IBM systems will feel about the company selling off the x86 server business. What can be said is that if this transition is not seamless and painless, then Cray, SGI, Hewlett-Packard, Dell, and others will be quick to exploit the situation to their own advantage.