Despite dumping its money-losing semiconductor business on GLOBALFOUNDRIES this week and a discouraging 3Q14 financial report, IBM appears determined to drive Moore’s Law. The law, which produced decades of price/performance gains for IT, will continue to achieve gains but it won’t be solely silicon-based.
How badly do we need Moore’s Law to continue? Dr. Bernard Meyerson, IBM Fellow, VP of Innovation, as part of a keynote presentation at Enterprise2014 put it this way: when it comes to data, there are 6-9 orders of magnitude on the horizon. Today’s zEC12 processor, the fastest commercial processor out there, hasn’t a chance of keeping up for long.
Still, Meyerson isn’t writing off silicon: “Silicon transistors will dominate Information Technology for decades to come but contribute little to its progress,” he declared. To augment the shortcomings of silicon, we will have to look to innovation and the resulting integrated solutions made up of specialized hardware, software, systems, architectures, and network functionality to compensate for lost technology benefits.
Ironically, data center managers actually may find themselves worrying about the slowness of the speed of light and the longer data paths that may result. His question to data center managers: Even at 300,000,000 meters/sec, is light fast enough to keep pace with technology? His answer: Not even close.
The answer lies in new innovative integrated solutions that include 3D integration, synaptic architectures, agile computing (autonomic acceleration), cognitive computing, neuromorphic systems and more. In the near term he suggests data centers to look into advances in FPGA (Field Programmable Gate Array) and GPU acceleration. In the new Power8 systems, FPGAs leverage CAPI to avoid a lot of overhead and delay.
IBM researchers explore new semiconductor materials
The offloading of IBM’s global commercial semiconductor technology business to GLOBALFOUNDRIES, at a cost of $1.5 billion over three years, doesn’t signal an IBM retreat from the semiconductor business. The deal commits GLOBALFOUNDRIES as IBM’s exclusive server processor semiconductor technology provider for 22 nanometer (nm), 14nm, and 10nm semiconductors for the next 10 years. You can bet the upcoming generation of System z will use 22nm chips and 14nm and 10nm chips for subsequent revs of the z, or whatever they are calling it by then.
This is combined with IBM’s previously announced $3 billion investment over five years for semiconductor technology research to lead in the next generation of computing. Between that investment and the offloading of the semiconductor fabrication business this week combined with the research described at Enterprise2014, you can be sure that IBM will stay involved in the CPU business. But one thing you should realize now, just throwing more silicon CPUs at the performance challenge, as IT has done for decades, will no longer work. Adds Meyerson: Brute force (more of the same) has run its course.
Full disclosure: DancingDinosaur is NOT a financial analyst. Still, the IBM 3Q14 financials released on Monday isn’t going to thrill IBM investors. Revenues were down across the board. Of most interest to DancingDinosaur were the results of IBM’s hardware group, STG, which dropped. Specifically, revenues from Power Systems were down 12% compared with the 2013 period. Revenues from System x were down 10%, but that’s now Lenovo’s worry. Revenues from System z fell 35% compared with a year ago while revenues from System Storage decreased 6%. The System z is due for a refresh in 2015, which will undoubtedly entail a significant price/performance gain plus whatever other goodies IBM will load on. This usually gives the z a revenue boost. Power just introduced some new POWER8 machines at Enterprise 2014, which should result in revenue increases in upcoming quarters.
Maybe not so coincidentally on the blog itjungle, a piece titled 2020 Processor Technology Could Unite Power And Mainframe Chips, by Dan Burger throws yet a few more possibilities at the processor question, especially as it pertains to IBM’s enterprise server platforms, Power and System z. Burger apparently was attending the same session as DancingDinosaur when Bernie Meyerson and others brought up the silicon question. He caught up with Ross Mauri, the general manager of System z, grabbing this interesting quote on the subject: “Now there will have to be investment in the next big thing and so it’s interesting to consider whether the Power chip and the mainframe CMOS chip will merge into one chip for both platforms,” adding “I don’t know what is next in that 2020 time frame based on the technologies we are looking at today—anything is possible.” BTW, 2020 is just a bit more than five years away; DancingDinosaur, who has been covering business and technology since 1975, may not even be retired by then.
Tags: 3D integration, agile computing (autonomic acceleration), CAPI, cognitive computing, FPGA (Field Programmable Gate Array), GLOBALFOUNDRIES, GPU acceleration, Moore's Law, neuromorphic systems, POWER8, semiconductor, semiconductor business, semiconductor technology, silicon transistors, synaptic architectures