When IBM first announced OpenPower several years ago, it seemed like a sad end to one of the only x86-alternative architectures still left standing. More recently, we’ve seen renewed interest from various companies, several vendor announcements, increased uptake for GPU and AI applications thanks to the Nvidia-designed NVLink and its superior bandwidth compared with PCI Express, and, in at least some cases, higher overall performance.
EETimes has a profile of Power9’s overall market performance as part of the Facebook Open Compute Project. Google, Alibaba, and Tencent are all working on Power9 systems, with Tencent claiming that Power9 offers 30 percent more performance than an equivalent x86 installation while using less rack space and fewer servers. There are also reports of Power9 being quietly prepped for deployment by at least one major web company and for a major data center customer. IBM reportedly wants to win at least 20 percent of the Linux server market for >$5,000 servers and the company has put a major push behind Power9 as a GPU compute and artificial intelligence research platform, with the launch of its AC922 “Newell” platform late in 2017.
This new platform offers a two-socket Power9 server (up to 22 cores per socket in the water-cooled variant and 20 in the air-cooled system) with up to six Nvidia Volta GPUs. Meanwhile, NVLink 2.0 doubles bidirectional bandwidth to 75GB/s in both directions between the various GPUs as well as the CPU GPU linkages.
IBM expects to continue refining Power9 with both scale-up and scale-out versions of the platform through 2020, with significantly more bandwidth arriving in later versions of the system and, one assumes, significantly higher power consumption as well. EETimes notes that while earlier versions of the Power9 roadmap included 10nm and 7nm versions, they’re nowhere to be found on later versions of the roadmap.
One major shift over the last few years has been the Power customers themselves. Initially, IBM thought it might create an ecosystem around the Power architecture, with chip designers taking out licenses to build compatible products. So far, the chip is finding much more success with OEMs than design firms, but it’s curious that the plans for a 10nm and 7nm version of Power have vanished from existing roadmaps. One possibility is that IBM believes it makes more sense to focus on refining its CPUs and platforms on a single, stable, mature 14nm node for now, rather than trying to shift to bleeding-edge nodes. Power9 CPUs are extremely complex and aren’t great designs for bringing up new nodes.
IBM isn’t going to break the hammerlock x86 has on the general server market, but any gains it can carve out in fields like artificial intelligence or machine learning could lead to further gains as those markets mature. As x86’s own long-term dominance illustrates, being the default solution is a powerful advantage. If IBM can keep Power competitive in this space, it has a fighting chance to carve out a meaningful market, even if x86 owns the more general server industry.
IBM’s AI-Enabled Smartphone Scanner Can Detect Counterfeit Goods
An experienced wine connoisseur can probably tell at a glance if that $1,000 bottle is the genuine article, but most of us cannot. A properly trained machine learning algorithm might be able to help with that.
IBM’s Project Debater to Set Stage for New Kind of Civil Society
Could this be how the Anthropocene age ends, not with a bang, but with a corporate AI composing fragmented arguments in favor of telemedicine?