Last week, I wrote about IBM’s OpenPOWER initiative and how it was attempting to recreate the Linux revolution, on hardware. I pointed out that Intel isn’t the problem that folks thought Microsoft was in the 1990s and that while a lot of software resources in companies could be converted to the fight with Linux, microprocessor skills are far more limited, creating a larger power imbalance between IBM’s effort and Intel than existed between the Linux effort and Microsoft. However, this was all done at the NVIDIA GPU Technology Conference, and a firm with the relative power of NVIDIA partnered with IBM could create something very different and potentially powerful enough to close the gap.
Let’s talk about NVIDIA and OpenPOWER.
It is hard to walk out of the GPU Technology Conference and not get that a huge sea change is going on in the server market. Unlike past events, which were mostly about gaming or design using graphics processors in their traditional form, this conference appeared mostly focused on Big Data analytics and deep learning.
I almost had to walk outside and make sure I was in the same conference. In fact, the biggest announcement at the event wasn’t actually a new card from NVIDIA, it was a joint announcement, which I also covered last week, between Dell, NVIDIA and VMware on desktop virtualization for engineers, animators and other heavy workstation users. This was a very unique server that could substantially change how large-scale engineering, graphics and animation are done worldwide by focusing more on “where” the computing was done rather than the “how.”
Deep learning is the new buzz word (well, words) when it comes to analytics. Ever since Watson, it has been clear that the next generation of analytics platforms needs to be intelligent and far more capable than the last. Watson is currently hampered by the time it takes to train the system, but deep learning techniques massively increase the accuracy while potentially dramatically reducing the time it takes to train the system. Part of this is that, more and more often, systems enabled for deep learning can take an active role in automating the training, which is why the time it takes can be reduced substantially.
GPU computing, which has already been instrumental in the indexing and analysis of unstructured data like photographs, is apparently ideal for deep learning because of its comparatively massive ability to parallel process information. It shouldn’t be a surprise that a system that was developed to create images might also be the best system to analyze them.
NVIDIA is clearly positioning itself as the company at the heart of the next wave of analytics platforms and it doesn’t like Intel very much.
Intel’s Achilles’ Heel
Intel’s historic weakness has always been graphics. Its last big graphics push was an effort called Larrabee, and it was ironically headed at the end by Pat Gelsinger (who runs VMware now and was a major contributor to the Dell announcement above). It failed spectacularly and the company has yet to come up with a true alternative to either AMD’s or NVIDIA’s graphics capability.
If the analytics market moves even more aggressively to deep learning platforms, the competitive advantage that any platform better tuned to NVIDIA’s technology than Intel’s (did I mention NVIDIA doesn’t like Intel?) could be massive. In this segment, performance is king. If that performance is massively better on a GPU tied to, say, a POWER processor, suddenly OpenPOWER has a beachhead that not only is defensible but one that could potentially create another competitive advantage for Watson and for every aligned POWER vendor tied to NVIDIA’s technology for analytics and deep learning.
This new wave of computing could break with POWER, and not Intel.
Wrapping Up: Alliances for the Win
Markets, and wars, are often won and lost through alliances. There is no way that OpenPOWER alone can significantly hurt Intel in markets that Intel now dominates with x86 technology without a market pivot. Such a pivot is happening with deep learning targeting the yet nascent AI and analytics opportunity (particularly analytics tied to unstructured data). An OpenPOWER/NVIDIA alliance targeting this young segment with massive potential could change the market and on this vector Intel is vulnerable.
Having OpenPOWER at the NVIDIA GDC may be a bigger deal than otherwise thought.
Rob Enderle is President and Principal Analyst of the Enderle Group, a forward-looking emerging technology advisory firm. With over 30 years’ experience in emerging technologies, he has provided regional and global companies with guidance in how to better target customer needs; create new business opportunities; anticipate technology changes; select vendors and products; and present their products in the best possible light. Rob covers the technology industry broadly. Before founding the Enderle Group, Rob was the Senior Research Fellow for Forrester Research and the Giga Information Group, and held senior positions at IBM and ROLM. Follow Rob on Twitter @enderle, on Facebook and on Google+