About two years ago, Alphabet's Google (GOOG 0.56%) announced that it was working on an open server architecture based on International Business Machines' (IBM -0.35%) then-upcoming POWER9 processors. Prior to 2013, IBM's chips were available only as part of an IBM Power system. IBM set out to change that with the OpenPOWER foundation, a consortium of companies working to build an open ecosystem around the POWER architecture. Both IBM and Google are founding members, and Google's commitment to POWER was a significant step forward for the initiative.
IBM officially launched its first POWER9 chip late last year. On Monday at the OpenPOWER Summit, Google systems hardware engineer Maire Mahony provided an update on Google's efforts to put POWER chips in its data centers. Mahony confirmed that Google's POWER9-based platform, named Zaius, had been deployed in Google's data centers. The platform is "Google-strong," according to Mahony.
An IBM POWER9 chip. Image source: IBM.
A big step forward for IBM
IBM's goal is to make POWER a significant player in the high-performance computing and hyperscale data center markets, eating away at Intel's (INTC -1.60%) dominance. Third parties can now build systems around POWER chips, and even design custom versions. Over 200 foundation members contribute technology to the initiative. That includes tech giants NVIDIA and Micron.
Back in 2015, an IBM executive quantified what IBM hoped to achieve. A 10% to 20% share of both the high-performance computing and hyperscale data center markets would be considered a victory. With Google finally using POWER9 chips for production workloads, IBM now has a foothold in hyperscale data centers.
Why, and for what, is Google using POWER9 processors? Google found that the performance of its web search algorithm, the heart and soul of the company, scaled well with both the number of cores and the number of threads available to it. IBM's POWER9 processor is a many-core, many-thread beast. Variants of the chip range from 12 to 24 cores, with eight threads per core for the 12-core version and four threads per core for the 24-core version. Intel's chips support only two threads per core via hyperthreading.
The bottom line is that IBM's POWER9 chips are ideally suited for workloads that fully take advantage of the large number of threads available. Google's web search is one such workload. They're not well suited for workloads that don't benefit from more threads, which is why the market-share ceiling for POWER isn't all that high.
Moving data fast
Mahony also talked about the importance of bandwidth. It doesn't matter how fast a processor is if it can't move data fast enough. IBM claims that one of its POWER9-based systems can transfer data up to 9.5 times faster than an Intel-based system, using OpenCAPI and NVIDIA NVLink technology. That's important for any kind of big data or artificial intelligence (AI) workload.
AI workloads are often accelerated by GPUs or other specialized hardware. Google developed its own accelerator, the Tensor Processing Unit, which it uses in its own data centers for AI tasks. But these accelerators still require a host processor that can move data fast enough. Here's how Mahony explained it:
And so, what that means is you've built this beautiful accelerator. You invested all of this effort, time, and resources to make that an amazing hardware platform. But if the host processor can't keep up with that, as my colleague said: "Did you really have an accelerator? Did you really accelerate the fleet?"
Early innings
Mahony didn't provide any hard numbers on the scale of the POWER9 rollout in Google's data centers, but she did say that the company has reached the phase where it's ready to scale up the number of applications and the machine count. Google does an awful lot of AI-related computing, so it's not a stretch to assume that this rollout will be a meaningful size.
Google's use of POWER9 chips in its data centers represents a big vote of confidence in the technology. It's taken a while, but IBM's efforts to provide an alternative to Intel in the data center are starting to bear fruit.
