On Nov. 18, NVIDIA (NASDAQ:NVDA) and IBM (NYSE:IBM) announced a strategic partnership. In particular, it seems that both companies will be collaborating to produce GPU-accelerated editions of IBM's enterprise software. In addition, IBM will be building enterprise systems that combine its POWER 8 CPUs along with NVIDIA's Tesla GPU accelerators. This, in this Fool's view, is the correct strategic decision for both NVIDIA and IBM, particularly in light of Intel's (NASDAQ:INTC) internal high performance computing ("HPC") accelerator efforts.
NVIDIA faces tough competition from Intel
NVIDIA has a little bit of an Intel problem. While the fears run rampant that NVIDIA's graphics processing units for PCs will eventually be displaced by integrated graphics processors on Intel's CPUs (this is another topic for another day, but it's not as likely as it seems), the real concern should be that Intel is making a very serious pass at the high performance computing accelerator market. As it stands today, NVIDIA's GPU accelerators plug into the same boards as Intel's CPUs, meaning that NVIDIA's products act as co-processors for Intel's chips – a mutually beneficial proposition.
However, as Intel aims to take more of the datacenter pie, it has committed to developing its own HPC accelerators under the brand "Xeon Phi." Clearly this is a problem for NVIDIA. While the first iteration of the Xeon Phi wasn't much of a threat to NVIDIA's Kepler-based Tesla chips, Intel has signaled that fairly dramatic upgrade for its next generation "Knights Landing" chip. Not only will Intel have much more competitive accelerators, but the next iteration of Xeon Phi will also be available as a stand-alone processor (and NVIDIA is unlikely to be able to sell accelerators into systems with this configuration).
NVIDIA's backup plans
Now, NVIDIA has signaled that its next generation GPU, codenamed "Maxwell," would have integrated NVIDIA-designed ARM-compatible cores on the die, which suggests that it, too, would be able to run as a stand-alone part. NVIDIA's CUDA is very well adopted in the HPC ecosystem, which could give its own parts a leg up, but at the same time much of the software is written primarily for X86 with the relevant portions offloaded to the GPU. The advantage for Intel is that its Xeon Phi is X86 compatible, potentially making utilization of the resources much simpler on the part of the software developers.
That being said, if NVIDIA's go-at-it-alone approach with stand-alone accelerators doesn't work or doesn't generate enough sales to replace lost revenue from Intel's Xeon Phi in Intel-based systems, it's good to have a backup plan: enter IBM. The sales of systems with IBM's POWER series of chips have continued to plummet as UNIX system sales have plummeted and as Intel CPU + NVIDIA GPU based systems continue to take share in HPC, so why shouldn't NVIDIA kill two birds with one stone by partnering with IBM?
NVIDIA ends up reducing its risk profile by gaining exposure to (and potentially helping to reinvigorate) systems with IBM's POWER and IBM gains the world's leading HPC accelerator vendor as a partner. It's a great, mutually beneficial match. Further, since IBM is a company with very legitimate server chops (from systems to the chips inside the systems) this isn't some fluffy, PR-driven partnership. There is real money to be made here for NVIDIA.
Foolish bottom line
NVIDIA has always been the company that finds a way to thrive even when it seems that everything is stacked against them. One of the biggest longer term threats – Intel in the HPC space – has been partially diffused with this IBM partnership. While NVIDIA should (and will) still fight diligently for spots in Intel-based servers, getting CUDA into POWER-based systems seems like a great hedge. The next thing to look for would be a similar deal with Oracle and its SPARC based systems.