Earlier this month, NVIDIA (NASDAQ:NVDA) and IBM (NYSE:IBM) announced a partnership that will put NVIDIA's Tesla line of GPUs into IBM's servers, as well as accelerate IBM's enterprise software using NVIDIA's hardware. With NVIDIA trying to expand beyond gaming graphics, and IBM trying to revive its slumping hardware business, this deal should benefit both companies. The loser in this deal is Intel (NASDAQ:INTC), which sells its own line of accelerators.
What this deal means for NVIDIA
Graphics cards are very good at doing simple computational operations an enormous number of times. This is perfect for games, for example, because thousands of polygons need to be transformed and drawn to the screen very quickly. It's also good for a variety of other applications, such as scientific simulations. NVIDIA created the Tesla line of GPUs specifically for these types of applications.
The Tesla business has been growing fast, but NVIDIA still relies on gaming GPUs for much of its revenue and profit. This deal is a big step toward a more diversified NVIDIA, and it opens up new doors for the company's products. Until now, the Tesla line has been used strictly for supercomputing applications, but this deal puts the product to use in data centers, a potentially huge market for NVIDIA. With big data analysis growing increasingly important for companies, NVIDIA's products allow for a massive performance boost over traditional systems.
What this deal means for IBM
Last quarter, IBM's hardware business fell off a cliff. Sales of Power systems, which use IBM's own Power architecture, plummeted by 38% year over year, leading the company to miss revenue expectations by more than $1 billion. It's become clear that IBM needed to differentiate its hardware from the competition, and the deal with NVIDIA does that.
IBM's bread and butter are its software and services, and having its enterprise software accelerated by NVIDIA's hardware gives IBM a distinct advantage. This move makes it easier for companies to reap the benefits of GPU acceleration, and it gives them a reason to buy IBM's Power servers.
What this deal means for Intel
While NVIDIA's GPUs offer big performance gains over Intel's processors for many applications, Intel has its own line of accelerators that compete directly with NVIDIA's Tesla. The Xeon Phi line of coprocessors is Intel's answer to NVIDIA's Tesla.
Xeon Phi chips currently being sold are in the form of add-on cards, exactly like GPUs. But since these chips are based on the same x86 architecture as Intel's normal CPUs, they can theoretically be used as the main CPU in a system as well. This is significant, because it would allow a system to consist of a single Xeon Phi chip for both running the OS and accelerating applications. This is opposed to a system with an add-on accelerator in addition to a CPU.
Intel has confirmed that the next generation of Xeon Phi chips, due sometime in 2015, will have this property. This is a big step for Intel, and it will give it a significant edge in the high-performance computer market. The IBM deal is certainly a negative, as it presents a big competitor to Intel-based systems, but Intel should become very competitive in the market once 2015 rolls around.
The bottom line
NVIDIA benefits the most from this deal, as it creates a market for its Tesla line for enterprise applications. IBM needs to boost the attractiveness of its Power-based systems, and the addition of GPU acceleration does exactly that. But the game changes in 2015 when Intel releases the next iteration of the Xeon Phi, and I suspect that Intel will be able to grow its market share significantly at that time.