Nvidia dollars lifted by Intel's Sandy Bridge ramp, payola
Tesla GPUs disappoint
Nvidia has frenemy Intel to thank for its improving financial situation in its first quarter of fiscal 2012.
Despite the fact that the graphics and CPU chip maker's revenues and profits for its first quarter ended May 1 were down – revenues fell 4 per cent, to $962m, and net income was down 1.7 per cent, to $135.2m – Wall Street will nonetheless be pleased. Why?
Because Intel and Nvidia buried the hatchet and cross-licensed each other's GPU technologies back in January, which brought Nvidia $22m in the quarter, a small portion of its $1.5bn haul from Chipzilla that will spread out over years. (That $22m was just one month's worth of the licensing money, in fact.) More importantly, the two buried the hatchet (some would say in common rival Advanced Micro Devices' skull) just in time for the "Sandy Bridge" Core ramp on desktop and notebook PCs, an upgrade wave that Nvidia is expecting to ride like a dude.
Nvidia's core GPU business was up 3.8 per cent in the first quarter, to $637.6m, and its consumer products business – bolstered by its Tegra 2 system on chips – rocketed up 78.2 per cent, to $122.6m.
The company has been winding down its CPU chipset business for years, which was "down significantly" in the quarter, according to Michael Hara, vice president of investor relations at Nvidia, who spoke on the conference call with Wall Street analysts. Hara added that notebook GPU sales for Sandy Bridge-based machines more than made up for the loss of the chipset biz, and that Sandy Bridge chips were expected to reach 75 per cent of Intel's shipments in the current quarter and that based on increasing discrete GPU attachment rates on these processors, Nvidia expected to gain share. Desktop GPUs were down sequentially from Q4, but did better than expected.
Nvidia also singled out the dual-GPU GTX590 discrete graphics card for gamers as being a good seller, and yes, it is supported on the new Crysis 2 and you can help rid New York City of aliens with it.
If there was one disappointment in the quarter, it was Nvidia's Professional Solutions unit, which sells GPUs for workstations and Tesla GPU co-processors for workstations and servers.
"Frankly, I wish our Professional Solutions business was growing faster, too," admitted Jen-Hsun Huang, president and CEO at Nvidia, in the call.
Huang said that Tesla GPUs were still in the early adoption phase, but that the lumpiness in this business was due more to a few big deals that happen some quarters and then not in others. Nvidia had just such a big order in its fourth quarter of fiscal 2011, and that is why the Tesla business was down sequentially. Huang said that the underlying, core Tesla business is growing steadily as software vendors port their code to CUDA and use GPUs to accelerate its performance, adding that if they change maybe 10 per cent of their code, they can see acceleration of between 5X and 10X on their applications.
"That's too compelling to ignore," Huang declared. And he is absolutely right. But coding for GPUs cannot be as simple as Nvidia makes it out to be with CUDA, even if it is a lot easier than for other accelerators in the past, or else it would have been done already.
Huang said that there are several million workstations sold each year, and that with CUDA and a mix of GPUs and GPU co-processors, users will be able to not just use their machines for visualization, but also for computation and simulation. The problem comes back to software, of course, but he was convinced that Nvidia and its workstation hardware and software partners would not be satisfied with a workstation business that grows in lockstep with gross domestic product, but would want to give workstations more computational capabilities than end users have ever been able to lay their hands on thanks to GPUs.
"No competition is in the path of our growth," said Huang, adding that it was really about demonstrating new capabilities with software partners.
Nvidia ended the quarter with 2.73bn in cash in the bank, and expects its acquisition of mobile chip maker Icera to complete in late May or early June. Huang did not talk much about Nvidia's integration plans for Icera's Livanto baseband chip, and said that the idea was to supply the two processors – Tegra and Livanto – for the post-PC era, not to come up with a single, integrated design. Huang did leave the door open for such integration if Nvidia and its OEM partners thought it was valuable. Icera has burned nearly $300m in venture capital funding designing its Livanto chips and software, and Nvidia is paying $367m in cash to acquire it.
Looking ahead, Nvidia said it expects sales to rise by 4 to 6 per cent sequentially in its fiscal second quarter, which would put sales at between $1bn and $1.02bn. ®