Kepler chip drought leaves Nvidia gasping for moolah

Supply shortages for 28 nanometer GPUs from fab partner Taiwan Semiconductor Manufacturing Corp adversely impacted GPU chip and coprocessor maker Nvidia once again in its most recent quarter.

Revenues for the first quarter of fiscal 2013 ended in April came in at the high end of the company's guidance, but nonetheless fell by by 3.9 per cent to $924.9m. Because Nvidia was adding people and carrying significantly higher sales, research, and development costs during the quarter (which would have been offset by sales of the new "Kepler" family of GPUs had they shipped on time late last year and been ramping in discrete graphics cards), the company's net income was chopped by more than half to $60.4m.

In a conference call with Wall Street analysts, Karen Burns, Nvidia's interim CFO, said that GPU revenues were down 6.7 per cent sequentially to $579.7m from the final quarter of fiscal 2012 ended in January, with a sales slump for both high-end desktop and notebook GPUs because demand for the new Kepler GPUs outpaced supplies. That said, Nvidia was still selling Fermi generation GPUs in conjunction with the "Ivy Bridge" PC ramp from Intel, which kicked off during the quarter. The problem is that the mix shifted to the midrange products because everyone knew the high-end was going to be revamped with Kepler GPUs, and thus margins suffered.

Professional Solutions, the unit of Nvidia that makes high-end GPUs for workstations as well as the Tesla GPU coprocessors for servers and supercomputer clusters, was down 4.2 per cent sequentially to $212.6m. Nvidia has not seen fit to break out Tesla sales separately from high-end GPUs, but someday, if this business grows as Nvidia hopes and many supercomputer centers are betting, that will change.

The Consumer Products group at Nvidia had $132.6m in revenues, up 20.8 per cent from the prior quarter. Consumer Products includes Tegra hybrid ARM-GPU chips for smartphones and tablets, Icera baseband processors, GPU and intellectual property sales to game console makers, and various embedded products. Tegra product revenues were up 50 per cent sequentially.

While the wait for Kepler is no doubt annoying to Nvidia's top brass, Wall Street, and customers wanting to get their hands on one, Nvidia has $3.13bn in the bank and a promising, if slow-motion, product ramp coming this year.

Looking ahead to its fiscal second quarter ending in July, Nvidia expects for sales to be somewhere between $990m and $1.05bn, which is slightly down from the $1.017bn Nvidia brought in during Q2 fiscal 2012. The company brought $151.6m to the bottom line in that quarter, which worked out to 25 cents per share. Nvidia does not forecast earnings per share, so it is not clear if it believes it can meet that profit level again.

Jen-Hsun Huang, president and CEO at the chip maker, said on the call that he expected for 28 nanometer Kepler parts to be supply-constrained through the remainder of the year. And ironically, Huang said that 28 nanometer product yields at TSMC were "probably the best of any new node that TSMC has ever done," and that it really is a matter of so many different companies ­wanting to have TSMC make their chips – Advanced Micro Devices and Oracle both use TSMC for their foundry and are asking for 28 nanometer chips, just to name two.

"There's just not enough capacity and the planning process, when we think about these new nodes, it just simply needs to be better," Huang said on the call. "And I don't think there's any way around that."

He added that in the coming quarter, 28 nanometer products would be about 30 per cent of the product mix and that he wanted it to be a lot higher.

There is no easy way to make a fab pump out more wafers more quickly, and chip designs are tied very closely to a fab's etching processes and machines. So TSMC's partners will just have to ride it out and chip sellers will just have to hope that their wafer agreements get them close to the front of the line at TSMC. ®