Back to top

Nvidia admits Fermi GPU supply problems

Nvidia CEO Jen-hsun Huang has admitted that the firm's Fermi architecture is struggling to get out the blocks, as it is having trouble making the 40nm chips.

At an Nvidian analysts' day yesterday, Huang was forced to admit what has been well known for ages: Fermi is such a big, cumbersome, transistor-heavy piece of silicon, that Nvidia's fab partner TSMC is struggling to get it out the door in any volume.

"From a supply perspective, we wish we had more 40nm capacity," Huang confessed. "We are working with TSMC really closely. They are doing a fab [sic] job. Yields are improving. Capacities are improving. But we are finding it hard to keep up. Everyone is clamoring to have Fermis out the door. We are working really hard to get Fermis out the door."

Nvidia formally launched its GeForce GTX 400 series, recently and while there are a few 470s kicking about, the flagship 480s are rarer than hen's teeth.

Even at 40nm, Fermi is a big chip and yields at TSMC are low. Huang, told an analyst meeting at the company–Ěs headquarters in Santa Clara, California yesterday that Nvidia had shipped "a lot" of 40nm chips although those launched as Fermi chips had the architecture tweaked in order to be able to supply.

Huang said Nvidia wasn't going to go sniffing around for a new fab. Rather it would tighten up its relationship with TSMC

Rival AMD also makes 40nm chips at TSMC. It has managed to get a full line-up of DX11 graphics chips out in the marketplace, although some of these are also experiencing supply constraints.

Tags: tsmc, nvidia, jen-hsun huang