Nvidia’s “in pretty good shape on supply” with next-gen GPUs on different nodes

Nvidia believes that by splitting manufacturing across different TSMC production nodes – 12nm for its current GPUs and presumably 7nm for its future Nvidia Ampere, Hopper, or whatever architectures – it can keep supply up with graphics card demand this year.

Though Nvidia CEO, Jen-Hsun Huang, has warned that it “surely won’t have ample supply” of graphics cards this year because of just how tight things are across the industry… mostly because of the PC world’s almost complete reliance on getting its finished silicon out of TSMC’s factories. But the ever-effusive Huang still believes the company is “in pretty good shape on supply” because of dispersing its manufacturing across different TSMC fabs and, more importantly, across different production nodes.

Speaking during Nvidia’s recent Q4 earnings call, Huang explained that the entire silicon industry is experiencing tight supply at the moment, so it’s not alone in its reliance on TSMC’s third-party fabrication facilities. But he claims that by working closely with Nvidia’s customers, making smart guesses on the level of future demand, and working very closely with TSMC, the green team should be in good shape when it comes to getting new GPUs into the hands of its customers.

But then, he would say that in front of investors and industry analysts, right? The fact that Huang’s been up front about the fact that it’s not going to have “ample supply” means that should anything go wrong with TSMC’s manufacturing graphics cards could dry up real quick.

All of Nvidia’s current best graphics cards are manufactured on TSMC’s 12nm FinFET process, from Turing’s lowly GTX 1650 all the way up to the Volta-powered V100 monster chip. That puts all of the green team’s green eggs into a single silicon basket, but with a new generation of graphics cards set for this year Nvidia is splitting its manufacturing across different production nodes.

“We’ve got a lot of different factories,” says Huang, “and several different nodes of process qualified. I think we’re in good shape. And so we just have to watch it closely.”

From that it looks like Nvidia is hoping that by splitting demand across multiple fabs and production lithographies it can keep the GPU train chugging along all year, barring manufacturing catastrophes within TSMC.

And what of these new architectures on those new production nodes? We were hoping to hear something official out of Nvidia in March at its Graphics Technology Conference – whether that’s confirmation of the Nvidia Ampere codename, or something else entirely we still don’t know – but fingers crossed GTC doesn’t go the way of MWC in Barcelona…