The question of when Nvidia would launch its next-generation GPUs has been a matter of some discussion of late. Earlier this year, we predicted a midsummer launch, only to have CEO Jen-Hsun Huang seemingly shoot down that idea with his declaration that the next GPU generation wouldn’t appear for “a long time.” Then Nvidia pulled out of a previous talk it was scheduled to give on its next-generation consumer GPU at Hot Chips (also in August). But later rumors and some controversial NDA changes that circulated around have both pointed to a release later this summer and additional evidence to that effect has now reportedly surfaced.
ExtremeTech is not bound by any NDA on this topic and is therefore free to cover this information.
The news comes via Overclockers3D. Gamer Meld, a YouTube channel, claims to have received an email from a GPU retailer that contained information on next-generation launch timing, which, as always, may or may not be accurate and should be taken accordingly. A copy of the email is below:
The email notes a launch date of 8/30/2018 for the GTX 1180, 9/30 for some variations of the GTX 1170 and a GTX 1180+, and 10/30 for the GTX 1160 as well as some other card variants. Staggering the launch out like this would make some rumors of a late Q4 refresh for AMD hardware make sense, though this is supposedly a 12nm shift for Polaris with a modest amount of actual performance uplift. Assuming the launch timing is accurate, it implies that these GPUs are built on TSMC’s 12nm optimization of its 16nm technology (12nm should be considered a marketing term for both TSMC and GlobalFoundries rather than a fundamentally new node). Then again, given how strong Nvidia’s Maxwell refresh was on the same node as the earlier Kepler, Nvidia fans are likely to feel pretty good about the idea of a new architecture without a node shrink.
Let me be clear, though — 12nm is our guess for this hardware, but it’s still a guess. Shipping a GPU as an early 7nm product would, however, be a daring move for Nvidia for several reasons. In the last few years, we’ve seen foundries shift to lower nanometers with smartphone SoCs first, and Apple typically leads the pack in that regard, even going so far as to reportedly pay TSMC to be first on 7nm. We know AMD is also building on 7nm at TSMC, but its Vega Machine Intelligence GPU isn’t expected until the end of this year (post-iPhone) and will also ship in markets that can afford to pay for cutting-edge silicon. Finally, we’ve seen GPU vendors struggle with availability when making node transitions, even when the nodes aren’t new. It took AMD and Nvidia months to ship 14nm in volume, even though that process node had been deployed for smartphone SoCs for over a year before Nvidia shipped its first 14nm GPU.
It seems quite unlikely that the company would hit 7nm on consumer now, but it’s not impossible.
Past that point, expectations for Nvidia’s first new generation in two years are generally high. Hopefully, the company’s new products will deliver.