Monday03 October 2022

NVIDIA Might Skip over the 20nm Half-Node and Move Straight to 16nm for New GPUs

Reading time is around minutes.

In the world of silicon size matters, but not in the traditional way. Manufacturers are always looking to make things smaller so they can stuff more into the same space. However there are two types of die shrinks. In the CPU world you see what it commonly referred to as full die shrinks while GPUs tend to do things in half-node steps.

To get a little technical, CPUs follow standard lithographic practices as defined by ITRS (International Technology Roadmap for Semiconductors). This follows an establish path for process shrinks and also can make the reduction easier to handle. For example Intel’s next shrink will go from 22nm to 16nm and then following on to 11nm. GPUs on the other hand use smaller half steps as a sort of stop gap between the established nodes: 28nm to 22nm to 20nm to 16nm etc.

This method has led to some challenges when working on the in-between steps. Right now both nVidia and AMD have had problems with getting their GPUs working at 20nm. AMD spent a considerable amount of time in trying to get things working at 20nm.  TSMC (Taiwanese Silicon Manufacturing Company) not really been able to get things going at 20nm. In April of this year they (TSMC) stated that they were not ready to meet the demand for 20nm for either AMD or nVidia.

Because of this we are now hearing that nVidia might skip over 20nm and push right into 16nm or the next full node process. If this is the case TSMC might find it easier to run at this process level. TSMC was already working with ARM on 16nm as far back as April 2013 so this process should be mature enough to handle nVidia’s needs.  It makes sense for nVidia to skip over the troublesome 20nm node and jump to the ITRS defined node and work with TSMC to get there.

Doing this will allow them to put new GPUs on a more stable process and begin to work on optimizing the design of their next generation GPUs for this smaller and more space efficient process. The 800 series GPUs from nVidia will stay at 28nm and hit the market in H2 2014 while we will most likely see the next gen drop in H2 2015 at 16nm.

It should be an interesting time in 2015 to see what the AMD Vs nVidia landscape looks like. Of course, there is the fact that AMD is also working on 14nm with TSMC… we wonder if that half-node step will work properly or if things are just getting too small for the half steps to keep being the right path.

Tell us what you think in our Forum

Last modified on Thursday, 10 July 2014 11:14

Leave a comment

Make sure you enter all the required information, indicated by an asterisk (*). HTML code is not allowed.