Hacker News Reader

Comment by ayngg | Hacker News Reader

I thought they are using TSMC for their gpu, which means they will be part of the same bottleneck that is affecting everyone else.

View on HN

That depends on how much of TSMC's capacity they've reserved, and which node they are manufacturing on. I'm guessing that a low end GPU doesn't need a 5nm node. I think these are 7nm, which is still going to be booked tightly but probably not as heavily bottlenecked as smaller nodes.

I believe nvidia doesn't use TSMC or not only use TSMC.

30 series Nvidia cards are on Samsung silicon iirc

Yeah, Samsung 8nm, which is basically Samsung 10nm++++.

10nm--?

The '+' in this case is a common process node trope where improvements to a node over time that involve rules changes become Node+, Node++, Node+++, etc. So this is a node that started as Samsung 10nm, but they made enough changes to it that they started marketing it as 8nm. When they started talking about it, it wasn't clear if it was a more manufacturable 7nm or instead a 10nm with lots of improvements, so I drop the 10nm++++ to help give some context.

The datacenter cards (which are about half of their revenue) are running on TSMC.

Yeah they use Samsung for their current series but are planning to move to TSMC for the next irrc.

Independent of the question around TSMC they are still affected as:

- Shortages and price hikes caused by various effect are not limit to the GPU chiplet but also most other parts on the GPU.

- Especially it also affects the RAM they are using, which can be a big deal wrt. pricing and availability.

Currently the bottleneck is substrates, not TSMC.

Except Apple

Reply on news.ycombinator.com