Intel Corp (INTC)

[simplytom]

In February 2016 there were first rumors that 16nm-TSMC-GPUs from nvidia is ready for prime-time.
A jump from 28nm-TSMC to 16nm-TSMC was a big leap forward.

So nvidia made a 'predictable' jump forward.
Nvidia had tailwind because of the new process.

All the fab-guys have problems with the next process-step.
This so called "half-node" steps are blurrying the view.

You can recognize a TSMC full-node when they are moving the GPU-folks to the next node.

So this means 28nm -> 16nm.
So what is the next TSMC full node?
All rumors in the next couple of quarters are about 16nm-gpu.
There are no 10nm-rumors.

Back to Intel.
All process-steps in the last 15years were full-node steps.
32nm -> 22nm -> 14nm -> 10nm

But as we look at it - it is slowing down.
I think that we will see a "more than linear jump" in performance.
But there is an efficiency-wall in the area of 4,5GHz to 5,5Ghz.
No CPU-Pipeline is able to jump over this wall.
So we have to live with it.

I think in the next couple of years we will see a wide pipeline approach.
Something like 10-issue-wide - but this means a completely new Cache-Pipeline a completely different FPU and so on and so on.

There will be an architecture based on the usage of HBM2-Memory.
But this has to be a tailored core.
The complete pipeline has to be destroyed and rebuild specifically for this memory.
I am quite shure that one "A-Team" is working on this approach.

Perhaps there already was one big fail inside the team - but Charly didn't recognize the signs.

If this CPU-monster is ready for primetime the current core can be tweaked for even smaller form factors.