undefined | Better HN

0 pointsdrivebycomment5y ago0 comments

You're not wrong, but most of HN threads are like this. 80% of comments are low information, Dunning-Kruger effect in action. But among that, there are still some useful gems, so despite what you said, HN is still worth it. If you fold the first two top level comments, the rest have some useful , informed perspective.

I don't see this as having that much of an impact on any short to medium term. ARM has too much intricate business dependencies and contracts nVidia can't just get out of.

My speculation is that nVidia might be what it takes to push arm to overcome the final huddle into more general purpose and server cpu, and achieve that pipedream of a single binary/ISA running everywhere. Humanity would be better off if a single ISA does become truly universal. Whether business/technology politics will allow that to happen and whether nVidia has enough understanding and the shrewdness to pull that off is to be seen.

0 comments

dahart5y ago

> Humanity would be better off of a single ISA does become truly universal.

This is an interesting thought. I think I would agree with this in the CPU world of the last 20-30 years, but it makes me wonder a few things. Might a universal ISA eliminate major pieces of competitive advantage for chip makers, and/or stall innovation? It does feel like non-vector instructions are somewhat settled, but vector instructions haven't yet, and GPUs are changing rapidly (take NVIDIA's Tensor cores and ray tracing cores for example). With Moore's law coming to an end, a lot of people are talking about special purpose chips more and more, TPUs being a pretty obvious example, and as nice as it might be to settle on a universal ISA, it seems like we're all about to start seeing larger differences more frequently, no?

> 80% of comments are low information, Dunning-Kruger effect in action.

I really liked all of your comment except this. I have a specific but humble request aside from the negative commentary & assumption about behavior. Please consider erasing the term "Dunning-Kruger effect" from your mind. It is being used here incorrectly, and it is very widely misunderstood and abused. There is no such effect. The experiments in the paper do not show what was claimed, the paper absolutely does not support the popular notion that confidence is a sign of incompetence. (Please read the actual paper -- the experiments demonstrated a positive correlation between confidence and competence!) There have been some very wonderful analyses of how wrong the Dunning-Kruger paper was, yet most people only seem to remember the (incorrect) summary that confidence is a sign of incompetence.

https://www.talyarkoni.org/blog/2010/07/07/what-the-dunning-...

drivebycommentOP5y ago

> Might a universal ISA eliminate major pieces of competitive advantage for chip makers, and/or stall innovation?

That's a good question. I didn't try to put all the necessary nuances in a single sentence, so you're right to question a lot of the unsaid assumptions. I don't know for sure at this point the innovation in ISA has run most of its course yet, but I do feel like we kind of have, given how relatively little difference it makes. I think a "truly" universal ISA, if it ever happens, would necessarily have to have a governance and evolution cycles, so that people will have to agree to the core part of the universal ISA, yet have a room and a way to let others experiment various extensions, including vector extensions for example, and have a process to reconcile and agree on the standard adoption. I don't know if that's actually possible - it might be very difficult or impossible for many different reasons. But if such can happen, that would be beneficial, as it would reduce certain amount of duplication, and unlock certain new possibilities.

> Please consider erasing the term "Dunning-Kruger effect" from your mind. It is being used here incorrectly

Duly noted.

j / k navigate · click thread line to collapse