Hacker News Reader

Comment by vessenes | Hacker News Reader

The actual quote is excellent (and already on this page). He's right -- the list of things to be done is obvious, and will take time.

And, I'd like to imagine he has some enjoyment at the thought -- if you saw his keynote showing off a precomputed(ish) raytraced walkthrough of their new office space, he was really enjoying himself and the possibilities.

What Hsun has is an incredibly good window into the silicon side of the problem, and he knows how many chip cycles he's got to address the amount of computation needed for physics, visual realism and presumably a latency budget to allow wireless support. Plus he can estimate what screen tech will be doing.

It's a prodigious amount of R&D, and it is a bit breathtaking. Finally having a reason to do it, and a taste of how appealing it can be is pretty cool.

View on HN

You mean how many chip cycles he needs, right? How many he's got we all know: maybe 2. Then CMOS scaling is over. (Not that things will completely stop evolving, but they'll surely slow down.)

well TSV have evolved greatly which means that 3D dies can come soon. 3D transistors already expanded the limit of how small parts can get before physics hits you in the face with a baseball bat. On top of that memristors and graphene could open a whole new world for chip designers. If we have only 2 cycles worth of process left we are doomed :)

> 3D transistors already expanded the limit of how small parts can get before physics hits you in the face with a baseball bat.

Yes, but we are just starting 10nm now. Go two more nodes (if I understand correctly that is meant here by cycles) and you are at 5nm.

https://en.wikipedia.org/wiki/5_nanometer

to put that into perspective: "An Australian team announced that they fabricated a single functional transistor out of 7 atoms that measured 4 nm in length."

Maybe we can make our way down to single atom transistors eventually, but who knows how long it will take to make that commercially viable.

We already have a single atom transistor (https://en.wikipedia.org/wiki/Single-atom_transistor), but you're right that it's far from commercially viable.

In 20 years? I think he's got at least 4, maybe 5. Of course, I don't know how big those evolutions will be in terms of advances, but I'm just a silicon dilettante, nothing like an expert.

I think the dispute here is that there are two different standards in play. He has time for 4-5 cycles, but Intel only has concrete plans for ~2 cycles. After that the existing die structures can't be meaningfully reduced anymore, and everything is up in the air. Maybe we get some wild, big-but-fast optical chips, maybe we get a breakdown of Moore's law.

Of course, 3d chips look fairly achievable and might eke out a few more cycles, but things will get interesting sometime before that 20 year deadline.

I don't think Intel plans for more than 2 fabrication cycles regardless of some gloom and doom omg quantum physics go away plox prophecy.

14nm FinFet GPU's are only coming in now, and they'll stay on 14nm for at least 3-4 years. 10nm GPU's will probably come out in 2019-2020, 7nm GPU's will come out in 2025 or so, and 5nm GPU's will come out in 2030, that's already ~15 years and considering that everyone is predicting that we'll have to stick to processes longer as the small we get the more time it takes to actually optimize them for good yields and performance you probably have 3-5 years there too.

So even if we do not adopt any major process changes like switching to graphene, solid state photonics to replace the metal pathways and even optical transistors or w/e else people are cooking up to enable us to continue to improve our integrated circuits he can still be correct about the time frame.

Also regardless of what happens or not he was talking about true VR immersion this means that allot has to change other than just how good and convincing the graphics are this also includes learning to develop VR applications and as VR is a very new media that alone can take 20 years.

This is a good summary, thanks. And yeah, I only meant to speak to the argument in the comments over how many cycles we can "fit" in the next 20 years.

I think a lot of VR's biggest hurdles for that time aren't dependent on processing power. Focus and resolution issues will side-step some of it (seriously, our eyes are terrible outside of their central focus region!) because there's simply no need to render everything flawlessly as on a TV. Beyond that, there are going to be all kinds of design hurdles. Right now, we have nausea-inducing roller coaster rides and people running into walls because we can't sync up the (already impressive) visual experience with inner-ear confusion and physical boundaries. Just working out what the most compelling designs are for our current VR tech is the work of years, and that's without doing it on the treadmill of ever-improving tech!

Do you have a link to the keynote video?

Just found it — think this is it.

https://youtu.be/Z9Og9xtAY3w?t=19m27s

Thanks.

Here it is https://www.youtube.com/watch?v=YCxFGxqLsHE

Reply on news.ycombinator.com