A one-dimensional transistor

HPE is still making sensational claims about the Machine project and its commercial viability.

The impression I got the last time I talked to an HPE person on the subject (so this is third-hand subjective, take it as you will) is that the hard technology has been ready and practical to some some sort of productization for a while now, but the entire project got boondoggled in trying to "truly revolutionize" and rebuild computing from first principles including brand new OSes and applications...

It's sounds like an interesting lesson in a sort of reverse Sagan Effect where "we have a new whipped cream" turns into "well, to really sell it we first must produce an apple pie from scratch... and do that well, how about we try a new big bang?"

Well, this means that we have a potentially great technology that possibly can make a difference 10 years from now (or more). It's good to know, but it's not something to expect in new consumer products by the coming Xmas.

Many things were invented way before their mass adoption. For instance, LED effect was known since 1907, first prototypes built in 1920s, but commercial LED indicators only took up in 1970s, and LED lighting, in 2010s.

So we just have to wait — unless you're a solid state physics researcher, then you can actually work on making it practical.

We have the ability to build eDRAM with latency under 2ns and directly attach mass quantities to a CPU.

But it's more cost-effective to have a fast L2 and let your L3 and main memory be slower. In a multi-CPU machine you have to wait around anyway for a lot of memory accesses.

It's not the memory technology that bottlenecks memory. It's locating and routing data. And that cares a lot about better transistors.

They will just market it as 3D anyway, and why not add a HD too. If a toothpaste company can do it, they can.

The average consumer won't really need to know which is better.