Hacker News Reader

Comment by hinkley | Hacker News Reader

I never really thought about this before, but a thermodynamics breakthrough in power supply design, heat management, or power generation could qualify as a state secret.

We know that the NSA makes custom chips, and it's probably harder to know how many calculations per second that hardware can manage than predicting the capacity of COTS hardware based on how hot the building is and how hot the nearest powerplant is getting.

One of the things I loved about the history of the Blackbird is that it held the record for fastest plane, and when a new pretender to the throne arrived, on a couple different occasions they sent up another Blackbird pilot and retook the crown. When a private individual breaks a record, they tend to go as fast as they can. Bragging rights for an military airplane are "mine is faster than yours" not "mine is exactly this much faster than yours". Just keep that classified.

View on HN

Note https://www.nitrd.gov/pubs/nsa/sta.pdf promising ~100Ghz cryogenic processors by 2010 with sufficient (but seemingly not ridiculous) government investment.

Fascinating. I see three possibilities:

1. the project succeeded and the government has kept cryogenic processors totally dark for a decade with no public leaks or hints to industry (far-fetched, though Skunkworks hid the Blackbird pretty well.)

2. they aggressively funded it as a black project but hit insurmountable design challenges (like quantum computers) or bureaucratic abyss (like the F35.)

3. the funding never materialized.

are any research teams working on superconducting classical computers, rather than quantum computers? does nonlinearity get in the way?

4. Attempts to create cryogenic CPUs were made well before 2005 and failed. The project was made somewhat public in order to misdirect adversaries into spending their budget and efforts on cryogenic processors.

5. The project was funded as a black project without the intention to actually deliver results but rather to justify RSFQ circuits R&D to later use those results elsewhere

I was trying to find an old article about someone who overclocked either a 486 or a Pentium to hell and gone [edit: with mineral oil and dry ice] so I could speculate on what sort of speed increase one could get from spending a ridiculous amount of time and effort on overclocking.

I didn't find any of what I was looking for, but I stumbled back onto the fact that Cray II's were cooled by running fluorinert over the components and that the NSA was one of their biggest customers at the time.

Whose to say they aren't still using fluorinert on servers?

But I've also heard rumors about TLAs getting access to pre-production versions of consumer chips (eg, when yields are still sad). And who knows what binning is going on at Intel. If I have a really good production run and I get 95 chips that have absolutely zero flaws, who do I sell them to? When you manufacture something you can't guarantee a certain yield to be above your wildest expectations, so would I even create a product number for the unicorns? Do I bin them with 'the best' chips even though I know they're better than the best?

Or do I just find someone with deep pockets who's happy to take whatever I've got, be it 3 or 300?

http://www.cpushack.com/2018/07/03/cpu-of-the-day-the-intel-...

Reply on news.ycombinator.com