Been there, done that. In 2003 the RFC 3526 defined standardized DH params up to 8192 bit length.
Best practice since years for IPSec is at least group 14 (2048 bit), but web servers are a problem: Apache’s support for params > 1024 bit was added in July 2015!
Bigger then!!! WE NEED BIGGER PRIMES!!!
It would be interesting to know, however, what actual percentage of available computing needed those address spaces in the 2000s, rather than by type of problem.
Also, the possibility exists that the RISC cores of these machines might have quite a different interface wrapped around them than the commercial one. Compatibility with the ordinary Intel instruction set isn’t necessary if you are not running an off the shelf OS. (I have the Transmeta processor in mind).
Another (farfetched?) option: They built their own chip fabrication facility.
While the hardware needed to build a prime cracking supercomputer is not a trivial amount, fundamentally it isn’t much different from any other supercomputer application. ASICs buy you lower power consumption and higher speed per chip but if you aren’t constrained by budget (as the NSA would not be) you can just throw more off the shelf CPUs at the problem.
I’m not even sure the NSA would have to hide their procurement. Its no secret that they own and operate a massive computational network. On a global scale, I wouldn’t be surprised if the NSA procurement is smaller than Amazon, Google, Apple, MS, etc.
The only Intel production figures I could find was that in 2012 Intel made about 350 million x86 chips (and apparently Intel was not running at 100% capacity). A quick glance at the top500 supercomputer list says that between 500k and 3 million cores are used for a top supercomputer so about 100k to 800k individual CPUs per super computer. Even if the NSA was secretly operating the most powerful supercomputer on Earth, it would only take Intel a few days to make the chips for it (assuming the figures I found were close to accurate).
I don’t think it’s that farfetched. If I can sit around and think up an ASIC that I want to build and actually go through with it, then they could have thousands of people designing ASICs for decrypto. They could have so many people on it that they could divide the problem and no single person would know what it’s really for.
They could be doing that same thing with software running on massive distributed clusters.
I read somewhere that the NSA was outsourcing some of their search “problem space” to Google. Can’t find a link right now though. 
But does that number include NSA purchases? I rather doubt it or somebody could do the math and know right quickly about how many chips weren’t accounted for by the public market.
And he shall judge among the nations, and shall rebuke many people: and they shall beat their cameras into plowshares, and their servers into pruninghooks: nation shall not lift up broken-crypto against nation, neither shall they learn subterfuge any more.
I’m going to repeat that at the Lodge tonight, word for word, of course I’ll have abided in large quantities of Whiskey first.
No! Using a laser to burn the optics from a distance is easier and less risky than either. I don’t know why more people don’t do it. Maybe I’ll need to start doing workshops or something…
That’s a nice animation of how a key works in a pin-tumbler lock. Attributing it to its creator, Deviant Ollam of TOOOL, might also be nice.
How much of a laser do you need to burn a camera?
I’m not positive, but it strikes me as very likely that this is not the kind of task that would be improved by an ASIC. GPU-accelerated, maybe, but those are just a whole lot of not very powerful general purpose CPUs running in parallel.
The cutting edge stuff always relies on economies of scale that a government couldn’t match, anyway, so I would be shocked if they don’t just buy the chips to make their supercomputers off the shelf like anyone else. You can’t make an ASIC that will get you more FLOPS than Intel’s or IBM’s latest.
That said, the NSA has put a whole lot of money into far out semiconductor R&D, particularly in fields related to lithography, in the past. I don’t know if they still do that, but all that research wound up in the hands of commercial companies before anyone made actual chips with it. It wasn’t like they developed their own chips, they just wanted to generally advance the tech for making the chips.
It depends upon the camera, and what kind of image sensor it uses. A laser not powerful enough to burn it will still in many cases blind it temporarily, allowing it to be disabled unseen (by that camera, at least). My guess is that a beam which is powerful enough to etch wood and plastic can be enough to damage most CCDs. Another factor is that the beam needs to pass through various camera optics.
Sounds like a fun area for experimentation by those who aren’t already too busy tinkering with too many other things.
Sure, on an individual hardware scale. But FLOP for FLOP, you can make many more ASICs for much cheaper than you can make that same amount of CPU. If all you are doing is running one algorithm over and over against different inputs, then an ASIC is the thing for that purpose. All the other shit that runs a CPU has been cleared away, so all it’s doing is a little i/o and that single algorithm. And you can make millions of them for the cost of hundreds of CPUs. An ASIC is highly scalable in terms of production, which translates to more FLOPS in the final tally. They are highly un-scalable in terms of redirecting them for other uses, of course.
If I were the NSA, I would be banking primes. I’d isolate a few good algorithms for generating them, and I’d have a massive array of computation generating new primes, secretly, and not ever tell anyone. Then I’d have them stored and on hand for when I need to use them later.
Using non-standard primes will definitely make you stand out, possibly enough to justify further investigation.
Good point. Done.
OK, then OPs, Mr. Stickler. You ought to hold yourself to the same high level of precision. We will be watching for when you slip up and pounce when you do.
