4 ms·
What you say about secret sites may once have been true. It is almost certainly no longer. It is unlikely that at this point, the NSA or others have as much g
by DannyBee 14y ago
What you say about secret sites may once have been true.
It is almost certainly no longer.
It is unlikely that at this point, the NSA or others have as much general purpose computing power as, for example, Google.
Why would they?
They don't need that much general purpose or floating point performance anymore, because it's reasonable for them to just fab their special purpose own stuff.
I'm sure they have some semi-specialized architectures made for doing their most important analysis tasks very quickly (which may or may not be fast in bitcoin-relevant ways), and for everything else, they have a smaller number of general purpose datacenters.
I can't think of any good reason the NSA would have 1-5 million x86-64 computers laying around, instead of 1-4 million more specialized processors, and a million x86-64.
- beering 14y agoWell, you have to have at least a few computers in order to store, index, and analyze all communications in the world, right? puts on tinfoil hat
- sneak 14y ago> Why would they? They don't need that much general purpose or floating point performance anymore, because it's reasonable for them to just fab their special purpose own stuff. Indeed, the NSA has their own fab.
- moconnor 14y agoWithout commenting directly on the fab, any codebreaking hardware could be expected to excel at computing bitcoin hashes. Perhaps more interesting is the codebreaking ability latent in the bitcoin network. That must be a tempting target for a variety of agencies around the world.
- rl3 14y ago> Indeed, the NSA has their own fab. They could always just fab architectures specialized in BTC mining, assuming their fabrication capacity is of sufficient scale to get the job done. I'm sure anyone who successfully performed a hostile takeover of BTC could find ways to recoup the initial investment in hardware. That said, I would assume the majority of their fab output and available computing power is busy with more important things.
- gwern 14y agoYes, they have their own fab, but I looked up what I could a while back on that fab ( http://www.gwern.net/Slowing%20Moore%27s%20Law#fn23 http://www.gwern.net/Slowing%20Moore%27s%20Law#fn23 ), and everything indicates the NSA chipfab is outdated, using very large-size features, and essentially meant for legacy system support - making old hard to replace chips, perhaps. If they're doing anything fancy, it'll be at partner chip fabs... but so many chip fabs have already left the USA that one wonders.
- michaelt 14y agoI'd imagine the economics of variable / nonrecurring engineering costs are different at that scale. Their tech might only let them make 200MHz chips instead of 2GHz, but couldn't they just make ten times as many of them? It's been a long time since CPUs have seen much of a bump in clock rates, after all - nowerdays it's all about smaller features allowing more cores/cache in the same space. If you don't care much about space as your process costs are dominated by setup rather than wafer costs, wouldn't you just use more space?
- iso8859-1 14y agoHeat dissipation is a limiting factor and larger chips require higher voltages. Which would require a even lower clock rate.
- michaelt 14y agoLet's say the NSA's chip fab is 10 years out of date - the same technologies as an Athlon XP "3000+" 2.1 GHz CPU with a 70 watt TDP. Worked just fine in a lot of consumer desktop PCs. If they needed a lower TDP, couldn't they just drop the clock rate and use wider features for lower gate leakage?
- DannyBee 14y agoAgreed. It's also likely the case that in practice, their TDP doesn't matter. Yes, they have a budget, but somehow i think their main problem is more finding sources of electricity and cooling, not paying for it.
- moconnor 14y agoThe article's comparison is invalid - in HPC computing speed is as much about the network infrastructure as it is the raw number of cores; simulations and graph-processing require interaction between the processing elements. The scores on the Top500 reflect this. In an embarrassingly-parallel application such as bitcoin mining, their peak performance is far higher. In fact, the theoretical peak of Titan's 19k GPUs would be around 90 single-precision petaflops, comfortably higher than the estimated peak of 2 million reasonably recent x86 processors in Google's data centers (unlikely to be top-of-the-range number crunchers). I've worked in HPC with a variety of actors for over a decade; I am in no doubt that classified machines exist with more power than Titan and Sequoia (#1, #2), which together make up most of the computing power in the top500 (it follows a power distribution, appropriately). An exaflop is still a really, really big number though. I can hazard a few guesses at non-public machines in the US that would reach or beat Titan. Perhaps the US Govt commands an exaflop of power spread amongst several agencies, but I wouldn't place any bets on it. That the bitcoin mining network has reached this scale is both astounding and depressing. That's an awful lot of computing power going to waste.
- kzrdude 14y agoIt's in the design of bitcoin -- its network power must grow enough to defend against attacks on the increasingly lucurative bitcoin market. The bigger the market, the more expensive attacks are worthwhile. Conversely, the hardware driving bitcoin could be part of its valuation.
- justincormack 14y agoHow could wasted hardware be part of a valuation?
- kzrdude 14y agoThe hardware working in the bitcoin network is it's capital backing. The more hardware backs it, the more resilient bitcoin is to attacks. If we valued one bitcoin at $1 million today, we would have a problem, because the hardware required to attack the network would be much cheaper than the potential gains of an attack.