Intel Demos Chip To Compute With Encrypted Data (ieee.org)
- Reference: 0180948466
- News link: https://it.slashdot.org/story/26/03/10/2022201/intel-demos-chip-to-compute-with-encrypted-data
- Source link: https://spectrum.ieee.org/fhe-intel
> Worried that your latest ask to a cloud-based AI reveals a bit too much about you? Want to know your genetic risk of disease without revealing it to the services that compute the answer? There is a way to do computing on encrypted data without ever having it decrypted. It's called fully homomorphic encryption, or FHE. But there's a rather large catch. It can take thousands -- even tens of thousands -- of times longer to compute on today's CPUs and GPUs than simply working with the decrypted data. So universities, startups, and at least one processor giant have been working on specialized chips that could close that gap. Last month at the IEEE International Solid-State Circuits Conference (ISSCC) in San Francisco, Intel demonstrated its answer, Heracles, which [1]sped up FHE computing tasks as much as 5,000-fold compared to a top-of the-line Intel server CPU.
>
> Startups are racing to beat Intel and each other to commercialization. But Sanu Mathew, who leads security circuits research at Intel, believes the CPU giant has a big lead, because its chip can do more computing than any other FHE accelerator yet built. "Heracles is the first hardware that works at scale," he says. The scale is measurable both physically and in compute performance. While other FHE research chips have been in the range of 10 square millimeters or less, Heracles is about 20 times that size and is built using Intel's most advanced, 3-nanometer FinFET technology. And it's flanked inside a liquid-cooled package by two 24-gigabyte high-bandwidth memory chips—a configuration usually seen only in GPUs for training AI.
>
> In terms of scaling compute performance, Heracles showed muscle in live demonstrations at ISSCC. At its heart the demo was a simple private query to a secure server. It simulated a request by a voter to make sure that her ballot had been registered correctly. The state, in this case, has an encrypted database of voters and their votes. To maintain her privacy, the voter would not want to have her ballot information decrypted at any point; so using FHE, she encrypts her ID and vote and sends it to the government database. There, without decrypting it, the system determines if it is a match and returns an encrypted answer, which she then decrypts on her side. On an Intel Xeon server CPU, the process took 15 milliseconds. Heracles did it in 14 microseconds. While that difference isn't something a single human would notice, verifying 100 million voter ballots adds up to more than 17 days of CPU work versus a mere 23 minutes on Heracles.
[1] https://spectrum.ieee.org/fhe-intel
I'm worried about the DRM potential of this (Score:2)
Now imagine this isn't your workloads running on a cloud server, but software you paid for running on your own computer.
Working DRM would be the worst invention in the history of computing, and specialized hardware that can hide things from a root user makes it much more plausible.
LOL, Intel? (Score:2)
Yeah, that's gonna happen.
I smell BS (Score:1)
How do you work on an encrypted set of data without decrypting it? This sounds like absolute BS and coming from Intel it probably is with their recent track record. I mean if a processor can work on encrypted data without decrypting it and knowing what the data actually is, what's to stop someone from writing code making that processor spit out all the plain text of that encrypted data?
Re: (Score:2)
I think it's real. This isn't the first time I've heard about it. So far it hasn't been practical. If this one actually works, trusting it is going to depend on trusting the implementation, not on understanding or auditing it.
Re: (Score:2)
It is real: [1]https://en.wikipedia.org/wiki/... [wikipedia.org]
But it is exceptionally slow. This is something you only do when you have no choice.
[1] https://en.wikipedia.org/wiki/Homomorphic_encryption
Re: (Score:2)
Ok, but I still think it's BS that Intel has anything close to working at a decent performance level in this field. If anything I would consider this field to be up Nvidia's alley and if they don't have anything going on here, I have few hopes for anything Intel has to be anything less than smoke and mirrors.
Here's a better idea (Score:4, Informative)
Don't use the cloud
The cloud is a trap
Run away
Will still be dog-slow... (Score:2)
For that reason, this is hardware that has very limited use.
Cannot trust (Score:2)
How can the end user trust such a system? Even if we ever find a way to ensure it runs the software it is supposed to run (does it really encrypts your data?), the inner working involves concepts that even an IT engineer do not master.
Re: (Score:2)
the end user delivers encrypted data and decrypts the result. if you trust the encryption method then you can trust the output.
now, i have no idea how this system in particular works and what its capabilities or intended use are (didn't rtfa), but for specific operations and given specific assumptions this is entirely possible.
Re: (Score:3)
Compute on encrypted data is more of a theoretical exercise (I know, I know, I mean "not particularly useful") than a practical one. The limitations are so many that it can hardly be called processing. You can't make decisions on data that is encrypted, because then you could figure out what the data is (think 20 questions). You can only do some limited math on specific scenarios.
It's interesting, but it can't really process your health data or much of any real world data imho.
Re:Cannot trust (Score:4, Informative)
[1]Fully homomorphic encryption [quarkslab.com] is mostly theoretical, but that's because it is incredibly slow and uses huge amounts of memory, not because you can't write conditionals.
You can compute anything using FHE that you can with any other turing machine. As long as you can [2]wait long enough [arxiv.org].
If Intel can provide 1000x+ speedups, some of this might become usable in limited ways. Because right now it costs multiple seconds to do a single FHE multiply, and it needs something like 20000x the memory space of unencrypted computation.
[1] https://blog.quarkslab.com/a-brief-survey-of-fully-homomorphic-encryption-computing-on-encrypted-data.html
[2] https://arxiv.org/pdf/2303.10877
Re: (Score:2)
> [1]Fully homomorphic encryption [quarkslab.com] is mostly theoretical, but that's because it is incredibly slow and uses huge amounts of memory, not because you can't write conditionals.
> You can compute anything using FHE that you can with any other turing machine. As long as you can [2]wait long enough [arxiv.org].
For certain (3 letter?) agencies, knowing that the data always stays encrypted may be worth the trade-off of slow. Not for all situations, of course, but for very specific targeted uses.
[1] https://blog.quarkslab.com/a-brief-survey-of-fully-homomorphic-encryption-computing-on-encrypted-data.html
[2] https://arxiv.org/pdf/2303.10877
Re: (Score:3)
You don't need to trust the system, that's the beauty of FHE.
You encrypt the data and send it to the FHE chip presumably in the cloud along with your code. It crunches the code and the output is still encrypted because it doesn't have the keys. You get it back and decrypt it.
Re: (Score:2)
Easy. The government will mandate them and we will let them do it because we're going to be busy worrying about trans girls in sports or satanists or violent video games or whatever shiny thing they dangle in front of us this time.
You won't have a choice but to trust it because it'll be rammed down your throat and you'll just get used to it because you won't have a choice and the alternatives would be to stop using computers.
Once a voting public gives up critical thinking you're pretty well and trul