Bitcoin Forum

An ASIC miner in every PC

http://software.intel.com/en-us/articles/intel-sha-extensions

...so CPU mining makes an unexpected comeback.

Is anyone who is capable of making an educated guess about how the hash/watt values Intel could achieve with this as compared to current ASICs, and who is willing to be quoted, is invited to send me a PM.

It seems like this was talked about some time ago and it wasn't going to work well for Bitcoin mining?!  But maybe I'm not remembering correctly?

It probably isn't very good. A single SHA-256 operation takes 139 instructions; 2 would take at least 278. Peak instruction rate is 177730 MIPS for the best processors, and this won't even approach the peak speed. So the speed would be far less than 600 MHz, which is hardly any better than a high-end GPU.

Is that per processor, or per core?

Per processor. Per core it's probably closer to 30000 MIPS and less than 100 Mhps.

So a CPU containing these instructions might produce about 6 Mhps/watt?

Probably far less, but yes, that's the upper bound. However, it's worth mentioning that CPUs are useful for non-mining-related activities, and so suffer less from depreciation than ASICs and GPUs.

Yeah specialized instructions help but a general purpose CPU is a giant bloated monster of transistors 99% of which are never used in hashing (massive banks of cache, out of order execution pipeline, branch prediction, high speed memory controller, high speed point to point connectivity to northbridge, floating point computations ,etc).  Nothing can make it cost effective compared even to a GPU much less dedicated ASICs.

Had these instruction been included in Sandy Bridge processors it would have been a big deal.

From what I've heard the target application for this is IPsec portion of the IP stack, together with the already existing AES instructions. In all the operating systems. Apparently Broadcom and/or Marvell had made some inroads into the Intel network chipset territory by doing IPsec at much lower watts per megabit per second.

It obviously doesn't compare to ASIC mining. I mean something like a KnC where you have, possibly, more transistors dedicated to SHA then an entire Intel CPU.

But an Intel CPU cranking out as many Mhash as a GPU is pretty damn impressive. If it weren't for Asics, you might very well see people build CPU rigs with cheap motherboards rather getting expensive motherboards and tons of GPUs.

LOL....

Like first time sex... lots of excited anticipation, but once you get round to implementation its all over very quickly followed by  depression.

sounds like padlock reloaded  ;)

At least the NSA will have the ability to get all your private keys now. 

Basically that - in a short while, we will finally see MS Outlook with inbuilt encryption - using, of course, the new Intel CPU optimised encryption.

You guys are idiots.  Since when is SHA-256 an encryption algorithm?

AES is, and Intel has recently optimized that.

code release SHA-256 processor intel https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=22357

Dree,


good insight and update !  ;)

According to heise.de the new SHA256 instuctions will be introduced with Nights Landing (Xeon Phi) in 2014, and later with Skylake (follow up of Broadwell, which is the follow up of Haswell) - problably in 2015.

I am not the most computer/crypto knowldgable person, but does this mean that someone has to write some code to allow people to run asics on intel processors? Will there a time in the future I can use my old intel Core i5 laptop for mining btc? (I know I can now but it is a huge waste of energy).

Yes.


No.  While this would make a CPU "more" efficient, it wouldn't make it efficient enough.  Say your CPU was 3x (and it likely wouldn't be that much of a speed up) as fast and used the same power.  Would it really matter?

i'm no expert on this but could cpu support for hashing also be used for the integration of encryption directly into operating systems/ software instead of mining?

i just had this strange idea of my cpus doing encryption of mails/chats etc. as a background process on-the-fly
or is this just a chimera and my thought approach is way off?  :P


 

Your CPU can do that right now.  These instructions just improve the speed/efficiency of various encryption algorithms.  They don't make possible something that was impossible before.

In fact, your CPU is doing that right now. The very act of visiting these forums, posting, etc., already involves this encryption.

thanks for your answers.
so with improving the efficiency/ speed of these operations new cpus are capable of doing more complex encryption processes in the background.
i try to reason what this could mean for the future development of encryption of communication/information especially with Prism/Tempora news in mind.

Not really.  SHA-256 isn't particuarly computationally intensive.  The average CPU is capable of tens of millions of hashes per second even without this instruction.  Far more than most users could ever hope to need.

ah ok...what a pity, for a moment i hoped that could represent a glance at newer and safer encryption standards in the future.
so in the end it's more about independent network providers and in what form a software uses the algorithms to push progress in that direction.
nevermind...learnt something new again.
(i read a thread some months ago about the possibility of transporting information via the blockchain that's where my brooding came from)

That's pretty cool, although it's not too much use until it's massively parallelized. AMD should add asic SHA hashing to their GPU's. Then we talk.

AMD CPUs would also make things interesting, especially in the server market.

The highest amount of AMD cores in a chip for the desktop market is 8. In the server market, it is 16, and you can get multi-chip motherboards. 64 cores, all with onboard SHA256 hashing.

With that amount of power potentially out there, SHA256 had better be secure...

Matthew:out

No good. GPU's have thousands of cores. In order for it to be of any real use, it would have to be parallelized to the hundreds if not thousands.

It would be interesting to see AMD produce SHA256 or scrypt specific GPUs... I like to imagine they will, considering what steps they're taking to return to profitability. Shares of AMD trade around 3.50 for the last several months... association with bitcoin/cryptocurrency could be very good publicity for them and create another revenue stream.

These will go nicely with those new motherboards built just for mining  ::)


Really these companies missed the boat for our neck of the woods, they're going after something else...

Are you suggesting the Intel implementation is less efficient (in GH/W) than in dedicated ASICs, or just that they are less powerful per die? Because I think what matters is how this will affect centralisation. If the incremental power needed to run a mining process in the background is less expensive than the BTC it yields, then many people might run it. And since there are so many general purpose computers in the world, this might mean significant percentage of the world's hashing power might come from PCs eventually. Dedicated miners would still run dedicated ASICs, but CPU mining might no longer be irrelevant from a decentralisation point of view.

This really isn't about regular general-purpose CPU's suddenly becoming competitive for mining, but what it does open up is the possibility that in the future, a multitude of devices can be doing a tiny amount of mining in the background all the time.

A very plausible consumer application (that would really represent the holy grail in Bitcoin usability) would be a credit-card sized kinetically-powered lightweight node that functions as a hardware wallet signing transactions. SHA 256 hardware implementation in a CPU is not necessarily about mining, but could form the basis of a low-power signing implementation.

As far as Intel's motivation, it's kind of naive to assume it's Bitcoin specifically, because SHA-1 and SHA-2 are used in a tremendous amount of applications that have nothing to do with cryptocurrency.

The guys over at freebsd had some words..
http://arstechnica.com/security/2013/12/we-cannot-trust-intel-and-vias-chip-based-crypto-freebsd-developers-say/