Math processor

I need a processor as fast as possible to handle 2048 bits numbers. Yes, its huge. I really cant use a PIC for example; I'm lost... any other tip?

thanks a lot for all answers,

n.

--
Lucas Zimmerman
netbogus@gmail.com

2048 bit integers? Is there a 2048-bit floating-point format?

The obvious question is, why this many bits?

I sure hope you don't have to do any divides.

If you need serious number crunching at this word length, an FPGA is probably the way to go. Not a weekend project.

John

And the Flops (FP operations per second) and accuracy required are?

Pretty much any 8 to 32 bit processor can do 32-bit FP in software, given enough time. If you want it in silicon the choices get much slimmer and the packaging gets more challenging to deal with in prototype quantities.

Best regards, Spehro Pefhany

--
"it\'s the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

You really need to spend more time around engineers. Let me illustrate.

My dad's company

makes parts for street rods. A guy was building a cool custom deuce roadster with Masarati running gear. My dad built a one-off custom body with a floorpan to fit the chassis (and suicide doors, but that's not part of the story).

When we got done with the body the customer took it to the local high-end paint shop, which is owned by a guy who does really really good work and who is an absolute perfectionist. The customer says "make the paint _perfect_" and leaves for 2 weeks on business out of country.

The paint shop gets busy making the paint _perfect_ (to their definition). Customer gets back from out-of-country and is shocked to find them still working on getting the hand-built and kinda rough floorpan up to show-quality paint standards. Clarification of the definition of the word "perfect" ensues.

So don't say "as fast as possible" or you'll end up with something custom built, possibly that requires water cooling.

_Do_ say how fast you need to do arithmetic with 2048 bit numbers, and how many you need to store. A PIC _will_ do this, but perhaps not as fast as you'd like. A new supercomputer is one interpretation of "as fast as possible" but it may be a wee bit too big and too expensive.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Define "handle". How fast? How much money you got? What are your other parameters?

You could use a PIC, if you're other requirements aren't too strict. If you need performance, a 64bit processor will be faster than a 32bit processor. Perhaps an FPGA would be faster yet. We don't know what your requirements are though. What are you trying to do?

The better the question, the better the answer.

--
  Keith

That's an unduly optimistic position to take.

Best regards, Spehro Pefhany

--
"it\'s the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

It depends a lot on how fast you want to do it and how many floating point numbers you want to keep around.

You can do floating point math on a PIC if you don't mind it taking a while. If you buy a cheap PC you get a processor that does floating point in a box with a power supply for about $300US. For some projects this may be the lowest total cost way to go.

--
--
kensmith@rahul.net   forging knowledge

256 bytes doesn't sound all that huge. "As fast as possible" is pretty vague.

What are you trying to make, some kind of concealed gambling/cheating device?

Have a look at ARM processors-- they easily have enough RAM, you can use the GCC toolchain etc.

Best regards, Spehro Pefhany

--
"it\'s the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

Key question is what speed you need. Just about _any_ processor can do floating point arithmetic, with a suitable library, but the performance will be relatively poor. Even quite a basic processor like a PIC, can manage several thousand basic arithmetic operations, dropping to a few hundred complex operations (like a sin) per second. If you need to perform hundreds of thousands of operations per second, then you need to be looking at hardware floating point units. The questions then are how much data the processor needs to address, what speed the arithmetic needs to achieve, and what other I/O is needed.

Best Wishes

I'm replying my own post trying to answer all at once; sorry if it's confusing.

So, let me make myself a little clearer; I understand I was vague. So, here is the thing... I'm finishing my graduation in computer engineering and I'll start next year a circuit to encrypt data. This is my final project. I choosed to implement RSA algorithm (which has operations like mod, -,+, *, ^, etc). I would like to work with keys from 640-2048 bits long. I was definetly wrong asking for floating point processors, by the way. I have just noticed that. So, like I said before the project actually starts next year. I'm still considering all stuff; even the algorithm... I came here trying to get general information about possible designs.

Once again, thank you all for all help I really appreciate.

my best regards,

lz.

2048? Do you want to do public key cryptography? Iin which case a DSP or a dedicated coprocessor may come in handy.

Wim

schreef in bericht news: snipped-for-privacy@f14g2000cwb.googlegroups.com...

I've never seen a processor that didn't do it's things as fast as possible.

--
Thanks, Frank.
(remove \'q\' and \'.invalid\' when replying by email)

There are CISC* machines, there are RISC** machines. Then there's the CDP1802, the only NISC*** machine I know of.

• Complex Instruction Set Computer

** Reduced Instruction Set Computer *** Never had an Instruction Set Computer

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

What public key cryptography uses floating points ?

Sylvain

Microsoft's?

martin

In article , Frank Bemelman wrote: [...]

Did you ever see the CDP1802?

I think someone went home chuckling after coming up with that instruction set.

--
--
kensmith@rahul.net   forging knowledge

There is an RSA library for the dsPIC that can do 2048bit RSA encryption:

It can do an 1024bit RSA encryption in 7ms. Probably costs a bit though, and I'm guesing you don't want an off the shelf solution, although that would make your project easy :->

You could do the same on any 8bit micro like PIC or AVR, but it won't be nearly as fast.

Perhaps you can slant your project towards something low end like "How to do RSA encrytion on a $0.50 8pin PIC micro in under 2KB". That might be more achievable than say "The worlds fastest 2048bit RSA encryption algorithm" implemented in a bleeding edge FPGA. There are plenty of people out there designing super fast encryption algorithms, so it might be hard to do something unique in that respect.

Dave :)

This was a case of spotting the right question. 2048 bits immediately suggests RSA

None that I know of.

I am not aware of modular operations for floating point. Do public key algorithms without modular operations exist?

Wim

Ok, how about; A better question will get better answers.

--
  Keith