Nokia SL3 with CPLD idea

[blagus]

I have 2 spare CPLDs, originally used for XBox 360 RGH modification.
Both are CoolRunner-II, one XC2C64 (Mtx Infinity 2), and other one XC2C128 (TX DGX v1.1). They have JTAG port and can be reprogrammed without problems.

I saw some ideas about using FPGA or ASIC for SHA1 BF, which reminded me of these two I got. Technically, it is possible to program it to calculate SHA1 hash, but how fast it would be? Did anyone try, to share thoughts and experience?

BR,
Blagus

[blagus]

I've just made some quick calculations, based on ighashgpu table, bitcoin FPGA experience and simple maths.

Nokia SL3 has a total of 10^15 combinations.

1) When compared to some of the fastest bitcoin FPGA setups, which perform more complex maths - double SHA-256 compared to our SHA1:
Let's say that FPGA has the speed of 60G/sec, at double SHA-256 hash bruteforcing.

60G = 60*10^9/sec

The time required to cycle out every possible SL3 combination:
10^15 / (60*10^9) = 16666,67 sec = 277,77 min = 4,629 h = 4h 37min

I won't get into how faster it would be with SHA1, these numbers are for double SHA-256.

2) Now, let's take a decent, low end PC (for a bruteforcer), for example, overclocked 4870, which would give ~1G/sec:
1000M = 1G = 10^9/sec

Time = 10^15 / 10^9 = 10^6 sec = 16666,67 min = 277,77 h = 11,57 days

3) Goal would be to get close to BF-freaks on this forum who get 10G+/sec with 3x or 4xHD7970.

10 000M = 10G = 10*10^9/sec

Time = 10^15 / (10*10^9) = 100k sec = 1666,67 min = 27,7 h = 1,16 days

Even better - double speed, half time:
20G= 20*10^9/sec

Time= 10^15 / (20*10^9) = 50k sec = 833,33 min = 13,89h = 13h 53min.


Keep in mind the extra low power consumption of CPLD - a couple of watts compared to several hundred watts of a regular PC.

[online-nokia]

SL3 isn't SHA-1 mate

[wgmmmx1]

Originally Posted by online-nokia

SL3 isn't SHA-1 mate

If you not know what you write then better no answer other users this forum.
SL3 is SHA-1 algorytm 100% sure.

[blagus]

Originally Posted by online-nokia

SL3 isn't SHA-1 mate

So, we all have been fooled by the description of SL3 hash being SHA1($pass.$salt). All of the codes calculated and entered to phone are actually fake. This all is just a glitch in the Matrix!

Back to the topic, I think I'll just write some code that will SHA1 all 10^15 combinations and blink the LED on each 10G combos processed just for the performance test.

I took a look at the SL3 FPGA code, it isn't complex at all - store the hash and salt into a few registers, then go through all combos, append salt and compare to the original hash. If it matches, send it to PC via JTAG.

[Advance-Box]

@blagus

I did this some years ago using virtex 4 and realized that it is still cheaper to just use video cards gpu. Also making PCB for BGA with 400+ pins is a nightmare.

You cannot even fit 1 sha-1 core in your coolrunner cplds, they are too "small" and does not have enough macrocells.

The only way to beat the GPUs is to make custom SL3 asic.

[blagus]

Originally Posted by Advance-Box

@blagus

I did this some years ago using virtex 4 and realized that it is still cheaper to just use video cards gpu. Also making PCB for BGA with 400+ pins is a nightmare.

You cannot even fit 1 sha-1 core in your coolrunner cplds, they are too "small" and does not have enough macrocells.

The only way to beat the GPUs is to make custom SL3 asic.

Thanks a lot for the info.

The PCB itself isn't a problem, XBox360 RGH solution is on a small board, QFP package, with JTAG pin-header.

But, you say that it's too small for SHA1 code, so I'll forget about CPLD.

Can you give me some details about how much macrocells/memory should a CPLD/FPGA have for SL3 calculation? I've actually just started learning about these devices, and SHA1 calculation isn't really a complex task, so I'd like to get something capable of doing it.
I have just two SL3 phones that are just fine now, sim-locked, but if I could unlock them and learn something along, it would be nice.

Of course, I wouldn't ruin (your) SL3 market, something with 2-3G/s (FPGA speed I found on this forum) can't catch up with 4x7970 *******s calculating >14G/s.

[..::SuB::..]

Application specific chip is the best option for you only need a lot of money and some people with extraordinary knowledge based on mathematics and crypto. Try to google for litecoin or bitcoin asics/fpga research topics. We can't compare LTC scrypt or sha256 with our SHA1 problem but you can clearly see how hard to develop an application specific device.

[jodge]

Originally Posted by blagus

Thanks a lot for the info.

The PCB itself isn't a problem, XBox360 RGH solution is on a small board, QFP package, with JTAG pin-header.

But, you say that it's too small for SHA1 code, so I'll forget about CPLD.

Can you give me some details about how much macrocells/memory should a CPLD/FPGA have for SL3 calculation? I've actually just started learning about these devices, and SHA1 calculation isn't really a complex task, so I'd like to get something capable of doing it.
I have just two SL3 phones that are just fine now, sim-locked, but if I could unlock them and learn something along, it would be nice.

Of course, I wouldn't ruin (your) SL3 market, something with 2-3G/s (FPGA speed I found on this forum) can't catch up with 4x7970 *******s calculating >14G/s.

maybe try with this board, i think it is easier when you only have to focus to the algo not both of them.
https://en.bitcoin.it/wiki/ZTEX_FPGA...Bitcoin_Mining

ohh and almost forgot the ******* is this board not the 4x7970
https://en.bitcoin.it/wiki/Xtreme_Miners_Lion

[Bph&co]

Originally Posted by blagus

Technically, it is possible to program it to calculate SHA1 hash, but how fast it would be? Did anyone try, to share thoughts and experience?

BR,
Blagus

Hi,

USB-FPGA Module 1.15x: Spartan 6 LX150 FPGA Board with USB 2.0 Microcontroller

https://github.com/fpgaminer/sha1_collider

Hope this helps.

BR

[jodge]

i know the coin mining is more profitable than sl3 but what about the 20digit calculation time with FPGA(1-2TH/s)?

[blagus]

Thanks for all the input guys.
I took a look at Spartan 6 SHA1 code, and it doesn't fit my pocket - but I won't quit, if I manage to do something on some cheap FPGA board I'll let you know

Originally Posted by jodge

i know the coin mining is more profitable than sl3 but what about the 20digit calculation time with FPGA(1-2TH/s)?

Simple math: 10^20 / 10^12 = 10^8 seconds = 1,6M minutes = 27,7k hours = 1157 days = too much time.

10TH would be a bit better as it gives you 115,7 days but it's still too much.
To make it reasonable you'd need something capable of 100TH/s to get 11,57 days, and the only thing that could do it would be some very optimized ASIC design.

[Brka]

Originally Posted by blagus

Thanks for all the input guys.
I took a look at Spartan 6 SHA1 code, and it doesn't fit my pocket - but I won't quit, if I manage to do something on some cheap FPGA board I'll let you know



Simple math: 10^20 / 10^12 = 10^8 seconds = 1,6M minutes = 27,7k hours = 1157 days = too much time.

10TH would be a bit better as it gives you 115,7 days but it's still too much.
To make it reasonable you'd need something capable of 100TH/s to get 11,57 days, and the only thing that could do it would be some very optimized ASIC design.

I think 10 x ZTEX 1.15y boards would be just fine for SL3
... if somebody manage to write some code for it + some Raspberry PI or something, as controller for them ...

And about those "TeraHash" values ... don't dream too much.
In fact, sl3 is not much different than sha-256/double round,
but would require customized ASIC, none of existing cant be used.


b.r.
Alex

[haloon]

I am already managing 22400Mh/s on 10x 1.15y. So nothing special if you look for the price of those boards.

[Brka]

Originally Posted by haloon

I am already managing 22400Mh/s on 10x 1.15y. So nothing special if you look for the price of those boards.

are you talking about SL3 on ZTEX boards ?