Affordable Encryption Accelerators

Martyr

does anyone know where to get these?
looked at one place so far, soekris.com

i was looking at the possibility of getting this at some point in future.

Encryption, 128/192/256 AES, DES, 3-DES and RC4 at 420 to 920 Mbps
Authentication, SHA-1 and MD5 at 650 to 720 Mbps

this is just a prototype, but it sounds pretty good.

the little brother is half the performance, around $80 dollars each and supported by most unix-variant operating systems.(obsd/fbsd/linux..etc)

i'm guessing these could be used to crack encryption too, anyone know if they can be used for that purpose?
or for cracking purposes, would it be better to use FPGAs like from Xilinx??

edit:just found article using xilinx fpgas for sha-1 at 2.5Gbps, not sure what they used though

Capt'n Midnight

Couldn't you just use a GPU ?

Martyr

i looked into this last year, and apparently there is only a small increase in performance of block ciphers because of the manipulation of bytes needed, and gpus are more suitable to parallel operations.
so, probably serpent,des,des3 and other bitslice implementations would work very well.(i wouldn't be an expert on it now)

perhaps there is someone that has improved on this work since though (must look again)
and i was told by a games programmer that dx10 cards may be more efficient for encryption.

i'm not sure about hash algorithms (most are designed for 32-bit processors), but the paper i mentioned (which i can't find a copy of) claims to produce 2.5 gbps on a Xilinx FPGA, and obviously some precomputations are happening.

i've seen implementations of some hash algorithms like md5 and sha-1 using MMX/XMM, for what seems like a good idea.
2 or 4 hashes at once are obviously better than one, of course...but mmx/xmm doesn't natively support bit rotations, which is where it really slows down.

its actually better to use 32-bit code in some situations, you can create 2 hashes at once, eliminating dependancies.

if it were possible to pre-compute some values, and overall eliminate bit-rotations, then a parallel algorithm using xmm would be very fast.
how to do this, i don't know, but maybe the article (which i can't find) shows.

i haven't read this yet, but might be worth a look

Martyr

obviously somebody here on the boards must have read Cracking WIFI faster

good read about crypto on hardware, another site is openciphers.org
sorry if i can't keep up with events..:p