some questions about the future

[1epi]

What success rate should your HUGE rainbow tables have ?
I'd go for 99% (miss 1 in 100), project-rainbowcrack has opted for 96.8 % (miss 1 in 31.25), and that 96% looks bad in the eye of a potential buyer.

I think the Bruteforce point (bfpoint) should be between 1000 and 10000, but it depends on the desired chainLen.

3 rainbow tables for total success rates under 99.5%, 4 rainbow tables for >99.9%.
It's kind of blurry, this area, because if you have a fixed success rate and bfpoint, and you create a 3 rainbow table set, it's smaller in size than a 4 rainbow table set, < 5. On the other hand it takes more to generate. I can give you an example if you want to find out exactly the tradeoff between them. For large keyspaces i suggest we create sets of 5 tables in order to generate them faster. But if you have time and CUDA enabled devices I'd say we should go for the best configurations : less rainbow tables / set.

[1epi]

give me a average speed for one cuda enabled device @100k @200k @300k @400k chainLen. (md5)
and the compression ratio of your RT format.

[Bitweasil]

Rates/sizes:

For MD5, figure 400-450M links/sec per GT200 core.

Table size: 32 bytes per chain, plus 8192 bytes of header (per table).

//EDIT: Fixed chain size... whoops.

[1epi]

Rates/sizes:

For MD5, figure 400-450M links/sec per GT200 core.

Table size: 16 bytes per chain, plus 8192 bytes of header (per table).

400-450M links/sec for any chainLen ? That's not possible.
quick question : GT200 core = one nvidia GT200 video card ?

[Sc00bz]

Table size: 16 bytes per chain, plus 8192 bytes of header (per table).

Isn't it 8192 bytes plus 32 bytes per chain.

400-450M links/sec for any chainLen ? That's not possible.
quick question : GT200 core = one nvidia GT200 video card ?

That's in links per second not chains per second.
There's a table of speeds on this page
http://www.cryptohaze.com/gpurainbowcracker.php

Code: Select all

GPU                         | SPs | Frequency | links/s | 100k regen | 200k regen
----------------------------+-----+-----------+---------+------------+------------
nVidia GTX295 (single core) | 240 |  1.24 GHz |    436M |      11.5s |      45.9s
nVidia GTX260               | 216 |  1.24 GHz |    394M |      12.7s |      50.7s
nVidia 8800 GTX OC          | 128 |  1.46 GHz |    272M |      18.4s |      73.5s

Regen is the prework time as you can see it is 4 times longer with a chain length twice as long.

[Bitweasil]

400-450M links/sec for any chainLen ? That's not possible.
quick question : GT200 core = one nvidia GT200 video card ?

Links per second, as Sc00bz said. Divide by chain length to get chains per second.

GT200 core = 1 GT200 core. The GTX200 series cards are mostly single core, except for the 295s, which have 2 GT200 cores on them.

Isn't it 8192 bytes plus 32 bytes per chain.

Regen is the prework time as you can see it is 4 times longer with a chain length twice as long.

Yeah... fixed that. Sorry, was thinking of the hash field and forgot about the password field.

Regen... roughly scales with the square of the chain length, though that's not 100% accurate. The table search time should be linear with chain length.