0.7 BETA - performance numbers

[mrCracker]

Windows Vista X64, Intel [email protected], 8Gb memory, single 9800GTX (128 cores) @1.69Ghz, running 0.7BETA with these parameters:
CUDA-Multiforcer-0.7.exe -h NTLM -f ntlm.txt --min=1 --max=7 -c charsets\charsetloweruppernumeric

612 unique hashes gives me these results (length 1-6 is *fast*, so i don't include those lines..)

Launching kernel for password length 7
Done: 100.00% Step rate: 153.7M/s Search rate: 94059.9M/sec
------------------------------------------
Compute done: Reference time 23325.4 seconds
Stepping rate: 151.0M MD4/s
Search rate: 92398.2M NTLM/s

That was my very first test with 0.7Beta, more to come. I guess i will be upgrading to GTX260 in a few weeks tops.

[Bitweasil]

You should have significantly better numbers (at the expense of UI responsiveness) with the following:

-m 500 -threads 256 -blocks 256

Vista has a VERY long kernel launch time (compared to Linux/XP), so longer kernels benefit Vista a lot more. See if those make a difference.

[mrCracker]

*MUCH* better performance:
Step rate: 416.9M/s Search rate: 255168.3M/sec (at length 7)

Hmmm... Feature request: discover and suggest optimal settings based on my setup (-check option?). With my config here the default gives a reasonable performance, optimized settings gives *WAY* better performance. I want the "way better" to be the default.

[Bitweasil]

The 0.6 version has performance tuning.

The 0.7 version doesn't have that working quite yet. I need a way to find the maximum thread count without manually setting things. I suppose I could just keep incrementing until I get a launch failure. It's a different way of doing things. I'd like to see if I can get it working again, as this is highly dependent on the right settings, especially for longer kernel lengths.

I have a Windows dev environment now, so I'll play with it as I have time.

[mrCracker]

I just ordered 2 Gainward GTX260 896Mb overclocked cards, should improve my performance a bit i guess over my existing single 9800GTX card.

Suggestions for optimal parameters highly appreciated - no, i don't care if my display goes sluggish as long as CTRL+C can kill it if it takes too long time.

Regards,
mrCracker

[Bitweasil]

Parameters for all the GTX series cards are the same.

-m 500 --threads 512 --blocks 512

If you get an error about insufficient resources, drop the thread count down (try 384 or 256).

[mrCracker]

Performance using stock GTX260 card (single) on 1534 NTLM hashes:

Launching kernel for password length 6
Done: 99.41% Step rate: 428.8M/s Search rate: 666281.1M/sec

------------------------------------------

Compute done: Reference time 135.9 seconds
Stepping rate: 418.0M MD4/s
Search rate: 649546.4M NTLM/s

[MaddGamer]

Windows Vista x64 with Intel [email protected] Ghz 4G ram; Quatro FX 4800;
CUDA-Multiforcer-0.7.exe -h SHA1 -f mypas.txt --min=1 --max=12 -c charsets\charsetfull -m 1000 --threads 512 --blocks 512

mypas.txt has 886 unique hashs
Launching kernel for password length 7
Done: 0.56% Step rate: 126.7M/s Search rate: 188946.6M/sec

[zJ0nk]

Win XP 32bit
i7
Tesla s1070

CUDA-Multiforcer-0.7.exe -h NTLM -f ..\6charNTLM.txt -o ..\6charNTLMoutput.txt --min=1
--max=5 -c charsets\charsetfull -m 1000 --threads 512 --blocks 512

...

Hash type: NTLM
CUDA Device Information:
Device 0: "Tesla C1060"
Number of cores: 240
Clock rate: 1.44 GHz
Single charset loaded.
Loading & sorting hashes. This may take a while.
Returning from sort.
Hashes loaded (2 hashes)
You may want to consider the FAST[hash] option if it exists.
Done with copy to constant

...

Launching kernel for password length 5
Done: 25.27% Step rate: 746.3M/s Search rate: 1492.7M/sec

[blazer]

wow impressive a tesla, how much is the one you're using may i ask? it seems similar to a GTX295 but slightly higher shader clock.