oclHashcat running on the new AMD hd7970
#1
Welcome!

The first thing I noticed, one could call it "a problem", is, that there is no linux driver available.

So, instead of worrying I directly went to my Win7/64 box and attached it on there.

After that, I installed the "amd_radeon_hd7900_win7_64.exe" special driver from AMD page. Installing the driver worked without any problems (had catalyst 11.12 installed before).

In the following steps I am using oclHashcat-lite v0.09 beta since oclHashcat-lite v0.08 does not have a "Tahiti" kernel. This new beta include some unpublished code optimization for MD5 which is around 2%. Here is what I did:

Quote:oclHashcat-lite64.exe -n 800 --gpu-loops 1024 -1 ?l?d?s?u --gpu-watchdog 0 --outfile-watch 0 --restore-timer 0 --runtime 28 --pw-min 8 -m 0 aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa ?1?1?1?1?1?1?1?1?1

This is a "Full-Power" command, its perfect for benchmarking.

Here is the output:

Quote:oclHashcat-lite v0.9 by atom starting...

GPU-Loops: 128
GPU-Accel: 800
Password lengths range: 8 - 55
Platform: AMD compatible platform found
Watchdog: Temperature limit disabled
Device #1: Tahiti, 3072MB, 0Mhz, 32MCU
[s]tatus [p]ause [r]esume [q]uit => s
Status.......: Running
Hash.Target..: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
Hash.Type....: MD5
Time.Running.: 20 secs
Time.Left....: 9 days, 9 hours
Plain.Mask...: ?1?1?1?1?1?1?1?1
Plain.Text...: ***aaaaa
Plain.Length.: 8
Progress.....: 163158425600/6634204312890625 (0.00%)
Speed.GPU.#1.: 8156.5M/s

Started: Wed Jan 11 19:23:13 2012
Stopped: Wed Jan 11 19:23:35 2012

MD5: 8156.6M/s, on a single GPU. Now that is great!

The funny thing is, even on extreme highly loaded workload (-n 800), which usually freezes the desktop, on this card the desktop was still useable. I do smell optimization potential on here.

Before I continue with the other algorithms, lets compare what we see to the expected theoretical results. For this, I will use my stock clocked hd7970 vs. stock clocked hd5970 (1 GPU only!):

hd5970: 1600sp * 725mhz (1160000) vs. 2048sp * 925mhz (1894400). This is a "theoretical" improvement of 63%.

MD5: hd5970 = 4940M/s, hd7970 = 8156M/s. This is a "real-life" improvement of 65%.

That means, the theoretical values perfectly fit the real-life values.

Ok, another test. This time SHA1. Same command, just switching to -m 100:

Quote:Speed.GPU.#1.: 2807.5M/s

SHA1: hd5970 = 1690, hd7970 = 2807. This is a "real-life" improvement of 66%.

Hits again the theoretical expectation. Awesome, I am starting to like this card.

Here are some more numbers without comments of some algos I think many people are interessted in:

Quote:vBulletin3: 2214M/s
vBulletin30: 1534M/s
NTLM: 15813M/s
SHA256: 1100M/s
SL3: 2825M/s

Now switching to oclHashcat-plus. I had to run this again on a beta version since the current 0.07 version did not work.

First thing to test, of course, WPA/WPA2:

Quote:Speed........: 127.9k c/s Real, 128.2k c/s GPU

This is just a single GPU card, the hd7990 will surely make 220k c/s

OK, the comparisation again:

WPA/WPA2: hd5970 = 79k c/s, hd7970 = 128k c/s. This is a "real-life" improvement of 62%.

Another thing that many people want to know is the multihash speed.

Quote:1 hash (not reversed): 5182M
2 hashes: 4550M
100 hashes: 4520M
1000 hashes: 4444M
10000 hashes: 4400M
100000 hashes: 4127M
1000000 hashes: 2970M
2000000 hashes: 2370M
4000000 hashes: 1037M

Conclusion: This card rocks, I am looking forward for the 7990!

Next steps: Wait for Linux drivers, then check out how to optimize GCN.


Messages In This Thread
oclHashcat running on the new AMD hd7970 - by atom - 01-11-2012, 09:17 PM