(Guide) Fglrx + AMD APP SDK + Pyrit + Cal++ for AMD Video Cards

[staticn0de]

Hallo,

indeed that isn't a persistant setting. My whole system is under construction and raw. But at the end ~/.bashrc will be set.

By the way, i got a chance to test one HD7990. Here is the result (after DISPLAY=:0):

Computed 195779.45 PMKs/s total.
#1: 'CAL++ Device #1 'AMD GPU DEVICE'': 111024.9 PMKs/s (RTT 1.1)
#2: 'CAL++ Device #2 'AMD GPU DEVICE'': 84497.1 PMKs/s (RTT 1.1)
#3: 'CPU-Core (SSE2)': 695.3 PMKs/s (RTT 3.0)
#4: 'CPU-Core (SSE2)': 701.9 PMKs/s (RTT 3.0)

The same problem.
Interesting is, how much will produce the third GPU?
Without DISPLAY=:0 each chip perform 50.000-60.000 PMks/s

Please let me know staticn0de if you find any way to boost the results on second GPU.

Hi there,

I've been working on this a bit tonight and have some good news and some bad news.

The good news is, you can get second card to be just about the same speed as the first! See below for my initial setup as per the guide before I made changes.

#1: 'CAL++ Device #1 'AMD CAYMAN'': 81018.0 PMKs/s (RTT 1.2)
#2: 'CAL++ Device #2 'AMD CAYMAN'': 47461.8 PMKs/s (RTT 1.5)
#3: 'CPU-Core (SSE2/AES)': 699.7 PMKs/s (RTT 3.0)
#4: 'CPU-Core (SSE2/AES)': 668.8 PMKs/s (RTT 3.0)
#5: 'CPU-Core (SSE2/AES)': 680.9 PMKs/s (RTT 2.9)
#6: 'CPU-Core (SSE2/AES)': 730.4 PMKs/s (RTT 2.9)
#7: 'CPU-Core (SSE2/AES)': 698.4 PMKs/s (RTT 2.9)
#8: 'CPU-Core (SSE2/AES)': 655.1 PMKs/s (RTT 3.0)

My results after changes

Computed 156874.93 PMKs/s total.
#1: 'CAL++ Device #1 'AMD CAYMAN'': 81071.2 PMKs/s (RTT 1.2)
#2: 'CAL++ Device #2 'AMD CAYMAN'': 80010.4 PMKs/s (RTT 1.2)
#3: 'CPU-Core (SSE2/AES)': 1196.1 PMKs/s (RTT 3.0)
#4: 'CPU-Core (SSE2/AES)': 1204.6 PMKs/s (RTT 2.9)

The bad news is, the settings we are changing are setup specific. What I chose through trial and error may not lead to the same result for you. I'll include a few combinations I tried to give you an idea.

1. For the fixing, the most important change I made is turning off hyperthreading. This gave me a massive increase. As it turns out, the CPU was the bottleneck after all.

Second, we have some values we can modify. Each time we chage them, we have to re-build and reinstall that component (not as hard as it sounds)

2. Changing the buffer size

If you followed the guide, this should work

nano +52 /root/pyrit_svn/cpyrit_calpp/_cpyrit_calpp.cpp

change the value between 1 and 4. I had the best performance on 2 and the worst on 1.

3. Rebuilding the cal module for pyrit

You need this step after each buffer size change

root@kali ~/pyrit_svn/cpyrit_calpp$ python setup.py build

root@kali ~/pyrit_svn/cpyrit_calpp$ python setup.py install

4. Modify CAL itself. This setting is either on, or it's off. I had the best speeds with it on.

nano +45 /root/calpp-0.90/include/cal/cal.hpp

Change the line by removing the leading '//' so it looks like this

#define __CAL_USE_NON_BLOCKING_WAIT 1

5. Rebuild call++

cd /root/calpp-0.90/

cmake .

make

make install

Give pyrit benchmark a go and let me know. The result I had after I made this changes. Remember, turning off hyperthreading in the bios was the biggest speed boost.

Best results for me:
Buffer set 2
HT off
cal.hpp removed slashes

Same as before, but buffer on 1

buffer set to 1
Running benchmark (121551.8 PMKs/s)... \ 

Computed 121551.82 PMKs/s total.
#1: 'CAL++ Device #1 'AMD CAYMAN'': 65645.1 PMKs/s (RTT 1.1)
#2: 'CAL++ Device #2 'AMD CAYMAN'': 57752.2 PMKs/s (RTT 1.7)
#3: 'CPU-Core (SSE2/AES)': 1227.4 PMKs/s (RTT 2.9)
#4: 'CPU-Core (SSE2/AES)': 1227.7 PMKs/s (RTT 3.0)
root@kali-corsair:~/pyrit_svn/pyrit#