low c/s and no fan speed

[jnet]

I've recently put together a new PC that uses a 5970 and two 5870s (4 GPUs total, crossfire bridges are not being used). This PC uses Ubuntu 12.04LTS and AMD v12.8 drivers that were installed per these directions. The only deviation I had from those instructions was issuing the command "sudo aticonfig -f --adapter=all --initial" so that all four GPUs would initialize.

Here is an output of lspci:

Code:

lspci | grep AMD
03:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI Hemlock [Radeon HD 5900 Series]
03:00.1 Audio device: Advanced Micro Devices [AMD] nee ATI Cypress HDMI Audio [Radeon HD 5800 Series]
04:00.0 Display controller: Advanced Micro Devices [AMD] nee ATI Hemlock [Radeon HD 5900 Series]
05:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI Cypress XT [Radeon HD 5870]
05:00.1 Audio device: Advanced Micro Devices [AMD] nee ATI Cypress HDMI Audio [Radeon HD 5800 Series]
08:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI Cypress XT [Radeon HD 5870]
08:00.1 Audio device: Advanced Micro Devices [AMD] nee ATI Cypress HDMI Audio [Radeon HD 5800 Series]

and fglrxinfo:

Code:

fglrxinfo
display: :0  screen: 0
OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: ATI Radeon HD 5900 Series
OpenGL version string: 4.2.11762 Compatibility Profile Context



display: :0  screen: 1
OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: ATI Radeon HD 5900 Series
OpenGL version string: 4.2.11762 Compatibility Profile Context



display: :0  screen: 2
OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: ATI Radeon HD 5800 Series
OpenGL version string: 4.2.11762 Compatibility Profile Context



display: :0  screen: 3
OpenGL vendor string: Advanced Micro Devices, Inc.
OpenGL renderer string: ATI Radeon HD 5800 Series
OpenGL version string: 4.2.11762 Compatibility Profile Context

and aticonfig --list-adapters

Code:

sudo aticonfig --list-adapters
* 0. 03:00.0 ATI Radeon HD 5900 Series
  1. 04:00.0 ATI Radeon HD 5900 Series
  2. 05:00.0 ATI Radeon HD 5800 Series
  3. 08:00.0 ATI Radeon HD 5800 Series

* - Default adapter

I was excited to see the capabilities/horsepower of my new PC, so I executed an md5 benchmark in hashcat-lite v0.10.

Here is the output of that benchmark:

Code:

sudo ./oclHashcat-lite64.bin -b --benchmark-mode=1 -n 160 --gpu-loops=1000 -m 0
oclHashcat-lite v0.10 by atom starting...

Password lengths range: 1 - 55
Watchdog: Temperature abort trigger disabled
Watchdog: Temperature retain trigger disabled
Device #1: Cypress, 512MB, 725Mhz, 20MCU
Device #2: Cypress, 512MB, 725Mhz, 20MCU
Device #3: Cypress, 512MB, 725Mhz, 20MCU
Device #4: Cypress, 512MB, 725Mhz, 20MCU

NOTE: Runtime limit reached, aborting...


Hash.Type....: MD5
Speed.GPU.#1.:  4931.7M/s
Speed.GPU.#2.:  4781.1M/s
Speed.GPU.#3.:  5114.0M/s
Speed.GPU.#4.:  5266.1M/s
Speed.GPU.#*.: 20066.2M/s

Wow! I was very happy with my bechmark results considering the budget for it was relatively low. I thought it was odd that the detected memory was half of what the GPUs have, but it didn't bother me too much.

To get an idea of oclHashcat-plus64 performance, I issued a simply 6 keyspace brute force attack against the example0.hash list that comes in the zip file package.

Here are the results:

Code:

sudo ./oclHashcat-plus64.bin -a 3 -i -n 160 --gpu-loops=1000 example0.hash ?a?a?a?a?a?a
oclHashcat-plus v0.09 by atom starting...

Hashes: 6494 total, 1 unique salts, 6494 unique digests
Bitmaps: 16 bits, 65536 entries, 0x0000ffff mask, 262144 bytes
Workload: 1000 loops, 160 accel
Watchdog: Temperature abort trigger set to 90c
Watchdog: Temperature retain trigger set to 80c
Device #1: Cypress, 1024MB, 725Mhz, 20MCU
Device #2: Cypress, 1024MB, 725Mhz, 20MCU
Device #3: Cypress, 1024MB, 725Mhz, 20MCU
Device #4: Cypress, 1024MB, 725Mhz, 20MCU
Device #1: Kernel ./kernels/4098/m0000_a3.Cypress_938.2_1.4.1741.kernel (568644 bytes)
Device #2: Kernel ./kernels/4098/m0000_a3.Cypress_938.2_1.4.1741.kernel (568644 bytes)
Device #3: Kernel ./kernels/4098/m0000_a3.Cypress_938.2_1.4.1741.kernel (568644 bytes)
Device #4: Kernel ./kernels/4098/m0000_a3.Cypress_938.2_1.4.1741.kernel (568644 bytes)

[s]tatus [p]ause [r]esume [b]ypass [q]uit => WARN: ADL_Overdrive5_FanSpeedInfo_Get(): -1

Status.......: Exhausted
Input.Mode...: Mask (?a?a?a?a?a?a)
Hash.Target..: File (example0.hash)
Hash.Type....: MD5
Time.Running.: 1 min, 22 secs
Time.Left....: 0 secs
Time.Util....: 82031.5ms/500.7ms Real/CPU, 0.6% idle
Speed........:  8961.1M c/s Real,  9209.2M c/s GPU
Recovered....: 56/6494 Digests, 0/1 Salts
Progress.....: 735091890625/735091890625 (100.00%)
Rejected.....: 0/735091890625 (0.00%)
HWMon.GPU.#1.: 97% Util, 83c Temp, N/A Fan
HWMon.GPU.#2.: 94% Util, 82c Temp, N/A Fan
HWMon.GPU.#3.: 83% Util, 79c Temp, N/A Fan
HWMon.GPU.#4.: 76% Util, 55c Temp, N/A Fan

There are three issues seen:

1. WARN: ADL_Overdrive5_FanSpeedInfo_Get(): -1
   
2. low c/s compared to lite's benchmark (8961.1M c/s Real, 9209.2M c/s GPU)
   
3. the two 5870s are only at 725mhz (I presume this is because of my 5970 though, so it's probably not a problem, just a configuration)
   

I overcame the fan issue by manually setting my fans to 100%. The command I used for that was sudo aticonfig --pplib-cmd "set fanspeed all 100". While this method works, it would be nice if I did not have to manually set this everytime. Has anyone else experienced this issue before? I tried searching the forums, but did not find anything.

The low c/s is very disappointing. Why would the benchmark in lite be so much higher? For grins and giggles, I purged my fglrx/12.8 installations and installed 12.10. Doing this was even worse because only GPU.#1 would work, so I rolled back to 12.8.

I love my new set up, but it seems the numbers are not where they should be.

Does anyone have troubleshooting suggestions? I've tried my best to investigate and troubleshoot on my own, but I'm completely stumped.

Thanks in advance for any assistance!

edit-- possible "DOH!" on my part.... could the reason be because this is 'technically' a mixed GPU set up?