Why is crossfire slower than a single card?

[rageltman]

Good afternoon. I just purchased a pair of ATI 6970M cards for my 'laptop' (clevo x7200). And i'm seeing some very strange behavior.

I run ubuntu 11.04, installed the newest ATI drivers from their site (11.11), configured crossfire, had to use the --acpi-services=off flag just to get GDM to load, and viola, it runs.

oclhashcat+ though, is a different story. It sees both of the GPUs, but runs slower with both than with one:

SINGLE RUN:

Hashes: 233
Unique digests: 233
Bitmaps: 11 bits, 2048 entries, 0x000007ff mask, 8192 bytes
Rules: 3141
GPU-Loops: 128
GPU-Accel: 40
Password lengths range: 1 - 15
Platform: AMD compatible platform found
Watchdog: Temperature limit set to 90c
Device #1: Barts, 2048MB, 0Mhz, 12MCU
Device #2: skipped by user
Device #1: Allocating 72MB host-memory
Device #1: Kernel ./kernels/4098/m1000_a0.Barts.64.kernel (1108212 bytes)

Status.......: Running
Input.Mode...: File (/tmp/51083936)
Hash.Type....: NTLM
Time.Running.: 6 secs
Time.Left....: 3 mins, 7 secs
Time.Util....: 6815.8ms/155.8ms Real/CPU, 2.3% idle
Speed........: 827.4M c/s Real, 894.1M c/s GPU
Recovered....: 0/233 Digests, 0/1 Salts
Progress.....: 5639461002/160454642976 (3.51%)
Rejected.....: 6282/5639461002 (0.00%)
HW.Monitor.#1: 96% GPU, 66c Temp


Hashes: 240
Unique digests: 240
Bitmaps: 11 bits, 2048 entries, 0x000007ff mask, 8192 bytes
Rules: 3141
GPU-Loops: 128
GPU-Accel: 40
Password lengths range: 1 - 15
Platform: AMD compatible platform found
Watchdog: Temperature limit set to 90c
Device #1: Barts, 2048MB, 0Mhz, 12MCU
Device #2: Barts, 2048MB, 0Mhz, 12MCU
Device #1: Allocating 72MB host-memory
Device #1: Kernel ./kernels/4098/m1000_a0.Barts.64.kernel (1108212 bytes)
Device #2: Allocating 72MB host-memory
Device #2: Kernel ./kernels/4098/m1000_a0.Barts.64.kernel (1108212 bytes)

Scanned dictionary /tmp/51083936: 495738527 bytes, 160454642976 words, starting attack...

[s]tatus [p]ause [r]esume [q]uit => s
Status.......: Running
Input.Mode...: File (/tmp/51083936)
Hash.Type....: NTLM
Time.Running.: 56 secs
Time.Left....: 2 mins, 23 secs
Time.Util....: 56691.8ms/1024.3ms Real/CPU, 1.8% idle
Speed........: 802.5M c/s Real, 809.5M c/s GPU
Recovered....: 7/240 Digests, 0/1 Salts
Progress.....: 45495589002/160454642976 (28.35%)
Rejected.....: 6282/45495589002 (0.00%)
HW.Monitor.#1: 49% GPU, 71c Temp
HW.Monitor.#2: 49% GPU, 71c Temp


Something is very wrong here. Config:
aticonfig --lscs
Candidate Combination:
Master: 4:0:0
Slave: 3:0:0
CrossFire is enabled on current device
CrossFire Diagnostics:
CrossFire can work with P2P mapping through GART
Dongle Capabilities: support PASSTHROUGH |INTERLINK_SW_AFR | INTERLINK_AUTO_AFR | INTERLINK_BLACKING | INTERLINK_SUPERAA

aticonfig --lscc

Master adapter: 0. 04:00.0 AMD Radeon HD 6900M Series
Candidates: 1. 03:00.0 AMD Radeon HD 6900M Series
aticonfig --lsch

CrossFire chain for adapter 0, status: enabled
0. 04:00.0 AMD Radeon HD 6900M Series
1. 03:00.0 AMD Radeon HD 6900M Series

[radix]

You shouldnt be running crossfire with oclhashcat's. Disable it.

[rageltman]

So i disabled crossfire on the cards, and now oclhashcatplus can't see the second card at all. What am i doing wrong here?

[KT819GM]

sudo aticonfig --adapter=all --initial -f
reboot

[rageltman]

sudo aticonfig --adapter=all --initial -f
reboot

Did it, X wont come up because it thinks that there's a display attached to the secondary adapter on PCI:3

Cleared out the second monitor section from xorg.conf and X started, still cant see the second adapter.

[rageltman]

Anyone else running a pair of these cards in a laptop? Seems that the only way both cards are seen is with crossfire, and that makes them slower. Complete bloody ATI BS it seems (been using nvidia for years, and i've lost enough of my life on ATI drivers). Still, apparently this is the best hardware i can get in my machine so i need to get it working...

[atom]

try with oclHashcat-lite v0.08 pls

[rageltman]

try with oclHashcat-lite v0.08 pls

Thanks Atom, but that fails even worse.

GPU-Loops: 1024
GPU-Accel: 160
Password lengths range: 4 - 55
Platform: AMD compatible platform found
Watchdog: Temperature limit set to 90c
Device #1: Barts, 1024MB, 0Mhz, 12MCU
Device #2: Barts, 1024MB, 0Mhz, 12MCU
ERROR: ./kernels/4098/m0000q_warp.Barts.64.kernel: No such file or directory

$ ./oclHashcat-lite64.bin -d1 4cf1c36009(removed 6 chars)20f2f607467beb0c ?s?s?s?s?s?s?s?s

same issue running with -d1

I'm starting to think that all the hype about ATI being better for hash analysis is a load of BS. They had horrible driver problems 12 years ago (check my nick) and they're STILL bloody incompetent. If they sell a SLI/Xfire product, then it should actually WORK that way.

Also, it appears that when using card2 in crossfire, the computational results are less than stellar. It just says every hash can be resolved with random binary output.

Does anyone on these forums use a dual ATI setup on a laptop with linux? Pentesters usually can't send hashes to a third party for analysis (clients dont even like them leaving the premises on an encrypted disk) so i'm probably not the first one to try and pack a nuclear solution to this problem in a portable manner.

[atom]

well all i can say is that if i would be a pentester, i would never use my notebook for cracking. i would just transfer the hashes to a workstation or cracking rig in my office and let crack it. its both cheaper and faster.

[atom]

oh and this:

ERROR: ./kernels/4098/m0000q_warp.Barts.64.kernel: No such file or directory

is mostly because of invalid use of unpacker. use 7z x file.7z

here you can see its there:

root@sf:~/xy/oclHashcat-lite-0.08/kernels/4098# ls -l m0000q_warp.Barts.64.kernel
-rw------- 1 root root 272774 Nov 25 10:10 m0000q_warp.Barts.64.kernel