What steps will reproduce the problem?

I have bt4 on Intel machine (i7 975) with ATI 10.2 drivers and HD5970.Stream SDK 2.01 succesfully installed and working.I'm using Pyrit 0.3.1-dev (svn r228)

What is the expected output? What do you see instead?

root@bt:~# pyrit selftest Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'OpenCL-Device 'Cypress''

2: 'OpenCL-Device 'Cypress''

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

9: 'Network-Clients'

Running selftest... Exception in thread Thread-2: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 111, in run self._testComputeFunction(101) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 101, in _testComputeFunction raise ValueError("Test-vector does not result in correct PMK.") ValueError: Test-vector does not result in correct PMK.

Exception in thread Thread-1: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 108, in run essid, pwlist = self.queue._gather(self.buffersize, timeout=0.5) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 552, in _gather self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Traceback (most recent call last): File "/usr/bin/pyrit", line 6, in <module> pyrit_cli.Pyrit_CLI().initFromArgv() File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 106, in initFromArgv func(self, **options) File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 854, in selftest cp.enqueue(essid, pws) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 484, in enqueue self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Help please Thanks

I'm not to sure what the problem is. Though lukas doesn't recommend the 10.2 drivers. http://pyrit.wordpress.com/2010/02/21/ati-catalyst-10-2/ Perhaps you should try with the 10.1 drivers?

1. Please try downgrading to 10.1

2. Can you successfully run the SDK-examples? Not just HelloCL but also the more complex ones like Mandelbrot

3. Please execute the following command in a shell and post the result

python -c "from cpyrit import _cpyrit_opencl; print [(i, _cpyrit_opencl.OpenCLDevice(0,i).solve('bla',('blablabla',)*3)) for i in xrange(_cpyrit_opencl.OpenCLPlatform(0).numDevices)]"

1 When i use ATI 10.2 I can suuccessfully run all SDK-examples

2 The output of comand under ATI 10.2 is [(0, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (1, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (2, ('\xc1\xed\xb1dJ\x93\xcb5sg0b\xfd\xc2\x8eO5\xef\xea*\x9c\xff\xc7I\xfc\x16\x14\x0e\xe3\x8f\xa3\xa8', '\x89XU"\xfa\xfb\xc7\x96\xda\x8c\xc5\xdd\xf4\xe5\xefG\xa4\xb5(B\xc2\x7f\xf6\xbc\xb7\xe0\xfdg\r\x99v<', '(h|L\xce\xceq\xd46#\xa2{\xcaq\xf7\xd7\xd7\xa0IT\xd1\r\xe4\x00\x1b(DCT\x18\xc1Q'))]

3 Now I'll try to downgrade to ATI 10.1

I have downgrade to ATI 10.1 but

root@bt:~# pyrit Segmentation fault

and all SDK-examples return in Segmentation fault

root@bt:/usr/local/opencl/samples/opencl/bin/x86# ./AESEncryptDecrypt Encrypting Image .... Input Image : input512.bmp Key : 15 201 51 89 92 34 96 66 11 225 161 96 81 211 108 124 Segmentation fault

Device 0 is your CPU, Device 1 and Device 2 are the two GPUs. Like I expected: The second GPU is producing wrong results. There may be several reasons for that:

• A bug in Pyrit's OpenCL-code (unlikely, there is a ton of error checking)
• A bug in ATIs OpenCL-implementation (very likely, given ATIs history of driver-quality)
• Your GPU is not working correctly (unlikely, but possible)

Please open /usr/lib/python2.5/site-packages/cpyrit/cpyrit.py, find line 375 and change it as followed:

for dev_idx in range(p.numDevices):

should read

for dev_idx in range(p.numDevices)[:1]:

This will cause only the first GPU to be used. Please see if 'selftest' is working after that change. If it is, change the line to

for dev_idx in range(p.numDevices)[1:]:

This will cause only the second GPU to be used. Again, see if 'selftest' is working.

If it is a bug in ATI's driver, we face a average of two months before ATI recognizes the problem...

I'm not sure, that about this:

I have downgrade to ATI 10.1 but root@bt:~# pyrit Segmentation fault

I had the same on Ubuntu 9.10 - this is known ATI SDK bug with kernels newer than 2.6.19 or so

by adding NOPAT to kernel boot string - eliminated segmentation fault in my case try to google on that issue

Now I'm trying with ATI 10.1 With nopat kernel boot option there's not Segmentation fault

I have edited /usr/lib/python2.5/site-packages/cpyrit/cpyrit.py, line 375 for the first GPU for dev_idx in range(p.numDevices)[:1]: this is the output of "pyrit selftest"

root@bt:/usr/local/opencl/samples/opencl/bin/x86# pyrit selftest Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'CPU-Core (SSE2)'

2: 'CPU-Core (SSE2)'

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

Running selftest...

All results verified. Your installation seems OK.

The GPU's are no longer recognized.

But if I edit /usr/lib/python2.5/site-packages/cpyrit/cpyrit.py, line 375 for the second GPU for dev_idx in range(p.numDevices)[1:]: this is the output

root@bt:/usr/local/opencl/samples/opencl/bin/x86# pyrit selftest Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'OpenCL-Device 'Cypress''

2: 'OpenCL-Device 'Cypress''

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

Running selftest... Exception in thread Thread-2: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 111, in run self._testComputeFunction(101) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 101, in _testComputeFunction raise ValueError("Test-vector does not result in correct PMK.") ValueError: Test-vector does not result in correct PMK.

Exception in thread Thread-1: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 108, in run essid, pwlist = self.queue._gather(self.buffersize, timeout=0.5) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 552, in _gather self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Traceback (most recent call last): File "/usr/bin/pyrit", line 6, in pyrit_cli.Pyrit_CLI().initFromArgv() File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 106, in initFromArgv func(self, **options) File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 854, in selftest cp.enqueue(essid, pws) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 484, in enqueue self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Now I'll try to revert to ATI 10.2,but I suspect that the result will be the same

bawww, I didnt take account for the CPU-Device. It should read:

for dev_idx in range(p.numDevices)[:2]

and

for dev_idx in range(p.numDevices)[2:]

ok Now with for dev_idx in range(p.numDevices)[:2]

root@bt:~# pyrit selftest Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'OpenCL-Device 'Cypress''

2: 'CPU-Core (SSE2)'

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

Running selftest...

All results verified. Your installation seems OK.

and with for dev_idx in range(p.numDevices)[2:]

oot@bt:~# pyrit selftest Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'OpenCL-Device 'Cypress''

2: 'CPU-Core (SSE2)'

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

Running selftest... Exception in thread Thread-1: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 111, in run self._testComputeFunction(101) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 101, in _testComputeFunction raise ValueError("Test-vector does not result in correct PMK.") ValueError: Test-vector does not result in correct PMK.

Exception in thread Thread-8: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 108, in run essid, pwlist = self.queue._gather(self.buffersize, timeout=0.5) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 552, in _gather self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Traceback (most recent call last): File "/usr/bin/pyrit", line 6, in pyrit_cli.Pyrit_CLI().initFromArgv() File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 106, in initFromArgv func(self, **options) File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 854, in selftest cp.enqueue(essid, pws) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 484, in enqueue self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

My card, Ati hd5970, is not crossfire but double GPU on single pcb...How can I test if the second GPU is fully functional?

Thanks for the information; this shows that the problem is the solely the second GPU and not interaction/multithreading. Is the result you posted above stable? Can you re-run the selftest with first or second GPU and get the same result?

For the moment, I suggest you stick with "for dev_idx in range(p.numDevices)[:2]" so only the first GPU is used. This is a workaround that gives you a working version of Pyrit. I'll see what ATI has to say about this.

The result of 'pyrit benchmark' on single-gpu would be of interest for me.

I can confirm that the test results are stable This is 'pyrit benchmark'

root@bt:~# pyrit benchmark Pyrit 0.3.1-dev (svn r228) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Running benchmark (34770.9 PMKs/s)... -

Computed 34877.81 PMKs/s total.

1: 'OpenCL-Device 'Cypress'': 30632.9 PMKs/s (RTT 1.9)

2: 'CPU-Core (SSE2)': 463.6 PMKs/s (RTT 2.9)

3: 'CPU-Core (SSE2)': 444.5 PMKs/s (RTT 2.9)

4: 'CPU-Core (SSE2)': 470.1 PMKs/s (RTT 2.9)

5: 'CPU-Core (SSE2)': 464.6 PMKs/s (RTT 2.8)

6: 'CPU-Core (SSE2)': 437.5 PMKs/s (RTT 2.9)

7: 'CPU-Core (SSE2)': 468.4 PMKs/s (RTT 3.1)

8: 'CPU-Core (SSE2)': 435.0 PMKs/s (RTT 3.1)

Thanks for help

Comment deleted

More information about this:

• I've dumped the .il and the .isa file that get generated for both GPUs on the HD5970. They are identical.
• It takes a very long time on the second GPU for clFinish() to return (seval seconds). It finally returns CL_SUCCESS (the call to clFinish() takes roughly a few milliseconds on the first GPU).
• CL_EVENT_COMMAND_EXECUTION_STATUS for clEnqueueNDRangeKernel() is always CL_COMPLETE
• The kernel-output on the second GPU is always wrong, but always the same every time the kernel is called. It behaves regardless of the input (!).

Maybe I need to disable internal CROSSFIRE ??

root@bt:/home/Sorgenti/pyrit/pyrit_svn/pyrit/test# aticonfig --lscc

Master adapter: 0. 06:00.0 ATI Radeon HD 5900 Series Candidates: 1. 07:00.0 ATI Radeon HD 5900 Series

root@bt:/home/Sorgenti/pyrit/pyrit_svn/pyrit/test# aticonfig --lscs Candidate Combination: Master: 6:0:0 Slave: 7:0:0 CrossFire is enabled on current device CrossFire Diagnostics: There is CrossFire Side port connection between GPUs CrossFire can work with P2P write through peer aperture Dongle Capabilities: support PASSTHROUGH |INTERLINK_SW_AFR | INTERLINK_AUTO_AFR | INTERLINK_BLACKING | INTERLINK_SUPERAA

root@bt:/home/Sorgenti/pyrit/pyrit_svn/pyrit/test# aticonfig --lsch

CrossFire chain for adapter 0, status: enabled 0. 06:00.0 ATI Radeon HD 5900 Series 1. 07:00.0 ATI Radeon HD 5900 Series

You can try this.

In any case, the second GPU should never return incorrect results without indicating an error-condition through the API.

No success Same error, same benchmark

• only to share my feeling - Im following this issue with pain ("pain" is referred to me not to the issue). Late in 2009, I choosed to wait Fermi, but now I hate the delay of NVIDIA with Fermi and his neverending sequence of fake announcements... ok, I say to me, I go with ATI... but then I see this so-low-level-driver-quality in 9.12, 10.1, 10.2 .... Bad time, bad time we - the customers - are living...

Comment deleted

@odian3

(note: I talk related to our needs, I mean pyrit user viewpoint, not gamer user viewpoint.)

I dont know you are talking about HARDware or SOFTware developers, anyway:

about NVIDIA. They said: "Fermi ready at end of 2009", then "Beginning of 2010", then "second quarter of 2010". During this period they show fake Fermi card, and take the attention up with "hype news" every week but it is just bla bla bla, smoke and hype. Result is: 6 months of delay in shipment Fermi card. NVIDIA official news "advise" future costomers to have a case ready to manage "a card that will be little hotter than usual": it means Fermi will be hot like hell. And last: they are working on Fermi 2: it means: actual Fermi is a "beta" chipset, the right one will be Fermi 2. More, they "price politic" is VOLTURE: prices are SO high without have HIGH performances. Have a looke ATI/NVIDIA similar card performances: there is a 100 Euro average difference (ATI=XEURO, NVIDIA=(X+100)Euro). My opinion: shame on you, NVIDIA.

about ATI. ATI has chiper, colder and powerfull graphic cards then NVIDIA, but their drivers are VERY unusable: have a look the the post of lukas related 9.12, 10.1 and 10.2 ATI drivers. The best card in the word has value=0 if drivers make her not usable. My opinion: shame on you, ATI.

Anyway, my money are not fake, my money are not alpha version. So, as my money are good, I have right to ask for good hardware/software. For the moment, I decided I will NOT buy ATI (till STABLE and USABLE drivers will be avaiable), I will NOT buy NVIDIA (till Fermi 2)

please take discussions elsewhere

ATI confirmed a problem on their side.

Is it mean, that odlan3's 5970 card is OK and ATI again has a bug?

Cpyrit Doesn't work with my HD 4870X2 too. I tried also with two HD4870 and if i disable Crossfire the second card disappears and it will be unusable. It is a problem of ATI or Cpyrit?

ATI. The driver is simply not working.

anyone tried the 10.3 "preview" driver?

I' m waiting for Linux version. I will check as soon as the new drivers and you'll know

10.3 is out. Any news on the problem?

I have started download now. Some minute and I'll tell you

I have to recompile pyrit ? If not I have the same error with Catalyst 10.3

pyrit selftest yrit 0.3.1-dev (svn r237) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Cores incorporated in the test:

1: 'OpenCL-Device 'Cypress''

2: 'OpenCL-Device 'Cypress''

3: 'CPU-Core (SSE2)'

4: 'CPU-Core (SSE2)'

5: 'CPU-Core (SSE2)'

6: 'CPU-Core (SSE2)'

7: 'CPU-Core (SSE2)'

8: 'CPU-Core (SSE2)'

Running selftest... Exception in thread Thread-2: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 111, in run self._testComputeFunction(101) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 101, in _testComputeFunction raise ValueError("Test-vector does not result in correct PMK.") ValueError: Test-vector does not result in correct PMK.

Exception in thread Thread-1: Traceback (most recent call last): File "/usr/lib/python2.5/threading.py", line 486, in __bootstrap_inner self.run() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 108, in run essid, pwlist = self.queue._gather(self.buffersize, timeout=0.5) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 552, in _gather self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

Traceback (most recent call last): File "/usr/bin/pyrit", line 6, in pyrit_cli.Pyrit_CLI().initFromArgv() File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 110, in initFromArgv func(self, **options) File "/usr/lib/python2.5/site-packages/pyrit_cli.py", line 941, in selftest cp.enqueue(essid, pws) File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 484, in enqueue self._check_cores() File "/usr/lib/python2.5/site-packages/cpyrit/cpyrit.py", line 410, in _check_cores raise SystemError("The core '%s' has died unexpectedly" % core) SystemError: The core 'OpenCL-Device 'Cypress'' has died unexpectedly

No, you don't. The OpenCL-code is compiled at run-time.

Please post the the result of the following command:

python -c "from cpyrit import _cpyrit_opencl; print [(i, _cpyrit_opencl.OpenCLDevice(0,i).solve('bla',('blablabla',)*3)) for i in xrange(_cpyrit_opencl.OpenCLPlatform(0).numDevices)]"

This is the output

[(0, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (1, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (2, ('\x98x\x80\xeaZ\x8d\xb2\x91\x17\xdc5\x1d0\xf6\xe6^\xdf\x10&\x94\xcf\x08s\xad\xb3\xbf\x85\xe2\xa6\\x9c\xdc', '\xefB\xd6\xa7aPp\x87r85x\x0e\x8d\xa4\xc6\x0c&\xa7A_]/\xad\xee\xce\xa8\xa6\xbcJ\xf3:', '\x80\xe5\xa4\x96J\x92\xd9\x8c\x0f\xed\xbe\x9fC+=?qLu\xb3\x87J\x0e\x01\xdb\xba\xc3\xc9dG\xed\x05'))]

so the second gpu is still producing random, incorrect results. If ATI does not provide a hotfix or a workaround in the meantime, you'll have to wait for 10.4 (in april) and try again .....

Thanks for support

odlan3,

Check out what worked for me over on the issue 137 thread (using 2 4850 cards, which wouldn't work in pyrit while crossfire'd). If you disable internal crossfire and try my setup, you may be able to use both cores before the next driver release (I'm using Cat 10.3).

BTW, I'm using the calpp version right now. I haven't tested it on OpenCL (which also didn't work crossfire'd) but the fix should work the same, as I stole it from a thread about OpenCL. After reading what the 5000 series cards are benchmarking at under calpp (included in latest svn), you may want to switch to that. I'd try my fix with OpenCL first, since you're already set up for that. Good Luck!

Thanks....It works!!!! pyrit benchmark Pyrit 0.3.1-dev (svn r245) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Running benchmark (64930.3 PMKs/s)... /

Computed 68346.30 PMKs/s total.

1: 'OpenCL-Device 'Cypress'': 32414.2 PMKs/s (RTT 1.9)

2: 'OpenCL-Device 'Cypress'': 31876.3 PMKs/s (RTT 1.9)

3: 'CPU-Core (SSE2)': 550.7 PMKs/s (RTT 2.9)

4: 'CPU-Core (SSE2)': 565.4 PMKs/s (RTT 2.9)

5: 'CPU-Core (SSE2)': 530.4 PMKs/s (RTT 3.1)

6: 'CPU-Core (SSE2)': 536.3 PMKs/s (RTT 2.8)

7: 'CPU-Core (SSE2)': 582.3 PMKs/s (RTT 2.8)

8: 'CPU-Core (SSE2)': 533.4 PMKs/s (RTT 3.0)

The CAL-Core should give you even higher performance...

I have same problem to compiling CAL-CORE

odlan3 please open a new issue for that

I have impression that comment #51 in issue #148 solve issue #123 :)

I keep this issue open until there is an official fix by ATI. The drivers still produce random results without ANY error indication....

I am guessing problem can be in xorg, BackTrack4 use xorg 7.4 I had segfault with BT4, so I wipe it out and go with debian.

I run: - debian 5.0.4 stable on amd64 - xorg 7.3 from .deb package - ATI driver 10.2 from ATI site - ATI SDK 2.01 from ATI site - python 2.6.4 from source tar.gz from python site, compiled and installed - boost 1.42.0 from source tar.gz from boost site, compiled and installed - pyrit with calpp-v2b-1 so as showed in issue #149.

hardware: 1 HD5770 on 1st PCI-X bus at x16 1 HD8770 on 3st PCI-X bus at x16

The system is up and running since saturday in neverending passthrough work, it is stable, no crach or dump or what else.

Hope this info can help.

Hey,now is fixed using cpyrit_opencl

python -c "from cpyrit import _cpyrit_opencl; print [(i,

_cpyrit_opencl.OpenCLDevice(0,i).solve('bla',('blablabla',)*3)) for i in xrange(_cpyrit_opencl.OpenCLPlatform(0).numDevices)]" [(0, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (1, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c')), (2, ('\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c', '\xb8\xb7\xf5E^\xfe\xb1\x8b\xf5\x98\xce\x95z$%\xd4\x05\xdb4\xb6E\xeb\xf4\xfd\x92\xf0\xa18\x18\xae\xaa\x1c'))]

This is what I have done aticonfig --adapter=0 --cfd (remove any crossfire chains) aticonfig --adapter=1 --cfd (remove any crossfire chains) aticonfig --adapter=all --initial -f (create new xorg.conf) export DISPLAY=:0

This whit ncpus-=2 pyrit benchmark Pyrit 0.3.1-dev (svn r250) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Running benchmark (68003.6 PMKs/s)... |

Computed 68829.74 PMKs/s total.

1: 'OpenCL-Device 'Cypress'': 33249.7 PMKs/s (RTT 1.8)

2: 'OpenCL-Device 'Cypress'': 32210.5 PMKs/s (RTT 1.8)

3: 'CPU-Core (SSE2)': 578.4 PMKs/s (RTT 2.8)

4: 'CPU-Core (SSE2)': 573.1 PMKs/s (RTT 2.9)

5: 'CPU-Core (SSE2)': 592.5 PMKs/s (RTT 3.2)

6: 'CPU-Core (SSE2)': 614.2 PMKs/s (RTT 2.8)

7: 'CPU-Core (SSE2)': 603.3 PMKs/s (RTT 2.9)

8: 'CPU-Core (SSE2)': 628.9 PMKs/s (RTT 3.1)

This whit ncpus-=4 pyrit benchmark Pyrit 0.3.1-dev (svn r250) (C) 2008-2010 Lukas Lueg http://pyrit.googlecode.com This code is distributed under the GNU General Public License v3+

Running benchmark (73917.1 PMKs/s)... |

Computed 76168.07 PMKs/s total.

1: 'OpenCL-Device 'Cypress'': 37503.7 PMKs/s (RTT 1.6)

2: 'OpenCL-Device 'Cypress'': 35855.9 PMKs/s (RTT 1.7)

Anyway version of CAL++ seems to be more performant

I think this issue can be closed as well. I think the "garbage" output from the second GPU was due to it being the secondary GPU in a crossfire setup. Once odlan de-crossfired his GPUs, his output was as it should be. The problem he had when producing garbage output for GPU #2 was due to internal crossfire being enabled. Also, when he disabled crossfire, the second GPU wasn't seen by pyrit. That got fixed in issue 137, so now everything works as it should. I don't see this as either a driver or pyrit issue. AMD admits that Stream doesn't work under crossfire. And the only reason pyrit couldn't see the 2d GPU was because we had to change the default DISPLAY variable from :0.0 to :0, so all GPUs would be used on one screen.

Unless of course, you saw another issue in your testing that I'm just not getting. (Always a possibility).

@robert.bleach I think you are right but last word is for the developpers Thanks to lukas, hazeman11 and thank you robert for help

odlan,

Absolutely. Lukas gets the final word. It's his baby. I just don't want him thinking that there's a bug where there isn't. I'm sure he has enough on his plate as is. It appeared that the test results he got from your second GPU bothered him, and I wasn't sure if he realized that those bad results were from when you were crossfire'd. Now that you're de-crossfired, your results are coming up correct for both GPUs, right?

And you're more than welcome for any help I've given. I wish I had lukas' or hazeman's programming knowledge so I could contribute more. As it is, I just got lucky finding the whole DISPLAY=:0 thing while scouring the ATI Stream SDK forums. At least I have tenacity working for me to make up for lack of programming skills. ;)

Bob

closed, bug in ati drivers. workarounds seem to exist.