Forum > GPU crunching
Latest drivers (NVidia and ATI)
Urs Echternacht:
--- Quote from: Frizz on 01 Nov 2011, 07:29:10 am ---...And some other nasty bugs for Linux.
--- End quote ---
Using still SDK 2.4 on some openSuse 11.3, 64bit.
With the switch from Cat 11.9 to Cat 11.10 there are now the same problems introduced, that are hindering LHD4K mod to work on windows. Now stops on my HD4670, which worked ok for the last few month, not only the GPU but the whole host. Screen showing, host frozen. Definitly a driver problem killing our OpenCL (1.0?) efforts.
Have to check my other Linux hosts (openSuse 11.4, Ubuntu 10.10) if similar will happen there.
Urs Echternacht:
On Suse 11.4 now autocorrs are overflowing immediately when using Cat 11.10. Will switch back to previous driver. Cat 11.6 is still the one of choice for my two GPU host.
<stderr_out>
<![CDATA[
<stderr_txt>
Number of period iterations for PulseFind setted to: 15
Running on device number: 0
OpenCL-kernels filename : MultiBeam_Kernels_r375.cl
OpenCL platform detected: Advanced Micro Devices, Inc.
Number of OpenCL devices found : 1
BOINC assigns 0 device, slots 0 to 0 (including) will be checked
Info : Building Program (clBuildProgram):main kernels: OK code 0
Linux optimized S@H v7 application (based on S@H Enhanced by Alex Kan)
Version info: SSE3 (AMD/Intel, Optimized v8-nographics), V5.13 by Alex Kan
SSE3 Linux64 Build 375 , Ported by : Jason G, Raistmer, JDWhale, Urs Echternacht
Original GPU DCT by Jason G
OpenCL version by Raistmer, r375
AMD HD5 version by Raistmer
Build: SSE3 System: Linux x86_64 Kernel: 2.6.37.6-0.7-desktop
CPU : Intel(R) Core(TM) i5-2500T CPU @ 2.30GHz
4 core(s), Speed : 3201.000 MHz
L1 : 64 KB, L2 : 6144 KB
Number of OpenCL platforms: 1
OpenCL Platform Name: AMD Accelerated Parallel Processing
Number of devices: 1
Max compute units: 6
Max work group size: 256
Max clock frequency: 800Mhz
Max memory allocation: 134217728
Cache type: None
Cache line size: 0
Cache size: 0
Global memory size: 536870912
Constant buffer size: 65536
Max number of constant args: 8
Local memory type: Scratchpad
Local memory size: 32768
Queue properties:
Out-of-Order: No
Name: Turks
Vendor: Advanced Micro Devices, Inc.
Driver version: CAL 1.4.1589
Version: OpenCL 1.1 AMD-APP-SDK-v2.5 (684.213)
Extensions: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_sharing cl_ext_atomic_counters_32 cl_amd_device_attribute_query cl_amd_vec3 cl_amd_printf cl_amd_media_ops cl_amd_popcnt
Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.008456
New autocorr logged:
score=0.5447, peak_power=140.2, bin=64484, fft_ind=0
New autocorr logged:
score=0.5253, peak_power=134.1, bin=14771, fft_ind=1
New autocorr logged:
score=0.5406, peak_power=138.9, bin=30001, fft_ind=2
New autocorr logged:
score=0.5439, peak_power=140, bin=16690, fft_ind=3
New autocorr logged:
score=0.5555, peak_power=143.7, bin=35093, fft_ind=4
New autocorr logged:
score=0.5482, peak_power=141.3, bin=20349, fft_ind=5
New autocorr logged:
score=0.572, peak_power=149.3, bin=65303, fft_ind=6
New autocorr logged:
score=0.5641, peak_power=146.6, bin=56811, fft_ind=7
New autocorr logged:
score=0.5527, peak_power=142.8, bin=64321, fft_ind=0
New autocorr logged:
score=0.517, peak_power=131.5, bin=44968, fft_ind=1
New autocorr logged:
score=0.5388, peak_power=138.3, bin=21885, fft_ind=2
New autocorr logged:
score=0.5372, peak_power=137.8, bin=37149, fft_ind=3
New autocorr logged:
score=0.5473, peak_power=141, bin=39324, fft_ind=4
New autocorr logged:
score=0.5484, peak_power=141.4, bin=27098, fft_ind=5
New autocorr logged:
score=0.5535, peak_power=143.1, bin=25439, fft_ind=6
New autocorr logged:
score=0.5332, peak_power=136.5, bin=43866, fft_ind=7
New autocorr logged:
score=0.5391, peak_power=138.4, bin=6430, fft_ind=0
New autocorr logged:
score=0.5415, peak_power=139.2, bin=51557, fft_ind=1
New autocorr logged:
score=0.5395, peak_power=138.5, bin=38147, fft_ind=2
New autocorr logged:
score=0.5419, peak_power=139.3, bin=37320, fft_ind=3
New autocorr logged:
score=0.5335, peak_power=136.6, bin=1918, fft_ind=4
New autocorr logged:
score=0.5517, peak_power=142.5, bin=20310, fft_ind=5
New autocorr logged:
score=0.5903, peak_power=155.7, bin=65349, fft_ind=6
New autocorr logged:
score=0.5332, peak_power=136.5, bin=43866, fft_ind=7
New autocorr logged:
score=0.5586, peak_power=144.8, bin=64321, fft_ind=0
New autocorr logged:
score=0.5227, peak_power=133.3, bin=18265, fft_ind=1
New autocorr logged:
score=0.5446, peak_power=140.2, bin=30001, fft_ind=2
New autocorr logged:
score=0.532, peak_power=136.1, bin=37352, fft_ind=3
New autocorr logged:
score=0.5485, peak_power=141.4, bin=17803, fft_ind=4
New autocorr logged:
score=0.5478, peak_power=141.2, bin=27058, fft_ind=5
OpenCL queue synchronized
SETI@Home Informational message -9 result_overflow
NOTE: The number of results detected exceeds the storage space allocated.
Flopcounter: 7903362043.636971
Spike count: 0
Autocorr count: 30
Pulse count: 0
Triplet count: 0
Gaussian count: 0
Wallclock time elapsed on cpu since last restart: 29.9 seconds
16:40:29 (7712): called boinc_finish
</stderr_txt>
]]>
</stderr_out>
Frizz:
I wonder why AMD drivers are of such bad quality (recently). Maybe because they are internally already adding support for the new GCN architecture? And break support for "older" cards?
Urs Echternacht:
--- Quote from: Frizz on 01 Nov 2011, 03:57:45 pm ---I wonder why AMD drivers are of such bad quality (recently). Maybe because they are internally already adding support for the new GCN architecture? And break support for "older" cards?
--- End quote ---
More that they never reached a "clean level" of having a minimum number of problems in a driver. Instead carrying this payload of problems with them makes it harder to implement something new(er) a way that it works. :-\
Have finished testing Cat 11.10 on 64 bit linux meanwhile.
Resumee :
Unusable for our OpenCL GPU apps. Even fails with same error picture on Kubuntu 10.10.
My suggestion for tested 64bit linux distros :
With one GPU Cat 11.9 works best, with two Cat 11.6 works better.
Raistmer:
We need to locate place of failure. I've seen the same with windows Cat preview.
still unknown where exactly data start to differ between driver versions.
Please, post your findings in corresponding thread on AMD forum too.
http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=155591&enterthread=y
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version