ATI/AMD GPU-only and HYBRID GPU+CPU Optimized AstroPulse 5.06 (r449-r456) for Windows OS Three editions available now at GPU Apps Public Downloads areaFirst of all:
This release intended for ADVANCED USERS. If you not sure you can repair your BOINC installation if something will go wrong - wait when this app will be included in Lunatics installer.
There are 3 different builds suitable for different types of ATI/AMD hardware.
1) OpenCL based application that can be used on HD4xxx/5xxx series of ATI GPUs.
2) OpenCL/Brook+ combo app that can be used on same GPU classes (HD4xxx/HD5xxx) but maybe provide different (better) performance/experience for some host setups
3) Brook+ GPU/CPU "hybrid" build - it suitable for older ATI GPUs and on-board chips starting from HD2xxx seria. It's update release for released before Hybrid AP rev393.
Requirements:
1), 2):
ATI GPU 4xxx or 5xxx,
ATI Stream SDK2.2 installed, CPU with SSE2 and better support,
Catalyst 10.7b drivers.
[
for some hosts with only HD4xxx GPU installed additional environment variable may be needed:
GPU_MAX_HEAP_SIZE=256To set it go My computer->Properties->Advanced->Environment variables (path can be slightly different in different windows versions). But before setting such variable run application and check stderr.
If you see line:
Global memory size:
with value >128MB - you don't need this environment variable
]
You can download SDK here:
http://developer.amd.com/gpu/ATIStreamSDK/Pages/default.aspxRelease notes for 1):
There are 2 command line switches that can be used for app performance tuning:
-ffa_block 8192 (default value) - defines how many different periods GPU will process per single kernel call
-ffa_block_fetch 2048 (default value) - defines how many threads will be used in FFA initial fetch kernel
Rules for using these values:
-ffa_block_fetch <number> can be used only if -ffa_block <number> already listed in command line
numbers should be even,better if they will be power of 2, ffa_block should be divisible by ffa_block_fetch.
If you experience lags during application execution try to decrease these values.
-disable_slot <slot_number>
It should allow to exclude broken device even if BOINC assigns task to it. For better use it's worth to set count value in coproc section from 1 to 2 (or as appropriate) to limit number of tasks launched by BOINC. If all GPUs work OK no need to use this option.
Known issues::
1) App will not work (no progress on secondary chip at all or will produce invalid results) on second core of dual-core GPUs like 5970 and will produce invalid results(or no progress at all) with CrossFire enabled on non-primary GPUs in system.
It's ATI SDK2.2 related bug, ask ATI for fix.
2) It can blink with popup console windows at each new task start. Again, it's long standing bug with ATI SDK:
http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=136672&enterthread=yAsk AMD for fix.
EDIT:
this bug was solved in Cat 10.12, update drivers.Requirements for 3) :
ATI GPU 2xxx or better,
Catalyst 9.x or better drivers
To get any AP tasks for GPU don't forget to enable "Send work from other apps" on web preferencies pageapp_info.xml section for this app (for testing on main):
app_info.xml file sections for using these apps:
1), 2): ( for 2) don't forget to change executable file name in provided sample. Sample can be used as is only for 1) app )
<app>
<name>astropulse_v505</name>
</app>
<file_info>
<name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r456.exe</name>
<executable/>
</file_info>
<file_info>
<name>AstroPulse_Kernels.cl</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse_v505</app_name>
<version_num>506</version_num>
<avg_ncpus>0.01</avg_ncpus>
<max_ncpus>0.01</max_ncpus>
<plan_class>ati13ati</plan_class>
<cmdline>-ffa_block 8192 -ffa_block_fetch 2048</cmdline>
<coproc>
<type>ATI</type>
<count>1</count>
</coproc>
<flops>15987654321</flops>
<file_ref>
<file_name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r456.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>AstroPulse_Kernels.cl</file_name>
<copy_file/>
</file_ref>
</app_version>
3)
<app>
<name>astropulse_v505</name>
</app>
<file_info>
<name>ap_5.06_x86_win_SSE_BROOK_NO_DOUBLE_r453.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse_v505</app_name>
<version_num>505</version_num>
<avg_ncpus>0.1</avg_ncpus>
<max_ncpus>0.1</max_ncpus>
<plan_class>ati13ati</plan_class>
<coproc>
<type>ATI</type>
<count>0.5</count>
</coproc>
<flops>6000000000</flops>
<file_ref>
<file_name>ap_5.06_x86_win_SSE_BROOK_NO_DOUBLE_r453.exe</file_name>
<main_program/>
</file_ref>
</app_version>
Please, edit <max/avg_ncpus> and <count> tags values as appropriate for your config.
For hosts participating in other ATI-based projects <count>0.5</count> can be changed to <count>0.1</count>
I would like to thank all Lunatics crew for continual and very helpfull support. W/o it this release could not be happen.
Author
Logged
