Welcome to MilkyWay@home

Posts by (retired account)

1) Message boards : Number crunching : Feature request: Run CPU versions of applications for which GPU versions are available - yes/no (Message 60501)
Posted 2 Dec 2013 by Profile (retired account)
Post:
Thanks again, Richard and Len LE/GE. Now everything works as intended. And I trust we will still read you on some other project forum then, Richard?

Here are the two sections for nbody single-threaded and mt application again, which now work fine for me (on a quadcore with HT enabled):

<app>
<name>milkyway_nbody</name>
</app>
<file_info>
<name>milkyway_nbody_1.38_windows_x86_64.exe</name>
<executable/>
</file_info>
<file_info>
<name>libgomp_64-1_nbody_1.38.dll</name>
<executable/>
</file_info>
<file_info>
<name>pthreadGC2_64_nbody_1.38.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway_nbody</app_name>
<version_num>138</version_num>
<api_version>6.13.0</api_version>
<flops>4.0e10</flops>
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>
<cmdline></cmdline>
<file_ref>
<file_name>milkyway_nbody_1.38_windows_x86_64.exe</file_name>
<open_name>milkyway_nbody.exe</open_name>
<main_program/>
<copy_file/>
</file_ref>
<file_ref>
<file_name>libgomp_64-1_nbody_1.38.dll</file_name>
<open_name>libgomp_64-1.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>pthreadGC2_64_nbody_1.38.dll</file_name>
<open_name>pthreadGC2_64.dll</open_name>
<copy_file/>
</file_ref>
</app_version>

<app>
<name>milkyway_nbody</name>
</app>
<file_info>
<name>milkyway_nbody_1.38_windows_x86_64__mt.exe</name>
<executable/>
</file_info>
<file_info>
<name>libgomp_64-1_nbody_1.38.dll</name>
<executable/>
</file_info>
<file_info>
<name>pthreadGC2_64_nbody_1.38.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway_nbody</app_name>
<version_num>138</version_num>
<api_version>6.13.0</api_version>
<flops>1.6e11</flops>
<avg_ncpus>3.0</avg_ncpus>
<max_ncpus>4.0</max_ncpus>
<plan_class>mt</plan_class>
<cmdline>--nthreads 4</cmdline>
<file_ref>
<file_name>milkyway_nbody_1.38_windows_x86_64__mt.exe</file_name>
<open_name>milkyway_nbody.exe</open_name>
<main_program/>
<copy_file/>
</file_ref>
<file_ref>
<file_name>libgomp_64-1_nbody_1.38.dll</file_name>
<open_name>libgomp_64-1.dll</open_name>
<copy_file/>
</file_ref>
<file_ref>
<file_name>pthreadGC2_64_nbody_1.38.dll</file_name>
<open_name>pthreadGC2_64.dll</open_name>
<copy_file/>
</file_ref>
</app_version>




2) Message boards : Number crunching : Feature request: Run CPU versions of applications for which GPU versions are available - yes/no (Message 60492)
Posted 30 Nov 2013 by Profile (retired account)
Post:
Thanks for your suggestions.

Richard, I ommitted a few details in my second post to make it not too lengthy. :) The applications and the dlls were already on my harddrive before I started to test the app_info, so I made a backup and copied them into the projects data directory together with the app_info.xml. However, I do appreciate your comment and the direct download links.

ylixia, I see your point here. Seems to be a viable workaround, although not perfect, as Link pointed out.

I'm surprised that nobody posted a working app_info section for nbody. Hasn't anybody succeeded or does nobody care? ;)

Anyway, I've choosen another solution now and will run nbody inside a Virtual Machine. Performance hit should be small (is nbody using AVX if present?) and there's even an advantage: I can easily limit the cpu cores used by the mt app, which avoids kicking out long running tasks like RNA World on the host.

My feature request remains on the table, btw. ;)

Regards
3) Message boards : Number crunching : Feature request: Run CPU versions of applications for which GPU versions are available - yes/no (Message 60467)
Posted 27 Nov 2013 by Profile (retired account)
Post:
I don't take the lack of reply so far as a no. Maybe this can not be implemented on short notice. I would appreciate a reply, though, even a 'no'.

In the meantime I'm trying to load the cannon. Can somebody help me with a working N-Body Simulation section for an app_info.xml? The following app_info.xml was created based on the corresponding section in client_state.xml:

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>102</version_num>
<api_version>6.13.0</api_version>
<flops>2.5e11</flops>
<avg_ncpus>0.2</avg_ncpus>
<max_ncpus>0.97</max_ncpus>
<plan_class>opencl_amd_ati</plan_class>
<coproc>
<type>ATI</type>
<count>0.5</count>
</coproc>
<cmdline></cmdline>
<file_ref>
<file_name>milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe</file_name>
<main_program/>
</file_ref>
</app_version>
<app>
<name>milkyway_nbody</name>
</app>
<file_info>
<name>milkyway_nbody_1.38_windows_x86_64.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway_nbody</app_name>
<version_num>138</version_num>
<api_version>6.13.0</api_version>
<flops>2.7e10</flops>
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>
<cmdline></cmdline>
<file_ref>
<file_name>milkyway_nbody_1.38_windows_x86_64.exe</file_name>
<open_name>milkyway_nbody.exe</open_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libgomp_64-1_nbody_1.38.dll</file_name>
<open_name>libgomp_64-1.dll</open_name>
</file_ref>
<file_ref>
<file_name>pthreadGC2_64_nbody_1.38.dll</file_name>
<open_name>pthreadGC2_64.dll</open_name>
</file_ref>
</app_version>

<app>
<name>milkyway_separation__modified_fit</name>
</app>
<file_info>
<name>milkyway_separation__modified_fit_1.28_windows_x86_64__opencl_amd_ati.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway_separation__modified_fit</app_name>
<version_num>128</version_num>
<api_version>6.13.0</api_version>
<flops>1.8e11</flops>
<avg_ncpus>0.2</avg_ncpus>
<max_ncpus>0.97</max_ncpus>
<plan_class>opencl_amd_ati</plan_class>
<coproc>
<type>ATI</type>
<count>0.5</count>
</coproc>
<cmdline></cmdline>
<file_ref>
<file_name>milkyway_separation__modified_fit_1.28_windows_x86_64__opencl_amd_ati.exe</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>

The first and the last section for the two openCL applications is ok, the problem is the part in the middle, here in orange. When I include the references for the dll-files, as above, then I get the following missing-file-error, respectively, and BOINC is deleting the main program, but not the dll-files.

27.11.2013 14:10:01 | Milkyway@Home | Found app_info.xml; using anonymous platform
27.11.2013 14:10:01 | Milkyway@Home | [error] State file error: missing application file libgomp_64-1_nbody_1.38.dll


27.11.2013 14:21:22 | Milkyway@Home | Found app_info.xml; using anonymous platform
27.11.2013 14:21:22 | Milkyway@Home | [error] State file error: missing application file pthreadGC2_64_nbody_1.38.dll


In case I don't include the references to the dll-files, I get an error. Example: http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=619064814

Thanks in advance for any help here.

Regards

4) Message boards : Number crunching : Feature request: Run CPU versions of applications for which GPU versions are available - yes/no (Message 60457)
Posted 25 Nov 2013 by Profile (retired account)
Post:
Maybe this has been discussed before, if so, my apologies. However, a quick forum search returned nothing. Is there any chance you could implement a switch like "Run CPU versions of applications for which GPU versions are available - yes / no"? This switch is available at Einstein@home and it allows to opt into every application while only crunching those applications on CPU where no GPU version is available.

Currently I want to crunch again some MilkyWay@Home N-Body Simulation on CPU, but I have to opt out from MilkyWay@Home and Milkyway@Home Separation (Modified Fit) otherwise I'll get CPU workunits for such applications, too (which I do not want for obvious reasons). I am aware that I could use an app_info.xml, but this is a rather complicated approach if you only want to choose the standard apps you're running. Like "shooting with cannons at sparrows", as a german saying goes. ;) For allocating GPU resources I am using an app_config.xml currently.
5) Message boards : News : New MilkyWay Separation Modified Fit Runs (Message 60240)
Posted 28 Oct 2013 by Profile (retired account)
Post:
Two work packages failed, not only on my rigs:

ps_modfit_16_3s_128wrap_1_1382698503_393776 (this is, however, a _16_3s already)
de_modfit_15_3s_bpl_128wrap_2_1382698503_420968
6) Message boards : Number crunching : Updated GPU Requirements (Currently not supporting GPU tasks) (Message 58412)
Posted 26 May 2013 by Profile (retired account)
Post:
- In general laptop AMD GPUs do NOT support doubles despite similar branding (e.g. A Mobility Radeon 5870 is not the same as a normal Radeon 5870). Only the Mobility Radeon 48xx are the only current ATI laptop GPUs that have doubles.


The HD 7970M supports DP and runs Milkyway. Specs. are very similar to the desktop HD 7870 (Pitcairn) but clock is lower. It appears it has also the same 1/16 divider between SP and DP processing power.

Regards
7) Message boards : Number crunching : GPU Requirements (Message 56714)
Posted 4 Jan 2013 by Profile (retired account)
Post:
Michel,

MW requires Double Precision arithmetic. The HD 67xx series doesn't support DP.

Joe


I can confirm this.

The 6xxxx series cards that do support DP are limited to the 6970 and 6950, and the same for the 7xxx series GPUs.


Not quite correct, the 7750, 7770, 7850 & 7870 all support DP. Just at 1/16th the single precision rate.


The mobile HD 7970M aka 'Wimbledon XT' also supports DP and works with MW. Technically it is similar to a desktop HD 7870 aka 'Pitcairn' but GPU clock is only 850 instead of 1000 MHz. Here's a current valid result:

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=374577906

and from stderr:

Using device 0 on platform 0
Found 1 CL device
Device 'Pitcairn' (Advanced Micro Devices, Inc.:0x1002) (CL_DEVICE_TYPE_GPU)
Driver version: 1084.4 (VM)
Version: OpenCL 1.2 AMD-APP (1084.4)
Compute capability: 0.0
Max compute units: 20
Clock frequency: 850 Mhz
Global mem size: 2147483648
Local mem size: 32768
Max const buf size: 65536
Double extension: cl_khr_fp64

(...)

Estimated AMD GPU GFLOP/s: 170 SP GFLOP/s, 34 DP FLOP/s
Warning: Bizarrely low flops (34). Defaulting to 100
8) Message boards : News : Separation updated to 1.00 (Message 54473)
Posted 20 May 2012 by Profile (retired account)
Post:
I've updated all of the separation applications to 1.00. For changes people might care about,
    - AVX will be used if available on Linux and Windows (64 bit only for Windows)



Apparently no AVX with Bulldozer and Windows 8 Developer Preview x64. Anyone using an Intel CPU w/ Windows 8 Dev. Preview x64 or a Bulldozer w/ Windows 8 Consumer Preview x64? Is AVX used then?

I compared my PC with two of the top 20 computers, all three currently running MilkyWay@Home v1.02 (opencl_amd_ati):
My PC #447441, AMD FX-8120, Windows 8 DP x64, BOINC 7.0.25 (x64), AMD Radeon HD 79x0 series (Tahiti) (3072MB) driver: 1.4.1720 -> Using SSE4.1 path
#365115, Intel i7-3960X, Windows 7 Prof. x64 SP1, BOINC 7.0.27 (x64), [4] AMD Radeon HD 6900 series (Cayman) (2048MB) driver: 1.4.1720 -> Using AVX path
#340810, AMD FX-8120, Windows 7 Home Premium x64 SP1, BOINC 6.12.34 (x64), [4] AMD Radeon HD 6900 series (Cayman) (2048MB) driver: 1.4.1607 -> Using AVX path

Regards
9) Message boards : News : Separation updated to 1.00 (Message 53081)
Posted 11 Feb 2012 by Profile (retired account)
Post:
    - The old CAL version is gone; it's replaced with the OpenCL application



Question: Has the nVidia application also switched from CUDA to OpenCL now? The application name suggests it.

Seems to work fine on a GTX 560 Ti w/ Win7 64bit and BOINC 6.12.34 (x64), first result is valid. The ATI onboard GPU (SP) is not causing problems, server tells me: "An ATI GPU supporting double precision math is required" and "ATI GPU R600 (R38xx) does not support OpenCL".


<core_client_version>6.12.34</core_client_version>
<![CDATA[
<stderr_txt>
<search_application> milkyway_separation 1.02 Windows x86_64 double OpenCL </search_application>
Guessing preferred OpenCL vendor 'NVIDIA Corporation'
Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4'
Error reading astronomy parameters from file 'astronomy_parameters.txt'
Trying old parameters file
Using SSE3 path
Found 1 platform
Platform 0 information:
Name: NVIDIA CUDA
Version: OpenCL 1.1 CUDA 4.1.1
Vendor: NVIDIA Corporation
Extensions: cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_d3d9_sharing cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll
Profile: FULL_PROFILE
Using device 0 on platform 0
Found 1 CL device
Device 'GeForce GTX 560 Ti' (NVIDIA Corporation:0x10de) (CL_DEVICE_TYPE_GPU)
Driver version: 285.62
Version: OpenCL 1.1 CUDA
Compute capability: 2.1
Max compute units: 8
Clock frequency: 1645 Mhz
Global mem size: 2147483648
Local mem size: 49152
Max const buf size: 65536
Double extension: cl_khr_fp64
Build log:
--------------------------------------------------------------------------------

ptxas info : Compiling entry function 'probabilities' for 'sm_21'
ptxas info : Function properties for probabilities
80 bytes stack frame, 76 bytes spill stores, 76 bytes spill loads
ptxas info : Used 62 registers, 100 bytes cmem[0], 56 bytes cmem[16]
--------------------------------------------------------------------------------
Build log:
--------------------------------------------------------------------------------

ptxas info : Compiling entry function 'summarization' for 'sm_21'
ptxas info : Function properties for summarization
0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
ptxas info : Used 13 registers, 2048+0 bytes smem, 48 bytes cmem[0]
--------------------------------------------------------------------------------
Estimated Nvidia GPU GFLOP/s: 842 SP GFLOP/s, 105 DP FLOP/s
Using a target frequency of 60.0
Using a block size of 4096 with 17 blocks/chunk
Using clWaitForEvents() for polling with initial wait of 12 ms (mode 0)
Range: { nu_steps = 640, mu_steps = 1600, r_steps = 1400 }
Iteration area: 2240000
Chunk estimate: 32
Num chunks: 33
Chunk size: 69632
Added area: 57856
Effective area: 2297856
Initial wait: 12 ms
Integration time: 542.787896 s. Average time per iteration = 848.106088 ms
Integral 0 time = 546.308173 s
Running likelihood with 109999 stars
Likelihood time = 2.219000 s
<background_integral> 0.000133824638386 </background_integral>
<stream_integral> 6.443820669765729 875.872833232415470 </stream_integral>
<background_likelihood> -2.940328087823295 </background_likelihood>
<stream_only_likelihood> -83.771552699247522 -7.323208286845183 </stream_only_likelihood>
<search_likelihood> -2.935410274019145 </search_likelihood>
11:38:38 (3120): called boinc_finish

</stderr_txt>
]]>




©2024 Astroinformatics Group