This is my current app_config.xml file (supporting WCG > hcc1)...
hcc1
10
0.20
0.20
I am running the E@H Binary Radio Pulsar Search (Arecibo) (opencl-ati)
How do I modify my app_config.xml file to best support this effort?
The current configuration is running 0.5C + 1 ATI GPU. I have two ATI GPUs and I'd like to restrict E@H to the use of just one GPU for the time being until HCC completes its work in about one month.
I modified GPU utilization factor of BRP apps to 0.5, but it doesn't seem to have had any effect.
Well, one of many perhaps :-). There's quite a few new ones joining every day!
Quote:
How do I modify my app_config.xml file to best support this effort?
You don't really need an app_config.xml as the defaults of 0.5C + 1 ATI GPU are really what is needed for best performance. If you only want to use 1 GPU, check out the option in BOINC client configuration. I haven't actually used this before but have a read and see for yourself if it will do what you need.
Quote:
I modified GPU utilization factor of BRP apps to 0.5, but it doesn't seem to have had any effect.
You actually have to download new work before the setting gets communicated to your BOINC client. All tasks in your cache will then use the new setting immediately. When two simultaneous tasks are running, one CPU core will be excluded and this will most likely give you the best GPU performance.
The only downside from this is that BOINC will still fetch CPU tasks (if you allow CPU crunching) as if all cores were still crunching. Over time, you tend to get too many CPU tasks. This was affecting one of my ATI endowed hosts at one point and my first workaround was to set a very low cache setting until the excess tasks had been worked off and then restore the desired cache until the problem returned.
Once the app config mechanism was released, I was able to solve the issue completely by using it to change the above-mentioned default to 0.45 CPUs + 1 ATI GPU. That allowed all CPUs to crunch with 2 GPU tasks but the GPU performance was significantly lower. So I used the local prefs through the GUI to exclude a core (75% for a quad core host). Since the client now knows that it's only using 3 cores for crunching, there's no over fetching of CPU work. The free core is available to support the ATI GPU so its performance is fully restored.
If you decide to try setting up an app_config.xml for E@H, the documentation is here. Note that this file will be quite separate from your WCG one and it will need to go in your E@H project directory. If you also use a cc_config.xml to exclude a GPU for particular projects, that one needs to be in the BOINC client directory.
Gary, thanks for the input. If you are correct as I suspect you are, the answers to my questions will be read and used by many GPU users.
Question: What is the short name of the application Binary Radio Pular Search (Arecibo)that is needed for the app_colnfig.xml file? BRPS perhaps? Or, put another way, here can I find its short name?
Gary, thanks for the input. If you are correct as I suspect you are, the answers to my questions will be read and used by many GPU users.
Question: What is the short name of the application Binary Radio Pular Search (Arecibo)that is needed for the app_colnfig.xml file? BRPS perhaps? Or, put another way, here can I find its short name?
Thanks
The short name is: einsteinbinary_BRP4
You can determine that names looking in the event log, when a task starts it says something like: "starting xxxx using einsteinbinary_BRP4 version zzzz".
The short name is: einsteinbinary_BRP4
You can determine that names looking in the event log, when a task starts it says something like: "starting xxxx using einsteinbinary_BRP4 version zzzz".
Feedback
You nailed it on the head, Horacio. Thank you.
I've got two 7770 GPU in this computer so to be fair I've assigned one GPU to WCG and one GPU to E@H. There is what the two app_config.xml files look like...
WCGs hcc1 file...
hcc1
5
0.20
0.20
E@Hs einsteinbinary_BRP4 file...
einsteinbinary_BRP4
5
0.20
0.20
This gives me five tasks running on both GPUs simultaneously and collectively using only one core. Time wise it takes an E@H task about two hours, but since five are running at the same time 120 minutes divided by 5 equals 20 minutes per task. I can live with that.
Of course the hcc1 five tasks take about 20 minutes. Divide 20 by 5 and you get 4 minutes per task.
I think by the time WCG finished with their HCC work, I'll be able to give XtremeSystems team a nice surprise. (:~)
Most likely just checking and telling you that you don't have an nVidia card so you won't get nVidia cuda tasks but as you know you do get CAL AMD Radeon HD 7700 (opencl-ati) tasks.
Maybe you have both boxes checked in your preferences.
This gives me five tasks running on both GPUs simultaneously and collectively using only one core. Time wise it takes an E@H task about two hours, but since five are running at the same time 120 minutes divided by 5 equals 20 minutes per task. I can live with that.
Looking at your results today I'd say this is not working well at all:
I'd dump the app_config and go back to running 2x. You also may need to allocate more than 1 core, possibly even 3 or 4 total on your i7 when running both Einstein and WCG on 2 GPUs. You can check this by monitoring your GPU usage in a program such as Afterburner. You'll get a lot more work done by sacrificing a CPU core or 2 (virtual in your case) and keeping the GPU at 93%+ efficiency.
Hello everyone, I'm the
)
Hello everyone,
I'm the new guy on the block at running E@H.
This is my current app_config.xml file (supporting WCG > hcc1)...
hcc1
10
0.20
0.20
I am running the E@H Binary Radio Pulsar Search (Arecibo) (opencl-ati)
How do I modify my app_config.xml file to best support this effort?
The current configuration is running 0.5C + 1 ATI GPU. I have two ATI GPUs and I'd like to restrict E@H to the use of just one GPU for the time being until HCC completes its work in about one month.
I modified GPU utilization factor of BRP apps to 0.5, but it doesn't seem to have had any effect.
Thanks for the assist.
Gandalf
RE: I modified GPU
)
This won't have any effect on tasks already in cache, only on newer tasks you download from now on.
RE: ... I'm the new guy on
)
Well, one of many perhaps :-). There's quite a few new ones joining every day!
You don't really need an app_config.xml as the defaults of 0.5C + 1 ATI GPU are really what is needed for best performance. If you only want to use 1 GPU, check out the option in BOINC client configuration. I haven't actually used this before but have a read and see for yourself if it will do what you need.
You actually have to download new work before the setting gets communicated to your BOINC client. All tasks in your cache will then use the new setting immediately. When two simultaneous tasks are running, one CPU core will be excluded and this will most likely give you the best GPU performance.
The only downside from this is that BOINC will still fetch CPU tasks (if you allow CPU crunching) as if all cores were still crunching. Over time, you tend to get too many CPU tasks. This was affecting one of my ATI endowed hosts at one point and my first workaround was to set a very low cache setting until the excess tasks had been worked off and then restore the desired cache until the problem returned.
Once the app config mechanism was released, I was able to solve the issue completely by using it to change the above-mentioned default to 0.45 CPUs + 1 ATI GPU. That allowed all CPUs to crunch with 2 GPU tasks but the GPU performance was significantly lower. So I used the local prefs through the GUI to exclude a core (75% for a quad core host). Since the client now knows that it's only using 3 cores for crunching, there's no over fetching of CPU work. The free core is available to support the ATI GPU so its performance is fully restored.
If you decide to try setting up an app_config.xml for E@H, the documentation is here. Note that this file will be quite separate from your WCG one and it will need to go in your E@H project directory. If you also use a cc_config.xml to exclude a GPU for particular projects, that one needs to be in the BOINC client directory.
Good luck with getting it all sorted :-).
Cheers,
Gary.
Gary, thanks for the input.
)
Gary, thanks for the input. If you are correct as I suspect you are, the answers to my questions will be read and used by many GPU users.
Question: What is the short name of the application Binary Radio Pular Search (Arecibo)that is needed for the app_colnfig.xml file? BRPS perhaps? Or, put another way, here can I find its short name?
Thanks
RE: Gary, thanks for the
)
The short name is: einsteinbinary_BRP4
You can determine that names looking in the event log, when a task starts it says something like: "starting xxxx using einsteinbinary_BRP4 version zzzz".
RE: The short name is:
)
Feedback
You nailed it on the head, Horacio. Thank you.
I've got two 7770 GPU in this computer so to be fair I've assigned one GPU to WCG and one GPU to E@H. There is what the two app_config.xml files look like...
WCGs hcc1 file...
hcc1
5
0.20
0.20
E@Hs einsteinbinary_BRP4 file...
einsteinbinary_BRP4
5
0.20
0.20
This gives me five tasks running on both GPUs simultaneously and collectively using only one core. Time wise it takes an E@H task about two hours, but since five are running at the same time 120 minutes divided by 5 equals 20 minutes per task. I can live with that.
Of course the hcc1 five tasks take about 20 minutes. Divide 20 by 5 and you get 4 minutes per task.
I think by the time WCG finished with their HCC work, I'll be able to give XtremeSystems team a nice surprise. (:~)
Referencing my Event
)
Referencing my Event log...
4/6/2013 10:57:26 PM | | Starting BOINC client version 7.0.56 for windows_x86_64
4/6/2013 10:57:26 PM | | log flags: file_xfer, sched_ops, task
4/6/2013 10:57:26 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
4/6/2013 10:57:26 PM | | Data directory: C:\ProgramData\BOINC
4/6/2013 10:57:26 PM | | Running under account Gandalf
4/6/2013 10:57:26 PM | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7 CPU 975 @ 3.33GHz [Family 6 Model 26 Stepping 5]
4/6/2013 10:57:26 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt nx lm vmx tm2 pbe
4/6/2013 10:57:26 PM | | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
4/6/2013 10:57:26 PM | | Memory: 20.00 GB physical, 39.99 GB virtual
4/6/2013 10:57:26 PM | | Disk: 119.14 GB total, 61.81 GB free
4/6/2013 10:57:26 PM | | Local time is UTC -7 hours
4/6/2013 10:57:26 PM | | CAL: ATI GPU 0: AMD Radeon HD 7700 series (Cape Verde) (CAL version 1.4.1741, 1024MB, 984MB available, 3200 GFLOPS peak)
4/6/2013 10:57:26 PM | | CAL: ATI GPU 1: AMD Radeon HD 7700 series (Cape Verde) (CAL version 1.4.1741, 1024MB, 984MB available, 3200 GFLOPS peak)
4/6/2013 10:57:26 PM | | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 7700 series (Cape Verde) (driver version 1084.4 (VM), device version OpenCL 1.2 AMD-APP (1084.4), 1024MB, 984MB available, 3200 GFLOPS peak)
4/6/2013 10:57:26 PM | | OpenCL: AMD/ATI GPU 1: AMD Radeon HD 7700 series (Cape Verde) (driver version 1084.4 (VM), device version OpenCL 1.2 AMD-APP (1084.4), 1024MB, 984MB available, 3200 GFLOPS peak)
4/6/2013 10:57:26 PM | | app version refers to missing GPU type NVIDIA
4/6/2013 10:57:26 PM | World Community Grid | Application uses missing NVIDIA GPU
4/6/2013 10:57:26 PM | Einstein@Home | Found app_config.xml
4/6/2013 10:57:26 PM | World Community Grid | Found app_config.xml
4/6/2013 10:57:26 PM | | Config: use all coprocessors
4/6/2013 10:57:26 PM | | Config: GUI RPCs allowed from:
4/6/2013 10:57:26 PM | | 192.168.1.100
4/6/2013 10:57:26 PM | | 192.168.1.102
4/6/2013 10:57:26 PM | | 192.168.1.103
4/6/2013 10:57:26 PM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 6983575; resource share 50
4/6/2013 10:57:26 PM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 1511352; resource share 100
4/6/2013 10:57:26 PM | Einstein@Home | General prefs: from Einstein@Home (last modified 06-Apr-2013 14:12:36)
4/6/2013 10:57:26 PM | Einstein@Home | Host location: none
4/6/2013 10:57:26 PM | Einstein@Home | General prefs: using your defaults
4/6/2013 10:57:26 PM | | Reading preferences override file
4/6/2013 10:57:26 PM | | Preferences:
4/6/2013 10:57:26 PM | | max memory usage when active: 20478.18MB
4/6/2013 10:57:26 PM | | max memory usage when idle: 20478.18MB
4/6/2013 10:57:26 PM | | max disk usage: 10.00GB
4/6/2013 10:57:26 PM | | (to change preferences, visit a project web site or select Preferences in the Manager)
4/6/2013 10:57:26 PM | | Not using a proxy
4/6/2013 10:57:27 PM | Einstein@Home | Restarting task h1_0484.80_S6GC1__S6BucketLVEa_484.913682292Hz_2184_1 using einstein_S6BucketLVE version 104 (SSE2) in slot 7
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0960128860447201107291318_0 using hcc1 version 705 in slot 3
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0960128860435201107291317_0 using hcc1 version 705 in slot 5
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0960128860441201107291318_0 using hcc1 version 705 in slot 8
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0960128860433201107291317_0 using hcc1 version 705 in slot 1
4/6/2013 10:57:27 PM | Einstein@Home | Restarting task p2030.20121222.G203.03-00.12.S.b0s0g0.00000_2848_2 using einsteinbinary_BRP4 version 134 (opencl-ati) in slot 10
4/6/2013 10:57:27 PM | Einstein@Home | Restarting task p2030.20121220.G174.75-01.12.S.b6s0g0.00000_2904_0 using einsteinbinary_BRP4 version 134 (opencl-ati) in slot 9
4/6/2013 10:57:27 PM | Einstein@Home | Restarting task p2030.20121220.G174.75-01.12.S.b1s0g0.00000_3016_1 using einsteinbinary_BRP4 version 134 (opencl-ati) in slot 0
4/6/2013 10:57:27 PM | Einstein@Home | Restarting task p2030.20121220.G174.75-01.12.S.b0s0g0.00000_2104_0 using einsteinbinary_BRP4 version 134 (opencl-ati) in slot 2
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0960128860445201107291318_1 using hcc1 version 705 in slot 6
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129081183201107281333_1 using hcc1 version 705 (ati_hcc1) in slot 12
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129081025201107281335_1 using hcc1 version 705 (ati_hcc1) in slot 14
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129081179201107281333_0 using hcc1 version 705 (ati_hcc1) in slot 16
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129081053201107281335_1 using hcc1 version 705 (ati_hcc1) in slot 13
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129080947201107281337_1 using hcc1 version 705 (ati_hcc1) in slot 15
4/6/2013 10:57:27 PM | World Community Grid | Restarting task X0930129081091201107281334_0 using hcc1 version 705 (ati_hcc1) in slot 11
____________________________________________________________________________
Can you explain this outtake from the log...
4/6/2013 10:57:26 PM | | app version refers to missing GPU type NVIDIA
4/6/2013 10:57:26 PM | World Community Grid | Application uses missing NVIDIA GPU
Thanks
Most likely just checking and
)
Most likely just checking and telling you that you don't have an nVidia card so you won't get nVidia cuda tasks but as you know you do get CAL AMD Radeon HD 7700 (opencl-ati) tasks.
Maybe you have both boxes checked in your preferences.
RE: E@Hs
)
Looking at your results today I'd say this is not working well at all:
http://einsteinathome.org/host/6983575/tasks&offset=0&show_names=1&state=5&appid=0
I'd dump the app_config and go back to running 2x. You also may need to allocate more than 1 core, possibly even 3 or 4 total on your i7 when running both Einstein and WCG on 2 GPUs. You can check this by monitoring your GPU usage in a program such as Afterburner. You'll get a lot more work done by sacrificing a CPU core or 2 (virtual in your case) and keeping the GPU at 93%+ efficiency.
Beyond wrote:Looking at your
)
@Gandalf_the_Grey
Try these app_config settings
WCGs hcc1 file...
hcc1
4
0.25
0.75
E@Hs einsteinbinary_BRP4 file...
einsteinbinary_BRP4
2
0.5
0.5