Raistmer, guys, after a few years of having it sit on the shelf while waiting for the prices of the processors I wanted to become more 'reasonable', I finally got the $ together and bought the ones I had planned to use since the beginning. It is an EVGA SR-2, with a pair of X5690 CPU's, (currently with, for testing purposes) 3 GTX950's and a GTX750 with 48 gb of RAM running on a pair of Intel 480Gb SSD's in a RAID 1 config. The system is ID: 7990258 - CrunchMonster.

I spent much of yesterday building, loading, configuring and updating it, and finally early this morning turned on the spigot and let the WU's flow! Just before that I spent a bit of time reading the current posts on how to configure this to provide the most efficient crunching and utilize the hardware optimally, including Can I further optimize multiple GPU calculations? and my post Getting back into it, advice appreciated, as well as this one.

I have set it to <count>0.5, and set the <avg_ncpus>0.35</avg_ncpus> and <max_ncpus>0.35</max_ncpus>, and was reading about polling as well, though haven't implimented it yet, don't want to change too much at once, before consulting the Oracle (the brain trust here on the forum!).

As currently configured, it is running 24 CPU tasks, which I am sad to say seem to be averaging over 3 hours, which if I remember correctly is Much, Much longer than my CAD system I just built (ID: 7949285 Zoom-PC). Taking a quick look at that one, it appears that GPU tasks have a Run time (sec) averaging 750-1500 and the CPU time (sec) of 175-320 and the CPU tasks Run time is 5400-5700 and CPU time 5100-5500 (with a few in the 7500 range).

For the new system, (so far, as only 21 have validated as I write this) GPU tasks average Run time 1500-2500 & CPU time of 460-680, and CPU tasks Run time 11,000-12,500 CPU time of 9,700-10300! I know there is a few gen difference between the CPU's, and also between the 980's/970 in the former and the 950's/750 in the latter, but it seems that there might be something more at play here? Could someone with better insight than I take a look at them and see if there is anything that might stand out as a configuration Gotcha that I can adjust to help optimize this new one for a little better performance? The video cards are slightly O/C'd (around 10%), otherwise everything else is currently running stock.

Thanks again, guys!

Looking at your host 7990258. You may want to reduce the number of CPU tasks. When I was running a dual Xeon E5645 system I found 20 threads of MB ran much better than 24. When we had a lot of VLARs running 24 of them at once made the system less responsive. That should also help your GPUs. As their CPU time seems high to me.

Currently with my 750 ti I am only using processpriority = high in my mbcuda.cfg. I did try pfblockspersm = 8 pfperiodsperlaunch = 200, but found those values were slower than default in my configuration. I figured I might giving the -polling option a try next to see how it fairs.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Opps that is GPU User Group

Here:

Mark, I don't see you making use of the mbccuda.cfg file. Someone correctly if I'm wrong, but I think you can boost performance there, both for you 780 and 980, they are running defaults.


For my 750 Ti I have:

SETI@home using CUDA accelerated device GeForce GTX 750 Ti
mbcuda.cfg, processpriority key detected
mbcuda.cfg, Global pfblockspersm key being used for this device
pulsefind: blocks per SM 16
mbcuda.cfg, Global pfperiodsperlaunch key being used for this device
pulsefind: periods per launch 400
Priority of process set to ABOVE_NORMAL successfully

You have:

SETI@home using CUDA accelerated device GeForce GTX 780
pulsefind: blocks per SM 4 (Fermi or newer default)
pulsefind: periods per launch 100 (default)
Priority of process set to BELOW_NORMAL (default) successfully

My times seem to be similar to yours, but I think you should be outperforming me noticeably.

I don't suppose there would be any way for you to translate that into english for me, would there? ;-)

---

Yes

Go into setiathome folder.

Find the mbcuda.cfg and right click.

Open with notepad.

At the bottom of the file you will see 3 lines that begin with ;

The first say processpriority
the second say pfblockspersm
and the last say pfperiodsperlaunch

You want to remove the ; so there is nothing before each of those words, that allows those commands to be executed.

at the end of each of those lines, is a either a word or value

what you want to do is change each to the following, making sure there is only 1 space between the word and symbols

as follows

processpriority = high
pfblockspersm = 8
pfperiodsperlaunch = 400

You can just copy the above and replace the last 3 lines in your mbcuda.cfg

The values are less than what Brent posted as those were for a Nvidia 980, for the 750Ti, you
should go with a lessor value just to make sure it doesn't lock up the system

Thank You! I will do that this evening, after my GPU task cache clears out, as I set it to NNTs earlier today, to hopefully be able to see in the tasks the differences the changes make. Though with the variability of them, that might not be apparent in such a small sample size, but it's worth a shot. Also, from the initial cache I received, it seems that I was overloaded on the GPU side, because just checking now, I have only 4 CPU tasks left to finish, and the rest of the page full is GPU, which look to be taking about an hour and 40 mins or so each currently. It will be interesting to see how the last bunch of them compares in processing time and such to the earlier ones when all the CPU's were being hogged, and see how much of a difference, if any, it made in their processing times. I really appreciate all your help guys, thanks!

---

Changing mbcuda.cfg shouldn't affect anything, it should read the file when the next tasks start.

I don't believe it requires a BOINC restart.

it does not require restart

when the next task starts, it will read the new command

Ok, added that last nite, didn't notice anything different while watching it, but it's probably not noticeable, it's a background process?

This morning I also added an app_config file to the setiathome dir:

<app_config>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>

As soon as I added that and had it re-read the config files, it blocked off 8 CPU cores for the GPUs, is this excessive, and might it make sense to back it off to 0.5 for the CPU usage, to save one CPU core for each physical GPU, instead of each GPU task that is running? And, should it have a similar section for the v7 as well?

As for my app_info file, it is:

<app_name>astropulse_v7</app_name>
<version_num>705</version_num>
<platform>windows_x86_64</platform>
<avg_ncpus>0.35</avg_ncpus>
<max_ncpus>0.35</max_ncpus>
<plan_class>cuda_opencl_cc1</plan_class>
<cmdline></cmdline>
<coproc>
<type>CUDA</type>
<count>0.50</count>

where I have set all the avg_ncpus and max_ncpus in the file to 0.35, and the count to 0.50, is this appropriate?

---

Keep the fractions "simple" - if you want to run 3 tasks per GPU use 0.33 in both app_config and app_info files.
1 CPU per MB task is definitely far too much, drop that to something in the range 0.04 to 0.075. AP on the other had do need more of a contribution from the CPU, so increasing this by a factor of ten (range 0.4 to 0.75).

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Change your cpu_usage to something like 0.25

Yes I know people say 0.04 but that is for astropulses that don't really use all that much CPU.

MB actually use anywhere from 0.2 to 0.35 of a CPU depending on how many instances you are running on the card at that time.

How do you know how much to reserve?

I use something called BoincTasks.

It shows you how much cpu the work unit is actually using.

So add up how much cpu is being used by all of the work units and make the value just enough to be greater than the next whole number.

What??

ie if you have 3 GPUs running 2 at a time and each MB is using 0.35 then 3 GPU x 2 work is 6 . 6 x 0.35 is 2.1 so you effectively have reserved 2 cores for those 6 work units.

If you have 4 GPU and 2 work units then 8 x 0.35 is 2.8 so only 2 again so you might want to increase the value to 0.38 which give 8 x 0.38 is 3.04 so it will reserve 3 cores for those 8 work units.

It rounds down, not up.

Good luck

Edit..

See new post about adding an extra core for multiple GPU systems.

Thanks for the advice, I'll make some adjustments this morning. Was just taking a look at some of the results after validation, those CPU tasks are taking 3-3 1/2 hours.. Ouch. I think once I get this dialed in a little bit better, I am going to next try overclocking the motherboard a little bit, I've seen some people post some settings that have worked for them that will OC it a bit, but still be stable under air, and keep the temps in the 60* range, which I feel is pretty reasonable and I don't feel like setting up a water cooler. Not sure how much difference it will make, but 3+ hours seems pretty crazy for a relatively high end CPU.

---

Sorry, early morning here. No coffee yet.

Forgot to add if you are running more than 1 GPU in a machine, then you want to add 1 cpu core to help feed those GPUs so change the percentage so that the total is 1 more than what is actually required.

going back to my example if you see that mb actually is using 0.23 and you are running 8 that is 1.84 core so it rounds down to 1.

Changing the value to .25 with 8 gives 2 core but you need that extra core to feed those 4 gpus so changing the value to 0.40 give 3.2 or 3 cores.

Bumped up all the 0.35's to 0.38 in the app_info file, I'll restart BOINC and let it run for a while and see how it goes. Oh, one question I've had for a long time but have neglected to ask, actually ever since they moved the message tab to the Event Log (have never liked that or understood why they had to change it, I want it all in one place, in one window, as another tab like it used to be. Oh well.) Is there a way to increase the size of it, so it will show a longer time frame of events? Not sure how long or many it currently displays, but I would like to make it much larger. I have a 400 gig HD, how much room can it take up? lol

Here is the startup after changing it, just for info in case anyone wanted to see what it said:

4/30/2016 9:16:23 AM | | Starting BOINC client version 7.6.22 for windows_x86_64
4/30/2016 9:16:23 AM | | log flags: file_xfer, sched_ops, task
4/30/2016 9:16:23 AM | | Libraries: libcurl/7.45.0 OpenSSL/1.0.2d zlib/1.2.8
4/30/2016 9:16:23 AM | | Data directory: C:\ProgramData\BOINC
4/30/2016 9:16:23 AM | | Running under account Me
4/30/2016 9:16:24 AM | | CUDA: NVIDIA GPU 0: GeForce GTX 950 (driver version 364.72, CUDA version 8.0, compute capability 5.2, 2048MB, 1939MB available, 1828 GFLOPS peak)
4/30/2016 9:16:24 AM | | CUDA: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 364.72, CUDA version 8.0, compute capability 5.0, 2048MB, 1960MB available, 1606 GFLOPS peak)
4/30/2016 9:16:24 AM | | CUDA: NVIDIA GPU 2: GeForce GTX 950 (driver version 364.72, CUDA version 8.0, compute capability 5.2, 2048MB, 1939MB available, 2061 GFLOPS peak)
4/30/2016 9:16:24 AM | | CUDA: NVIDIA GPU 3: GeForce GTX 950 (driver version 364.72, CUDA version 8.0, compute capability 5.2, 2048MB, 1939MB available, 2158 GFLOPS peak)
4/30/2016 9:16:24 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 950 (driver version 364.72, device version OpenCL 1.2 CUDA, 2048MB, 1939MB available, 1828 GFLOPS peak)
4/30/2016 9:16:24 AM | | OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 364.72, device version OpenCL 1.2 CUDA, 2048MB, 1960MB available, 1606 GFLOPS peak)
4/30/2016 9:16:24 AM | | OpenCL: NVIDIA GPU 2: GeForce GTX 950 (driver version 364.72, device version OpenCL 1.2 CUDA, 2048MB, 1939MB available, 2061 GFLOPS peak)
4/30/2016 9:16:24 AM | | OpenCL: NVIDIA GPU 3: GeForce GTX 950 (driver version 364.72, device version OpenCL 1.2 CUDA, 2048MB, 1939MB available, 2158 GFLOPS peak)
4/30/2016 9:16:24 AM | SETI@home | Found app_info.xml; using anonymous platform
4/30/2016 9:16:24 AM | | Host name: CrunchMonster
4/30/2016 9:16:24 AM | | Processor: 24 GenuineIntel Intel(R) Xeon(R) CPU X5690 @ 3.47GHz [Family 6 Model 44 Stepping 2]
4/30/2016 9:16:24 AM | | 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 aes syscall nx lm vmx smx tm2 dca pbe
4/30/2016 9:16:24 AM | | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
4/30/2016 9:16:24 AM | | Memory: 47.99 GB physical, 95.98 GB virtual
4/30/2016 9:16:24 AM | | Disk: 446.90 GB total, 339.72 GB free
4/30/2016 9:16:24 AM | | Local time is UTC -5 hours
4/30/2016 9:16:24 AM | SETI@home | Found app_config.xml
4/30/2016 9:16:24 AM | | Config: use all coprocessors
4/30/2016 9:16:25 AM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 7990258; resource share 100
4/30/2016 9:16:30 AM | SETI@home | General prefs: from SETI@home (last modified 03-Apr-2013 23:59:56)
4/30/2016 9:16:30 AM | SETI@home | Computer location: home
4/30/2016 9:16:30 AM | SETI@home | General prefs: no separate prefs for home; using your defaults
4/30/2016 9:16:30 AM | | Preferences:
4/30/2016 9:16:30 AM | | max memory usage when active: 24571.59MB
4/30/2016 9:16:30 AM | | max memory usage when idle: 46686.01MB
4/30/2016 9:16:30 AM | | max disk usage: 100.00GB
4/30/2016 9:16:30 AM | | (to change preferences, visit a project web site or select Preferences in the Manager)
4/30/2016 9:16:30 AM | SETI@home | Sending scheduler request: To report completed tasks.
4/30/2016 9:16:30 AM | SETI@home | Reporting 1 completed tasks
4/30/2016 9:16:30 AM | SETI@home | Not requesting tasks: "no new tasks" requested via Manager
4/30/2016 9:16:31 AM | SETI@home | Scheduler request completed


*edit* I just checked, I have 20 out of 24 cores crunching CPU tasks right now, so it should be pretty happy at this point?

---

Let it run for a while, see what your times do.

Don't know about changing the size of the event log

Is there a way to increase the size of it, so it will show a longer time frame of events?

You will need to create/modify a cc_config.xml file. Go here and scroll to the Options section (middle of page). It's called:

 <max_event_log_lines>N</max_event_log_lines>
    Maximum number of lines to display in BOINC Manager's Event Log window (default 2000, 0 means no limit). New in 7.1.2

Remember, each time BOINC is closed, the Event Log screen will start anew. However, you can also increase the size of the log file that is maintained on the drive located at (in Win7)

C:\ProgramData\BOINC\stdoutdae.txt

Default size = 2 MB
This option is four items below the above option, and is called:

<max_stdout_file_size>N</max_stdout_file_size>

If I remember correctly, once the max file size has been reached, BOINC will rename the file to "stdoutdae.old", and create a new "stdoutdae.txt" file. The process repeats, but the stdoutdae.old will be overwrittien. You can keep the history by renaming it (ex. stdoutdae[_DATE].txt). FYI: This is done automatically by the BOINC add-on SETISpirit. Here is a thread on this add-on.

Change your cpu_usage to something like 0.25

Yes I know people say 0.04 but that is for astropulses that don't really use all that much CPU.

MB actually use anywhere from 0.2 to 0.35 of a CPU depending on how many instances you are running on the card at that time.

I've never run AP, but in all the years I've run MB the only time I've had to reserve more CPU time is in the last week or so since making use of Tbar's suggestion of using -poll.
Without that option, MB WUs on CUDA just don't use much CPU time, even when running 3 at a time on 2 video cards.

cpu_usage of 0.04 is appropriate for CUDA MB work, even with multiple WUs on multiple cards.

EDIT- with 4 GPUs it may be necessary to reserve 1 CPU core.
I personally use Process Explorer to see just how much CPU time is being used.
If reserving a core results in more CPU time going to the CUDA applications & no System Idle resulting, then it's worth it. If it doesn't, it's better to have that CPU crunching CPU work than being idle.

---

Grant
Darwin NT

Half a day, took longer than I thought it would...lol

Al, the choice is yours.

I only give you options.

Was just taking a look at some of the results after validation, those CPU tasks are taking 3-3 1/2 hours.. Ouch.

Why ouch?
I'd expect most of those to be VLARs, so 3-3.5 hours are fairly reasonable times, given that particular CPU doesn't support AVX and has to run SSSE3 code.

My Core 2 Duo at 2.4GHz takes 5.5-6hours to process Guppies, mostly because it has 2 GPUs running 2 WUs at a time which takes CPU time away from the CPU WU processing. Reserving 1 CPU for the GPUs doesn't improve their processing time, and while it does improve CPU processing time the loss of 1 CPU core (out of 2) results in less work being done over all.

You could reserve a CPU core or 2 for your GPUs, but I don't know if it would improve your GPU crunching times. And while it would improve the CPU crunching times there's a good chance the improvement in time won't offset the loss of the 2 cores so your actual work done per hour may decline, not improve.

---

Grant
Darwin NT

First of all, thanks for the all the ideas/suggestions. Here is my cc_config file currently:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<save_stats_days>10000</save_stats_days>
<max_event_log_lines>0</max_event_log_lines>
<max_stdout_file_size>500</max_stdout_file_size>
</options>
</cc_config>

I think that pretty much covers my questions about my settings/options, though after thinking about it, I should probably reduce the max size of my stdout file, because at some point it will be too large for any editor to open, and if I would make a guess, it would be _way_ below 500 meg... I think I'll try 20 and see how that goes, because it is just a text file, and that is a heck of a lot of text in one place.

I'm not sure about the core reserve business right now, it is running on all CPU cores except one being reserved for each physical GPU. Is that a waste of CPU computing potential? Possibly, but I am going to let it run as currently configured and see how things shake out before tweeking anything more.

It seems to be happily running, not sure if any task times are dropping at all, I can check that later, and I guess it is true that this isn't the latest proc, so maybe the CPU processing times aren't out of line. I was just comparing them to my i7, which probably isn't fair to do as it it a pretty top of the line current generation, so I'll be happy that it is happy at this point.

Now to just get a few thousand tasks under it's belt, and then start adjusting a thing here and there, one at a time, slowly... Thanks again guys!

---

Ok, I seem to be fighting the app_info.xml battle again with a freshly loaded computer. I changed the count to .5 to run 2 tasks per GPU, and I was trying to save at least 3 of the 8 cores for the GPUs, but whatever I change in the x.x values in the <gpu_usage>x.x</gpu_usage> and <cpu_usage>x.x</cpu_usage> lines, it always gives me the .04 CPU and .5 GPU while running tasks. I make the changes, save the file as a .xml, and then close and restart BOINC. It's like it is ignoring the changes, if not the file, I don't know enough to tell. Here is the startup log if it has any clues:

6/5/2016 9:17:54 AM |  | Starting BOINC client version 7.6.22 for windows_x86_64
6/5/2016 9:17:54 AM |  | log flags: file_xfer, sched_ops, task
6/5/2016 9:17:54 AM |  | Libraries: libcurl/7.45.0 OpenSSL/1.0.2d zlib/1.2.8
6/5/2016 9:17:54 AM |  | Data directory: C:\ProgramData\BOINC
6/5/2016 9:17:54 AM |  | Running under account Flash
6/5/2016 9:17:55 AM |  | CUDA: NVIDIA GPU 0: GeForce GTX 950 (driver version 368.22, CUDA version 8.0, compute capability 5.2, 2048MB, 1940MB available, 2061 GFLOPS peak)
6/5/2016 9:17:55 AM |  | CUDA: NVIDIA GPU 1: GeForce GTX 670 (driver version 368.22, CUDA version 8.0, compute capability 3.0, 2048MB, 1959MB available, 2915 GFLOPS peak)
6/5/2016 9:17:55 AM |  | CUDA: NVIDIA GPU 2: GeForce GTX 670 (driver version 368.22, CUDA version 8.0, compute capability 3.0, 2048MB, 1884MB available, 2915 GFLOPS peak)
6/5/2016 9:17:55 AM |  | CUDA: NVIDIA GPU 3: GeForce GTX 950 (driver version 368.22, CUDA version 8.0, compute capability 5.2, 2048MB, 1940MB available, 2158 GFLOPS peak)
6/5/2016 9:17:55 AM |  | OpenCL: NVIDIA GPU 0: GeForce GTX 950 (driver version 368.22, device version OpenCL 1.2 CUDA, 2048MB, 1940MB available, 2061 GFLOPS peak)
6/5/2016 9:17:55 AM |  | OpenCL: NVIDIA GPU 1: GeForce GTX 670 (driver version 368.22, device version OpenCL 1.2 CUDA, 2048MB, 1959MB available, 2915 GFLOPS peak)
6/5/2016 9:17:55 AM |  | OpenCL: NVIDIA GPU 2: GeForce GTX 670 (driver version 368.22, device version OpenCL 1.2 CUDA, 2048MB, 1884MB available, 2915 GFLOPS peak)
6/5/2016 9:17:55 AM |  | OpenCL: NVIDIA GPU 3: GeForce GTX 950 (driver version 368.22, device version OpenCL 1.2 CUDA, 2048MB, 1940MB available, 2158 GFLOPS peak)
6/5/2016 9:17:55 AM | SETI@home | Found app_info.xml; using anonymous platform
6/5/2016 9:17:55 AM |  | Host name: FlashFlyer
6/5/2016 9:17:55 AM |  | Processor: 8 GenuineIntel        Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz [Family 6 Model 58 Stepping 9]
6/5/2016 9:17:55 AM |  | 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 aes f16c rdrandsyscall nx lm avx vmx tm2 pbe fsgsbase smep
6/5/2016 9:17:55 AM |  | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
6/5/2016 9:17:55 AM |  | Memory: 31.96 GB physical, 63.91 GB virtual
6/5/2016 9:17:55 AM |  | Disk: 223.47 GB total, 139.02 GB free
6/5/2016 9:17:55 AM |  | Local time is UTC -5 hours
6/5/2016 9:17:55 AM |  | Config: event log limit disabled
6/5/2016 9:17:55 AM |  | Config: use all coprocessors
6/5/2016 9:17:55 AM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 8017700; resource share 100
6/5/2016 9:17:55 AM | SETI@home | General prefs: from SETI@home (last modified 03-Apr-2013 23:59:56)
6/5/2016 9:17:55 AM | SETI@home | Computer location: home
6/5/2016 9:17:55 AM | SETI@home | General prefs: no separate prefs for home; using your defaults
6/5/2016 9:17:55 AM |  | Reading preferences override file
6/5/2016 9:17:55 AM |  | Preferences:
6/5/2016 9:17:55 AM |  | max memory usage when active: 16361.52MB
6/5/2016 9:17:55 AM |  | max memory usage when idle: 31086.89MB
6/5/2016 9:17:55 AM |  | max disk usage: 100.00GB
6/5/2016 9:17:55 AM |  | (to change preferences, visit a project web site or select Preferences in the Manager)
6/5/2016 9:17:55 AM | SETI@home | Sending scheduler request: To fetch work.
6/5/2016 9:17:55 AM | SETI@home | Requesting new tasks for CPU and NVIDIA GPU
6/5/2016 9:17:57 AM | SETI@home | Scheduler request completed: got 0 new tasks
6/5/2016 9:17:57 AM | SETI@home | Not sending work - last request too recent: 213 sec

Thanks for your thoughts guys.

---

Where are you placing your xml file to?

Did you use notepad?

Did you save it as *.* and not as .txt