Help optimizing app config file with Core i7 GTX750ti

Message boards : Number crunching : Help optimizing app config file with Core i7 GTX750ti
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 · Next

AuthorMessage
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1796960 - Posted: 18 Jun 2016, 0:36:12 UTC
Last modified: 18 Jun 2016, 0:46:25 UTC

So I have been banging my head around trying to find the best settings for my rigs.

3 dedicated SETI@home machines;

Core i7 3770, GTX750TI 2GB, Intel HD Graphics
Core i5 2400, GTX750 1GB
Core i3 3140t, GTX750 on the way, Intel HD Graphics

2 other part time seti machines:
Core i7 3770, R9 280x 3GB, Intel HD Graphics
Core i7 2600, HD6950 2GB,

As of now my app config file is this:

<app_config>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>

I am running 3xWUs but not sure if I am utilizing my CPU and GPU the best.

Also have it set to use 90% in BOINC manager.

How many WUs does a GTX750 seem to best at?
What CPU usage should I set my i7,i5,i3 config files?

Just want to make sure I am getting the most bang for my buck to help the SETI@home community.

Any help would be greatly appreciated.
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1796960 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1796975 - Posted: 18 Jun 2016, 1:17:39 UTC - in response to Message 1796960.  

<app_config>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>


Prior to work version 8 and greenbank data, this would have been an easy question to answer

It would have been 2 work units per card and possible 3 CPU per Quad core.

However, we now have new apps and new data. That being said. You need to take into account how much CPU each work unit will need.

I know several people here that run 750s so their input will be helpful, but until then.

I would suggest 2 work units per card (as the SoG app can take a full core for each work unit) and no more than 2 CPU work units.

If you start to see lagging or lock up then I would decrease the number of CPU work units to 1.

The <cpu_usage>.1</cpu_usage> doesn't seem to affect the breakthough listen data when they are crunched on the new app (OpenCl SoG) as they appear to need a full core.

Let us know how you do.

Zalster
ID: 1796975 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1796985 - Posted: 18 Jun 2016, 2:47:05 UTC - in response to Message 1796960.  

I run 2 750 Ti's in my systems.

First I would install Lunatics applications http://mikesworldnet.de/download.html

Then mb_cmdline_win_x86_SSE3_OpenCL_NV.txt put in:
-use_sleep -sbs 256 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp

(There are debates on the command line, but that works well for me)

ap_cmdline_win_x86_SSE2_OpenCL_NV.txt put in:
-use_sleep -unroll 10 -oclFFT_plan 256 16 512 -ffa_block 12288 -ffa_block_fetch 6144


<app_config>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.4</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>0.6</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>

</app_config>
ID: 1796985 · Report as offensive
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1797064 - Posted: 18 Jun 2016, 7:22:16 UTC
Last modified: 18 Jun 2016, 7:52:23 UTC

Currently running on my setup see how it goes:

**************************************************
Core i7 3770, R9 280x 3gb, Intel HD Graphics 4000,
90% CPU
**************************************************
Core i5 2400s, GTX 750ti 2gb,
60% CPU anything higher and it goes to 100% cpu in task manager
60% keeps it at 75%-90% in task manager
**************************************************
Core i3 4130t, GTX 750 1gb, Intel HD Graphics 4400
90% CPU
**************************************************
Core i7 2600, HD6950 2gb,
90% CPU
STRANGE issue, GPU load starts and stops, 0%-99% for about 10 seconds then stops? and ideas, using driver 15.7.1, supports OpenCL 1.2
**************************************************

<app_config>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.4</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>0.6</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>

</app_config>

I did not install lunatics lots i did not know what to check for install, when I did my intel gpu stopped responding and being recognized not sure what that is about.

I reformated with
Windows 7 PRO x64 SP1
disabled most services, firewall, security center, superfetch, media player sharing, system restore, themes.

Trying to keep task manager cpu at 90%
GPU seems maxed on all machines at 90-99%

I want max performance with out any lag or processes having to wait for work cause the machine is too busy.
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1797064 · Report as offensive
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1797319 - Posted: 19 Jun 2016, 4:39:26 UTC
Last modified: 19 Jun 2016, 4:43:25 UTC

UPDATE:

<app_config>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

</app_config>

Can you tell me with a chart what these numbers mean it would help, I see so many different results on my systems.

I know .5 gpu is two wus, and .33 is 3

cpu im CoNfUsEd....

Can you give me a little chart with a little reference.

app config
gpu
.1
.5
.33
etc...

cpu
.1
1
.04
.6
.4
etc.......

that would greatly help my configurations just kind of lost been gone for a while when system switched, so I am really really new to the new app config

what is recommended with i3, i5, i7 for cpu?

I posted my system specs above in the other posts, I wont repost incase that annoys people haha. all have 1 CPU and 1 GPU, plus some with 1 Intel GPU

Would it be wise to assign the Intel GPU less tasks than the GTX750 since the intel gpu is still using cpu resources?

thank you very much with any help

ITEHQ
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1797319 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1797322 - Posted: 19 Jun 2016, 5:13:58 UTC - in response to Message 1797319.  
Last modified: 19 Jun 2016, 5:17:51 UTC

<app_config>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
Can you tell me with a chart what these numbers mean it would help, I see so many different results on my systems.


As you mentioned, <gpu_usage> determines the number of WUs to run on a GPU.
1.0 for 1
0.5 for 2
0.33 for 3
0.25 for 4
etc.

<CPU_usage> determines how much of a CPU core to give to each WU being crunched on the GPU (it doesn't have anything to do with CPU crunching other than the fact that the more cores you give to the GPU, the less you have for CPU crunching).

0.5 means that with 2 GPU WUs being crunched, 1 CPU core will be reserved for both of them (ie no CPU crunching will occur using that core). So if there are 4 GPU WUs being crunched, 2 CPU cores will be reserved for them.
0.25 up to 4 GPU WUs can be crunched, and only 1 core will be reserved for them. Do 5 GPU WUs and 2 cores will be reserved.
1.0 means that each GPU WU being crunched will have it's own core; so 6 GPU WUs being done, 6 CPU cores reserved for them.


what is recommended with i3, i5, i7 for cpu?

It depends.
Most applications will require the use of 1 CPU core for each GPU WU being processed to get the absolute maximum performance.
Other applications require very little CPU time for good overall perfromance, so you will often see them with a <CPU_usage> value of 0.04.
So it depends on how much performance you want, how many WUs you run at a time on the GPU, and how many GPUs you have.

Generally, for maximum GPU performance you need to reserve 1 CPU for each WU you are crunching, and crunch multiple WUs on the GPU (how many is best, depends on the GPU; low end, mid range or highend).
However maximum performance on the GPU that provides your display can result in stuttering screen re-draws, lost characters while typing & general system sluggishness.


Would it be wise to assign the Intel GPU less tasks than the GTX750 since the intel gpu is still using cpu resources?

Generally it's best to disable iGPU processing, as the CPU has to share power, memory & heat resources with it. The end result being that both the CPU & iGPU output is much less than when running just one or the other.
If you disable iGPU crunching, and use that for you screen display, then you can run the addon GPUs at (or very close to) maximum possible output, and not have system usability issues.
Grant
Darwin NT
ID: 1797322 · Report as offensive
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1797714 - Posted: 21 Jun 2016, 5:04:51 UTC

UPDATE:

UPDATE:

<app_config>

<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.33</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

I turned the iGPU off in SETI@home settings on the web gui.

Seems to be performing really well, see numbers soon.
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1797714 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798455 - Posted: 24 Jun 2016, 17:27:39 UTC - in response to Message 1796985.  

Hello Brent,

I tried overnight your suggested command line settings on my main rig that has 2 GTX 750 Ti: ...
mb_cmdline_win_x86_SSE3_OpenCL_NV.txt put in:
-use_sleep -sbs 256 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp

(There are debates on the command line, but that works well for me)

ap_cmdline_win_x86_SSE2_OpenCL_NV.txt put in:
-use_sleep -unroll 10 -oclFFT_plan 256 16 512 -ffa_block 12288 -ffa_block_fetch 6144

In the morning, I found that my GPU WU being crunched had more than doubled in time (once they were completed) and that my RAC had slightly decreased instead of slowly increasing (since it hasn't reached its peak yet).
Obviously, a 7-hr shift is not long enough to know with any certainty if it was an improvement or not, especially since it was at the same time that a few AP WUs started coming through.

Q1. Are your 2 rigs with a GTX 750 Ti in each producing more than the Cuda 50 app with 2 WU/GPU?

Q2. Are there any command line settings that I should tweak when running 2 identical GTX 750 Ti?

Q3. How long should I run a test with new settings and what/where should I look to measure the diff?

Fyi, I ran the Lunatics v0.45 Beta3 setup 3.5 days ago with SoG and was allocating mostly 1 WU per GPU (except when I was at the rig to do a few 2 WU/GPU tests).
I haven't looked into the command line options yet as the learning curve is still increasing faster than my RAC!

Also, you might be interested in taking a look at rigs with identical CPUs crunching without GPUs. Here are the top performers (by RAC at the end of May when I imported host.gz into MS Access)
- i5-4460S: 7649851 7641587 7563528 7563352
- 4200+ [Family 15 Model 75 Stepping 2]: none with a RAC above 700

Any feedback is greatly appreciated.
Cheers,
Rob :-}
ID: 1798455 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798471 - Posted: 24 Jun 2016, 18:59:00 UTC - in response to Message 1797714.  

Hello ITEHQ,

Here's a few of my "lessons learned" during the last 2 months for (as you put it in your original post):
trying to find the best settings for my rigs

There's probably a few errors that others can point out and I am looking forward to their feedback.

The following is related to my experience with my primary rig that started in early May with the purchase of a Zotac GTX 750 Ti, and now trying to optimize with two identical GPUs (as per my previous post above).
1. Concentrate on 1 or 2 computers to optimize, or take very good notes as to what you did on each one (if your medium term memory isn't well above average);
forgetting to undo a small change from the night before can be a waste of a few days to come.

2. Learn which variables to look out for before making a change in order to have a baseline to compare to (I'm still learning on this one) ...and don't make changes on Mondays or Tuesday if your metric is based on a 24hr test that relies on data from the S@H servers (especially if you're expecting a big RAC increase).

3. Learn to use statistics websites such as: boincstats.com
...and use only the ones where you can figure out the age of the host data, such as: http://boincstats.com/en/stats/0/project/detail/ (see 2nd to last line).

4a. For CPUs and GPUs that are fairly recent, try to find "Top computers/Top Hosts" with the same CPU and GPU in order to compare yours to. I did this by trying to go through the top 1000 list in the stats section: https://setiathome.berkeley.edu/top_hosts.php
I was lucky as I found 2 identical rigs with the same S@H user who replied to a Private Message. (I didn't PM until last week to see if there was anything else that I hadn't considered that was an easy tweak.)
For your 8020966 ( i7-3770 with one GTX 750 Ti ), there's: 7147692 7836487 7210536 (based on my end of May host.gz import)

4b. If your CPU can collect more than 3.5k in credits per day, try to find a "Top computer" running without an added GPU. Currently, the RAC for the one in the last listed position has a RAC of ~3,800 (see: https://setiathome.berkeley.edu/top_hosts.php?sort_by=expavg_credit&offset=9980 )
Here are the top i7-3770 performers without an added GPU: 6841066 7379487
Their RAC is around 7K but I haven't looked on BoincStats.com to see if they are constant 24/7 crunchers.

5. Run your rig 24hrs/day for 3-4 days with the stock setup. Currently, on a GTX 750 Ti, you'll likely be running mostly the SoG app on the GPU, and your RAC should rise.

5b. Focus on the top crunchers using the same OS as yours (mostly Windows vs Mac vs Linux) and figure out which apps are used by the top crunchers identified above. It's likely they are using the stock BOINC/S@H installation or the Lunatics v0.44 (on Windows only) customized setup.

6. For nvidia cards, install Lunatics v0.44 with CUDA and let it run 1-2 days before trying to run more than 1 WU at a time. It will likely be a bit slower on the GPU but the CPU WU should be much faster.

6b. Configure: app_config.xml (as suggested by Brent above)...but without anything in the command line files. After a few days, your daily credits should be above or close to the same "Top computers" with the same CPU/GPU identified in item 4a above.

6c. learn about command lines (above my current knowledge)

7. Repeat above for your other rigs but with cutting out wait times from days to hrs.

8. Learn more from the forum posts and ask Qs (on topic) or start a new thread.

I hope I haven't missed something important. Please correct me if I have!

Cheers,
Rob :-}
ID: 1798471 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1798490 - Posted: 24 Jun 2016, 20:33:36 UTC - in response to Message 1798455.  

Rob,

I looked at some of the tasks you have completed and those command lines are not being applied to the tasks.

Make sure you have the correct names for them and use notepad to edit them. contents MUST be a single line with no carriage returns.
ID: 1798490 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798504 - Posted: 24 Jun 2016, 21:26:19 UTC - in response to Message 1798490.  

Hey Brent,
From what I can remember of yesterday before going to bed, I shut down Boinc and copied command lines into the specified files without a carriage return, and I also copied the provided content into app_config.xml
After starting Boinc, I did: Options | read config files
...and the GPUs started a 2nd WU for each device.

The only thing different is that I used: Notepad++
I made a copy of those 3 xml files before reverting back this morning and the only thing I can think of wrt the 2 command files is that I did not put a carriage return at the end of the provided line. Is that necessary?

Here are a few that took twice the avg time for their type:
- 13dc10ae.18834.4157.3.30.178_0
- blc2_2bit_guppi_57451_62689_HIP116936_0003.13801.416.17.26.90.vlar_0
I compared the last one to one of your guppi, and in mine there is no "Sleep() & ..." so obviously I did or didn't do something.
I'll try again tonight a few hours before hitting the sack.

If there's anything else you can think of, let me know,
R :-}
ID: 1798504 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1798508 - Posted: 24 Jun 2016, 21:42:44 UTC - in response to Message 1798504.  

Oh, you reverted back this morning, that would explain why they aren't being applied to what I looked at.

I don't see any command line applied to either of those 2 files.


The only thing different is that I used: Notepad++
I made a copy of those 3 xml files before reverting back this morning and

The command line files are .TXT not .XML
You don't need to read files/restart for them, they will be applied when the next task(s) start.

BTW, BoincTasks is a nice program for seeing runtime history.
ID: 1798508 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798573 - Posted: 25 Jun 2016, 1:44:48 UTC - in response to Message 1798508.  

Here's the 1st one I found: 15my10ab.25290.61567.8.35.154_0
that includes the "Sleep() & ..." this time around. YaY!

I'll leave it running overnight.
Fyi, the first few WU were running in parallel to 2 AP GPU WU that had started before your recommended change so anything sent before 2am UTC might be affected.

My initial observations with 2 WU/GPU as compared to 1WU/GPU are:
- the Arecibos are crunching in less than double the time; ~16mins before, and ~27mins now.
- the blc...guppi...vlar are running well over double the time. If I remember correctly: ~25-29mins before change, and now over 1 hr and they're only at ~75%; that's not so good :-/

I'm looking forward to trying your settings and Mr Kevvy's script to swap guppis on GPU with arecibos on CPU, see: https://setiathome.berkeley.edu/forum_thread.php?id=79836&postid=1798559

Let me know what you think,
Rob :-}
ID: 1798573 · Report as offensive
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1798585 - Posted: 25 Jun 2016, 4:08:32 UTC

UPDATE:

<app_config>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>.25</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
</app_config>

I am now running 4 gpu tasks, my thought process....

I am running several PCs:
Corei7 3770, R9 280x
Corei7 3770, GTX750ti
Corei7 2600, GTX750
Corei7 2600, HD6950 no working yet...

Decided to run 4 gpu tasks since the gtx750s have 5 compute unites, and I have 4 cores, plus 4 HT.

see how it does using 90% cpu.

What numbers allow me to setup higher than 4 wu. I want to max out my GPU wu with the amount of compute units the rx480 has....
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1798585 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1798587 - Posted: 25 Jun 2016, 4:40:47 UTC - in response to Message 1798585.  

What numbers allow me to setup higher than 4 wu. I want to max out my GPU wu with the amount of compute units the rx480 has....

Not much point IMHO.
As it is, running 4 WUs on the GTX 750Tis will most likely result in a significant drop in work output.
Prior to the Guppie WUs, it was generally best to run no more than 2WUs at a time as the increase in run time for shorties was so bad that it would result in less WUs done per hour even if there were only a few of them every few hours.
With the current lack of shorties & the introduction of the Guppies, 3WUs at a time appears to be giving the most WUs crunched per hour (although depending on the application & it's settings 4 may be OK, or you may be better off with only 2).
Grant
Darwin NT
ID: 1798587 · Report as offensive
Profile ITEHQ
Avatar

Send message
Joined: 10 Jun 16
Posts: 38
Credit: 1,810,243
RAC: 0
United States
Message 1798605 - Posted: 25 Jun 2016, 5:49:08 UTC
Last modified: 25 Jun 2016, 5:49:56 UTC

About 30-40 minutes for each GPU WU

About 1-2.5 hrs for each CPU WU
Monitoring the Universe for signs of intelligent life, as we have not found any on Earth yet (^_^)
ID: 1798605 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1798741 - Posted: 25 Jun 2016, 23:27:42 UTC - in response to Message 1798573.  

Rob, how many WUs are you running per card?
How many reserved cores per WU?
ID: 1798741 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798745 - Posted: 25 Jun 2016, 23:48:09 UTC - in response to Message 1798741.  

As I mentionned in a PM I sent you 1-2hrs ago, overnight I ran 2 WU/GPU with the suggested command lines.
In general, blc...guppi...vlars were taking 4-5 times longer (~2:20) than my previous settings of 1 WU/GPU (25-30 mins)...so I reverted back at around 2pm UTC (10am Eastern) to: 1 WU/GPU with no command lines for MB and AP.

To answer your last Q, the app_config was as per your suggestion above.

I think a new thread might be needed (if there isn't one already) since this seems more about "Lunatics v0.45 Beta3 with command lines" than ITEHQ's initial post.
ID: 1798745 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1798780 - Posted: 26 Jun 2016, 4:52:56 UTC - in response to Message 1798745.  

Rob, try running cpu_useage 1.0 for awhile, see if the CPU is the bottleneck.

Also remember that running guppie task with MB tasks will slow the MB tasks down greatly. Try suspending task so you can run them in 'same' pair to get a comparison.
ID: 1798780 · Report as offensive
Profile Stubbles
Volunteer tester
Avatar

Send message
Joined: 29 Nov 99
Posts: 358
Credit: 5,909,255
RAC: 0
Canada
Message 1798787 - Posted: 26 Jun 2016, 6:25:22 UTC - in response to Message 1798780.  

Overnight Trial #3 started!

Luckely, I had 3 times more Arecibos assigned to GPU than guppis so I supended the guppis for the night.

As I wasn't too sleepy, I stuck around for the 1st batch to go thru; all 4 went thru in ~29-30mins while the average for 1 WU/GPU of that type is ~16mins.

On average, the CPU load is around 60% with 4 WU running on the CPU.
I noticed that the CPU utilisation went up and down as the 4 GPU assigned WUs were finishing. Two WU were just 20 secs apart and the top CPU% was in the low 80s. That is interesting since I remember this morning before reverting back to my regular 1WU/GPU settings that the CPU % was in the mid 90s as I had 6 WU running on the CPU.
If tonight's test is a success, I might try Friday night's settings again but with 5 WU on the CPU instead of 6.

Rob out!
ID: 1798787 · Report as offensive
1 · 2 · 3 · 4 · Next

Message boards : Number crunching : Help optimizing app config file with Core i7 GTX750ti


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.