Running Stock Experiment

Message boards : Number crunching : Running Stock Experiment
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 5 · Next

AuthorMessage
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1501242 - Posted: 8 Apr 2014, 6:02:11 UTC

I got this idea from watching Stens RAC fall from running all AP to running just MB work.

My I7 920 with a GTS 250 card averages about 5400 RAC running a mix of MB and AP on the CPU , And MB on the GPU, Using lunatics Op apps.

Here is what I plan on doing.
1. Run stock apps on CPU doing only MB work. Will let RAC get stable.
2. Run stock AP on CPU only. Let RAC get stable.
3. Run stock MB and AP on CPU. And run stock MB on gpu. Until stable.

Any thing I have missed?

I still have two AP's running for about another 14 hours. So I will probally start this after I get home from work at 11:00 PM EDT.

I will check off the boxes so I dont get any GPU work, And the box for getting other work.

My oringal thought was just unistalling lunatics, But Sten told me that just
changing the AP_config .xml file to another name would be better.

Any thoughts on that?

This is basically a lets see what happens thing.
[/quote]

Old James
ID: 1501242 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1501298 - Posted: 8 Apr 2014, 8:52:02 UTC - in response to Message 1501242.  
Last modified: 8 Apr 2014, 8:55:48 UTC


3. Run stock MB and AP on CPU. And run stock MB on gpu. Until stable.

Any thing I have missed?


Only one thing to note - with what this experiment should be compared.
First two should be compared between each other.
Third will definitely give higher RAC cause you will use additional device in this experiment.
But too higher or not enough higher? We could not say. some control experiment should be carried in addition then. What it could be? Big discrepance between GPU MB and GPU AP credits already known.
Maybe comparison between GPU stock MB and GPU opt MB then?
I don't expect big difference in performance between these releases, but if you would see big difference in RAC that would probably mean that anonymous platform vs stock RAC generation has some additional issues. Worth to check ?

EDIT: if such plan would be accepted then worth to go with CPU stock MB in 3 and 4, cause CPU AP rac may differ (will see in experiments 1 and 2) so more homogenious environment for GPU part will be to use only CPU stock MB on CPU.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1501298 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1501304 - Posted: 8 Apr 2014, 9:09:17 UTC - in response to Message 1501298.  


3. Run stock MB and AP on CPU. And run stock MB on gpu. Until stable.

Any thing I have missed?


Only one thing to note - with what this experiment should be compared.
First two should be compared between each other.
Third will definitely give higher RAC cause you will use additional device in this experiment.
But too higher or not enough higher? We could not say. some control experiment should be carried in addition then. What it could be? Big discrepance between GPU MB and GPU AP credits already known.
Maybe comparison between GPU stock MB and GPU opt MB then?
I don't expect big difference in performance between these releases, but if you would see big difference in RAC that would probably mean that anonymous platform vs stock RAC generation has some additional issues. Worth to check ?

EDIT: if such plan would be accepted then worth to go with CPU stock MB in 3 and 4, cause CPU AP rac may differ (will see in experiments 1 and 2) so more homogenious environment for GPU part will be to use only CPU stock MB on CPU.

Im looking to see what will be the higher RAC for my machine. I know everyones milage will vary. Im curious to know what will happen.
[/quote]

Old James
ID: 1501304 · Report as offensive
Profile William
Volunteer tester
Avatar

Send message
Joined: 14 Feb 13
Posts: 2037
Credit: 17,689,662
RAC: 0
Message 1501330 - Posted: 8 Apr 2014, 10:59:26 UTC

Renaming app_info.xml to something else is a quick and easy way to go stock and preserve the app_info.xml for future reuse. Yes, I'd recommend that, especially if you did tweaks to the app_info.xml.
Just make sure you have run down your cache before you rename it - you should get everything back, with sent lost results enabled, but it's neater to start with an empty cache.

ad 3) stock MB on GPU will be a slight bugger with the multiple cuda versions. Ah, I see you already have established APRs on that host. Not a problem then.
A person who won't read has no advantage over one who can't read. (Mark Twain)
ID: 1501330 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1501334 - Posted: 8 Apr 2014, 11:37:42 UTC - in response to Message 1501242.  

...My I7 920 with a GTS 250 card...Any thing I have missed?

Well, considering my 250 produces more points running MB than it does AP, you may not receive the results you are seeking. The G92 cards are just terrible at running APs. You might want to consider using a newer nVidia card for this test...
ID: 1501334 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1501345 - Posted: 8 Apr 2014, 12:30:02 UTC - in response to Message 1501330.  

Renaming app_info.xml to something else is a quick and easy way to go stock and preserve the app_info.xml for future reuse. Yes, I'd recommend that, especially if you did tweaks to the app_info.xml.
Just make sure you have run down your cache before you rename it - you should get everything back, with sent lost results enabled, but it's neater to start with an empty cache.

ad 3) stock MB on GPU will be a slight bugger with the multiple cuda versions. Ah, I see you already have established APRs on that host. Not a problem then.

Make sure you keep a backup copy of all the programs, DLLs and other files referenced in your app_info - they can't be downloaded automatically if you decide to revert back again. Not saying that you'll need them, but you know what they say about Murphy's Law...
ID: 1501345 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1501571 - Posted: 9 Apr 2014, 4:34:45 UTC - in response to Message 1501334.  
Last modified: 9 Apr 2014, 4:35:31 UTC

...My I7 920 with a GTS 250 card...Any thing I have missed?

Well, considering my 250 produces more points running MB than it does AP, you may not receive the results you are seeking. The G92 cards are just terrible at running APs. You might want to consider using a newer nVidia card for this test...

I dont run AP on that GTS 250 card its just a MB card.
Edit- good point Richard.
[/quote]

Old James
ID: 1501571 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1501587 - Posted: 9 Apr 2014, 5:28:59 UTC - in response to Message 1501571.  

...My I7 920 with a GTS 250 card...Any thing I have missed?

Well, considering my 250 produces more points running MB than it does AP, you may not receive the results you are seeking. The G92 cards are just terrible at running APs. You might want to consider using a newer nVidia card for this test...

I dont run AP on that GTS 250 card its just a MB card.
Edit- good point Richard.

Okay, I see now. You're not going to test the GPU at all. I was hoping to see someone with a mid-range nVidia 6 series card run the same test as Sten.
ID: 1501587 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1501589 - Posted: 9 Apr 2014, 5:36:24 UTC

The test has started. I renamed my app_info.xml file. set my perfferances for MB7 work,No GPU and no to getting other types fo work.

I have received a ton of VLAR's. I assume Im doing just stock as the times are repoted at 7 hours Plus for each.

So now we waite, And waite:)

And before someone mentions 3 aborted work units, Just tribute that to being stupid. After changing my preffs, I did not do an update first before I hit get new tasks. And got 3 GPU work units. I dont abort work but I didnt want to skew up my test results.
[/quote]

Old James
ID: 1501589 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1501590 - Posted: 9 Apr 2014, 5:39:16 UTC - in response to Message 1501587.  
Last modified: 9 Apr 2014, 5:42:34 UTC

...My I7 920 with a GTS 250 card...Any thing I have missed?

Well, considering my 250 produces more points running MB than it does AP, you may not receive the results you are seeking. The G92 cards are just terrible at running APs. You might want to consider using a newer nVidia card for this test...

I dont run AP on that GTS 250 card its just a MB card.
Edit- good point Richard.

Okay, I see now. You're not going to test the GPU at all. I was hoping to see someone with a mid-range nVidia 6 series card run the same test as Sten.


I think I might let the GPU run free with MB before Im done, But i have never run Ap on it and dont intend to. I dont think the card would handle it. Edit- In August it will be 5 years old.
[/quote]

Old James
ID: 1501590 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1501603 - Posted: 9 Apr 2014, 6:01:29 UTC - in response to Message 1501590.  
Last modified: 9 Apr 2014, 6:27:19 UTC

...My I7 920 with a GTS 250 card...Any thing I have missed?

Well, considering my 250 produces more points running MB than it does AP, you may not receive the results you are seeking. The G92 cards are just terrible at running APs. You might want to consider using a newer nVidia card for this test...

I dont run AP on that GTS 250 card its just a MB card.
Edit- good point Richard.

Okay, I see now. You're not going to test the GPU at all. I was hoping to see someone with a mid-range nVidia 6 series card run the same test as Sten.


I think I might let the GPU run free with MB before Im done, But i have never run Ap on it and dont intend to. I dont think the card would handle it. Edit- In August it will be 5 years old.

The GTS 250 should run an unblanked AP in around 4.5 hours. For an older card, that's not too bad, considering the present extreme low end cards. I just changed logic boards on one machine and ran a few APs on the even older 8800 GT, it runs them in right at 5 hours. In contrast, an AMD 7750 runs one in about an hour, and uses less than 75 watts. However, both the 8800 & 250 run a MB almost as fast as the 7750.
ID: 1501603 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1501621 - Posted: 9 Apr 2014, 7:47:14 UTC

Even my low power 9800GT's will do an AP in an average of 6hrs compared to 10hr average per Q6600 core (my 3570K & 2500K take 6hr average per core).

After I find the peak average on that Q6600 rig I'll switch the 9800GT's over to MB only and see what happens.

Cheers.
ID: 1501621 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1502010 - Posted: 10 Apr 2014, 5:17:29 UTC

As I expected Rac has taken nose dive, But not as bad as I thought.
My average RAC on this machine running Lunatics on the CPU and GPU. Was around 5450 Plus or Minus a few hundred. And that was with AP and a MB mix.

Just running MB stock at the moment it is at 4800. So I will give it some time and see what happens.
[/quote]

Old James
ID: 1502010 · Report as offensive
Profile William
Volunteer tester
Avatar

Send message
Joined: 14 Feb 13
Posts: 2037
Credit: 17,689,662
RAC: 0
Message 1502082 - Posted: 10 Apr 2014, 8:23:13 UTC
Last modified: 10 Apr 2014, 8:38:21 UTC

RAC has a half-life of one week - asymptotic curve.

IOW after 2 weeks you still have ~1300 points from before the switch contributing to RAC. It may take a few weeks to level off, not counting pendings.

Your GPU tasks take roughly 1/10 of the time of CPU, so relative contribution should be around 500 from CPU and 5000 from GPU (not counting AP) - that brings us to at least 4 weeks before old RAC has decayed enough for 'MB CPU only' RAC to be predominant. Aybody mentioned you need patience for this?

edit: I forgot of course about it being 8 core. so maybe half half. more around 2 weeks then.
You ain;t looking at a smooth system of course, so might be worth running extra to get a better idea of where it levels of (and don't talk to me about error margins).

this host - to save me having to look it up again each time ;)

BTW I hit my head pretty hard yesterday, if I'm talking BS you can take it as a sign that cognitive functions suffered.
A person who won't read has no advantage over one who can't read. (Mark Twain)
ID: 1502082 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1502085 - Posted: 10 Apr 2014, 8:51:03 UTC - in response to Message 1502082.  

RAC has a half-life of one week - asymptotic curve.

IOW after 2 weeks you still have ~1300 points from before the switch contributing to RAC. It may take a few weeks to level off, not counting pendings.

Your GPU tasks take roughly 1/10 of the time of CPU, so relative contribution should be around 500 from CPU and 5000 from GPU (not counting AP) - that brings us to at least 4 weeks before old RAC has decayed enough for 'MB CPU only' RAC to be predominant. Aybody mentioned you need patience for this?

edit: I forgot of course about it being 8 core. so maybe half half. more around 2 weeks then.
You ain;t looking at a smooth system of course, so might be worth running extra to get a better idea of where it levels of (and don't talk to me about error margins).

this host - to save me having to look it up again each time ;)

BTW I hit my head pretty hard yesterday, if I'm talking BS you can take it as a sign that cognitive functions suffered.

I get your point. Two weeks minimun then:)Maybe more as needed.
I did forget about the pendings.
[/quote]

Old James
ID: 1502085 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1502289 - Posted: 10 Apr 2014, 18:43:46 UTC - in response to Message 1502085.  

RAC has a half-life of one week - asymptotic curve.

IOW after 2 weeks you still have ~1300 points from before the switch contributing to RAC. It may take a few weeks to level off, not counting pendings.

Your GPU tasks take roughly 1/10 of the time of CPU, so relative contribution should be around 500 from CPU and 5000 from GPU (not counting AP) - that brings us to at least 4 weeks before old RAC has decayed enough for 'MB CPU only' RAC to be predominant. Aybody mentioned you need patience for this?

edit: I forgot of course about it being 8 core. so maybe half half. more around 2 weeks then.
You ain;t looking at a smooth system of course, so might be worth running extra to get a better idea of where it levels of (and don't talk to me about error margins).

this host - to save me having to look it up again each time ;)

BTW I hit my head pretty hard yesterday, if I'm talking BS you can take it as a sign that cognitive functions suffered.

I get your point. Two weeks minimun then:)Maybe more as needed.
I did forget about the pendings.

Having 8 cores doesn't actually affect the RAC half life, but it does mean that the 60 or so pendings from before you started the test will fade in importance sooner. Looking at your validated work for the last 24 hours has a mix, of course, but it's already over two thirds from stock.

For ten stock CPU validated VLAR WUs from that list, the average run time was 32182 seconds and the average credit grant 102.5 cobblies. With 8 cores the host would do 21.5 of those per day, so that data forecasts an eventual RAC of about 2200. What kind of mix of angle ranges it will get during the test, and how it will react to those differences, could shift the RAC considerably. How often the wingmates are running stock CPU will also affect it, and getting out of the VLAR range will of course mean many pairings with GPUs.
                                                                  Joe
ID: 1502289 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1502316 - Posted: 10 Apr 2014, 20:00:16 UTC

I normally consider 4-6 months a good amount of time to find how a change will effect a systems RAC. That way I can compare it to tends that might be ongoing. Such as shorty storms or lots of VLAR tasks.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1502316 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1502534 - Posted: 11 Apr 2014, 5:50:59 UTC
Last modified: 11 Apr 2014, 5:54:33 UTC

Rac is still dropping like I expected, It is below 4400 now. I have 68 pendings and nine of them are v7 so I have a way to go getting them cleared up.

@Joe- I looked at my other two machines and they have a lot of VLAR's but it seems this host is pretty much flooded with them. Is that because of running stock or just the luck of the draw?

@HAL- I will run untill my pendings are gone and RAC stabilizes. But Id rather not run each test for 4-6 months:)

Im posting this now and will edit in (this host ) So William wont get mad at me:).
[/quote]

Old James
ID: 1502534 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1502588 - Posted: 11 Apr 2014, 7:59:37 UTC - in response to Message 1502534.  

@Joe- I looked at my other two machines and they have a lot of VLAR's but it seems this host is pretty much flooded with them. Is that because of running stock or just the luck of the draw?

Luck of the draw - depends what's around when you request work.

But hosts running CPU only do get a high proportion, because they pick up the leftovers from NVidia machines. Somebody's got to do them, and the project (with a bit of nudging from a subset of GPU users) have decided that they're best done on CPUs.
ID: 1502588 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1502749 - Posted: 11 Apr 2014, 17:20:50 UTC

Thanks for the info Richrard. I remember back in the day my GTS 250 could crunch one in 2.5 hours. I dont recall any ill effects from running them other than some screen lag if I was trying to play solitaire.
[/quote]

Old James
ID: 1502749 · Report as offensive
1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Number crunching : Running Stock Experiment


 
©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.