Just wondering if there's any good reason to upgrade anything?


log in

Advanced search

Message boards : Number crunching : Just wondering if there's any good reason to upgrade anything?

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 7 · Next
Author Message
Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1215774 - Posted: 8 Apr 2012, 19:38:54 UTC - in response to Message 1215763.
Last modified: 8 Apr 2012, 19:39:39 UTC

Is there any particular reason why an AP 6 WU running on a GPU should only use 6% of the GPU?

ITs the first GPU astropulse task wu rig has recieved and I'm a little surprised by the low GPU usage.

Regards,

You're running 285.62 drivers, to fully utilise the GPU you're got to free a core because of the Nvidia OpenCL 100% CPU Use Bug,
otherwise the app will slow to a crawl, or downgrade to Cuda 3.2 drivers (263.xx to 267.24)

Claggy

Horacio
Send message
Joined: 14 Jan 00
Posts: 536
Credit: 60,296,621
RAC: 91,303
Argentina
Message 1215791 - Posted: 8 Apr 2012, 20:02:01 UTC - in response to Message 1215714.


The only other thing to do is to make sure AstroPulse v6 is enabled in your SETI@home preferences

Claggy

Thanks for the heads up, I took a look and they were all checked, so I'm good to go. It'll be interesting to see if I get any perf improvement. Is there anything I need to do to make sure the AP tasks go to the vid card instead of the CPU? I've heard that they can run faster, depending on the vid card of course, and as I have GTX260's, is there any way other than babysitting it, forcing tasks one way and then the other, and doing the compares, to know which is best?


Without babysitting, the easy way is to look here: http://setiathome.berkeley.edu/host_app_versions.php?hostid=5873188 (your host app details page)
Look for the average processing rate of the AP v6 in CPU and GPU (it will take some time to appear, as it requires 10 validated task per app with less than 10% rfi blanking). Right now it seems that you have received just one WU for GPU, so there is no data for the CPU version... It will appear when you get some CPU AP.

The APR of both apps will tell you wich one is faster (the greather APR)...

Not a very precise meassure, but easy to get... If you want a more accurate comparison, you will need to look in the results page of each AP WU crunched in each device and compare the crunching times of WUs with the same percentage of RFI blanking...

At the end if you want to run only on the GPU (or the CPU) you will need to remove the undesired app version from the app_info... (there are some trick to remove an app without losing WUs, but I think it is better to delay the full explanation until the time in which you have efectively choosen to remove it)

____________

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1215796 - Posted: 8 Apr 2012, 20:09:46 UTC - in response to Message 1215774.

Is there any particular reason why an AP 6 WU running on a GPU should only use 6% of the GPU?

ITs the first GPU astropulse task wu rig has recieved and I'm a little surprised by the low GPU usage.

Regards,

You're running 285.62 drivers, to fully utilise the GPU you're got to free a core because of the Nvidia OpenCL 100% CPU Use Bug,
otherwise the app will slow to a crawl, or downgrade to Cuda 3.2 drivers (263.xx to 267.24)

Claggy


Thanks, I'll free up a core, I dont particularly want to start backtracking drivers right now.

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1215888 - Posted: 9 Apr 2012, 0:08:21 UTC - in response to Message 1215774.
Last modified: 9 Apr 2012, 0:12:54 UTC

Well with one core free I get 10% GPU usage:-/ And the 267.24 drivers will not load on my rig, cant find compatible hardware.. Those drivers are for 7xxx or 8xxx not Gforce 5:-( The earliest I can get for 500 series GPU's is 275.27..
Dunno if that one is any better or not..

And an ordinary search on nvidia's site only gives access to the 267.24 driver if you select GF 7/8, select GF 4 or 5 and there is nowt before 285..

I'll just let it run at 10% and when its done I'll remove the ap 6 cl app and revert to CPU for AP 6.

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1216056 - Posted: 9 Apr 2012, 9:14:23 UTC - in response to Message 1215888.
Last modified: 9 Apr 2012, 9:15:06 UTC

Well with one core free I get 10% GPU usage:-/ And the 267.24 drivers will not load on my rig, cant find compatible hardware.. Those drivers are for 7xxx or 8xxx not Gforce 5:-( The earliest I can get for 500 series GPU's is 275.27..
Dunno if that one is any better or not..

And an ordinary search on nvidia's site only gives access to the 267.24 driver if you select GF 7/8, select GF 4 or 5 and there is nowt before 285..

I'll just let it run at 10% and when its done I'll remove the ap 6 cl app and revert to CPU for AP 6.

Regards,

The one NV AP Wu you had, did have a huge amount of Blanking, the more Blanking a Wu has the lower the GPU load, and the longer it takes:

single pulses: 4
repetitive pulses: 30
percent blanked: 99.85

Try doing some more, you should find minimaly Blanked tasks take ~45 minutes on a GTX460,

Claggy

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216059 - Posted: 9 Apr 2012, 9:20:23 UTC - in response to Message 1216056.

[quoteThe one NV AP Wu you had, did have a huge amount of Blanking, the more Blanking a Wu has the lower the GPU load, and the longer it takes:

single pulses: 4
repetitive pulses: 30
percent blanked: 99.85

Try doing some more, you should find minimaly Blanked tasks take ~45 minutes on a GTX460,

Claggy[/quote]

I'll give it a go, but the GPU's are 560's:-) So just maybe a tad faster..

As it stands I will have to wait and see, I 'think' I have 1 more WU of that type..

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1216062 - Posted: 9 Apr 2012, 9:44:48 UTC - in response to Message 1216059.

The one NV AP Wu you had, did have a huge amount of Blanking, the more Blanking a Wu has the lower the GPU load, and the longer it takes:

single pulses: 4
repetitive pulses: 30
percent blanked: 99.85

Try doing some more, you should find minimaly Blanked tasks take ~45 minutes on a GTX460,

Claggy


I'll give it a go, but the GPU's are 560's:-) So just maybe a tad faster..

As it stands I will have to wait and see, I 'think' I have 1 more WU of that type..

Regards,

You might also want to increase your parameters, -unroll 4 -ffa_block 2048 -ffa_block_fetch 1024 is quite low for a top end GPU,

I'm using: -instances_per_device 1 -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 -sbs 256 -hp on my GTX460

Claggy

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216073 - Posted: 9 Apr 2012, 10:21:21 UTC - in response to Message 1216062.

Thanks Claggy,
I'll bear that in mind, I only have the one AP6 cl task in my cache, and its in the queue, so I'll tweak the ap_info before it hits the GPU.

Cheers,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216185 - Posted: 9 Apr 2012, 16:35:41 UTC - in response to Message 1216062.

Hi Claggy,
Used your settings, now getting 95% GPU usage:-) There's only one problem, GPU temps are running at 80-82C.

My GTX560's are factory overclocked so I suppose that might well have something to do with the high temps.

But the AP 6 WU was completed in under 25 mins..

Nevertheless I think I'll try and underclock the GPU's, boinc always manages to pick the GPU most likely to get hot, device 2 has its fan partly obscured by a sata 6gbps card, and I need to have that card for my SSD's..

I've got nvidia inspector, I'll have a go at underclocking that GPU and see what happens with the next AP6 WU in the pipeline.

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1216189 - Posted: 9 Apr 2012, 16:38:29 UTC - in response to Message 1216185.

Hi Claggy,
Used your settings, now getting 95% GPU usage:-) There's only one problem, GPU temps are running at 80-82C.

My GTX560's are factory overclocked so I suppose that might well have something to do with the high temps.

But the AP 6 WU was completed in under 25 mins..

Nevertheless I think I'll try and underclock the GPU's, boinc always manages to pick the GPU most likely to get hot, device 2 has its fan partly obscured by a sata 6gbps card, and I need to have that card for my SSD's..

I've got nvidia inspector, I'll have a go at underclocking that GPU and see what happens with the next AP6 WU in the pipeline.

Regards,

I've got my fan fixed to 70%, that keeps my temps ~60°C

Claggy

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216201 - Posted: 9 Apr 2012, 16:52:52 UTC - in response to Message 1216189.
Last modified: 9 Apr 2012, 16:53:16 UTC


I've got my fan fixed to 70%, that keeps my temps ~60°C

Claggy


My fans are running at 100% all the time, and even at max fan GPU1 still went up to 82C.
My GTX 460 also used to run a lot cooler than the GTX560's Both are Palit cards and come factory overclocked.

I've just tried using nvidia inspector to down clock that GPU, but GPUZ is still telling me its overclocked:-/

As there are still a load of cuda_fermi GPU tasks in the cache I dont know when the next AP WU will get loaded... I found a 3rd WU hiding in the cache:-) But short of suspending over 100 other GPU tasks and forcing it to get loaded, I dont think I'll ever be able to determine when its going to get to the head of the queue..

Would cutting the params you gave in half drop the GPU usage by a half as well?

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1216202 - Posted: 9 Apr 2012, 16:56:59 UTC - in response to Message 1216201.

Would cutting the params you gave in half drop the GPU usage by a half as well?

You can try dropping the -unroll value to 8 or 6, that'll lower Memory usage, and maybe utilisation,

Claggy

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216205 - Posted: 9 Apr 2012, 17:05:51 UTC - in response to Message 1216202.

Would cutting the params you gave in half drop the GPU usage by a half as well?

You can try dropping the -unroll value to 8 or 6, that'll lower Memory usage, and maybe utilisation,

Claggy


I've downclocked the GPU, found out why it wasnt doing as told, I've had msi afterburner controlling the fans, and tucked away was a little box to synch all GPU's that was ticked:-) Unticked and the downclock worked.

But I'll keep in mind the -unroll values, btw what does the -hp at the end of the line do?

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Claggy
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 3964
Credit: 31,880,997
RAC: 10,945
United Kingdom
Message 1216209 - Posted: 9 Apr 2012, 17:12:04 UTC - in response to Message 1216205.
Last modified: 9 Apr 2012, 17:12:41 UTC

btw what does the -hp at the end of the line do?

High Priority for the process thread:

Priority of worker thread raised successfully
Priority of process adjusted successfully, high priority class used

Claggy

Richard Haselgrove
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8275
Credit: 44,981,343
RAC: 13,875
United Kingdom
Message 1216215 - Posted: 9 Apr 2012, 17:59:21 UTC - in response to Message 1216209.

btw what does the -hp at the end of the line do?

High Priority for the process thread:

Priority of worker thread raised successfully
Priority of process adjusted successfully, high priority class used

Claggy

Note that this has nothing whatsoever to do with the 'high priority' you sometimes see against a task name in BOINC Manager.

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216227 - Posted: 9 Apr 2012, 18:44:18 UTC - in response to Message 1216215.
Last modified: 9 Apr 2012, 18:45:12 UTC

btw what does the -hp at the end of the line do?

Priority of process adjusted successfully, high priority class used

Claggy

Note that this has nothing whatsoever to do with the 'high priority' you sometimes see against a task name in BOINC Manager.

Hi Richard,
OK, After a little thought, I suspended all non AP GPU tasks, and loaded the next AP 6 GPU WU.. And the dratted thing is highly blanked, or so it would seem from the GPU usage, which is alternating between about 6% and 43% so far.. Well I have another one after the present one maybe it wont have so much blanking and I'll get an idea if underclocking the GPU will work at reducing heat on that card.
As it stands I cant afford to let the heat problem slide, if an AP WU gets loaded when I'm not around to check on it I could end up with a fried card, which I cant afford to replace right now.

Regards,
____________
Cliff,
Been there, Done that, Still no damm T shirt!

Josef W. Segur
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4140
Credit: 1,004,670
RAC: 253
United States
Message 1216231 - Posted: 9 Apr 2012, 18:50:13 UTC

I've held back for awhile to allow the specific system upgrade issues to settle out, but IMO it's time to inject a more general reply to the subject question.

As noted early in the thread, Al was running the AK_v8b CPU application from the Lunatics v0.38 installer. Last November a small bug was discovered in those apps and Lunatics produced the v0.39 installer with corrected AK_v8b2 CPU applications.

Our hope is that whenever we release new apps, all our users will be paying close enough attention to know whether they ought to upgrade.

Joe

Profile cliff
Avatar
Send message
Joined: 16 Dec 07
Posts: 322
Credit: 2,509,590
RAC: 0
United Kingdom
Message 1216247 - Posted: 9 Apr 2012, 19:31:29 UTC - in response to Message 1216231.

I've held back for awhile to allow the specific system upgrade issues to settle out, but IMO it's time to inject a more general reply to the subject question.

As noted early in the thread, Al was running the AK_v8b CPU application from the Lunatics v0.38 installer. Last November a small bug was discovered in those apps and Lunatics produced the v0.39 installer with corrected AK_v8b2 CPU applications.

Our hope is that whenever we release new apps, all our users will be paying close enough attention to know whether they ought to upgrade.
Joe

Hi Joe,
Its a tad moot:-) When a new app is published, and is in use apparently without too many issues, then people will upgrade to that app.. I would have thought that a good many people will have subscribed to the appropriate forums and thus be aware when a new app is out.

As I see it there are always going to be some of us that will have problems/niggles with the software..

I've yet to see any program thats been proved to work on any & all systems:-)

And there will always be some who for whatever reason miss the news that there is a new set of apps out or a new specific app for this or that..

Doubt anything can be done about it, other than perhaps the S@H server posting a warning to users that if they are using anon platform, its time to upgrade.

I suspect however there will still be some who dont get the message for whatever reason.

Which is I suppose one of the reasons these forums exist:-)

Regards,

____________
Cliff,
Been there, Done that, Still no damm T shirt!

Al
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 48,996,657
RAC: 30,191
United States
Message 1217504 - Posted: 13 Apr 2012, 3:29:43 UTC
Last modified: 13 Apr 2012, 3:39:48 UTC

I have a couple systems that were set and forget, and I had this one off line for a number of months for various reasons. I'm glad to be back at it, but there are times that it's hard to keep up on the latest and greatest, as life's pretty busy at times. I do try to pay attn if there are issues with what I am running, and would prefer to keep it up to date if possible, when it makes sense to do so. I believed it did at this time, as the AP5s are history, and the AP6s are the rule. I'll probably try to keep it updated on Intel's infamous time frame, namely Moore's Law, and do a major upgrade every 18 months or so. That should keep me current enough, but still not have to worry about each and every update.

Oh, and as far as I can tell, things seem to be working out alright with the updated software. I want to let it run thru a number of AP tasks, they are running on both CPU & GPU, so after I get a bunch of them under my belt, I'll dig in and compare results, and determine which I would like to run it under, then force them all to whichever turns out to do them the best on this machine.
____________

Profile perryjay
Volunteer tester
Avatar
Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 13,716,529
RAC: 12,170
United States
Message 1217661 - Posted: 13 Apr 2012, 15:51:03 UTC - in response to Message 1217504.

Al, don't forget, we have a new type of work unit coming out very soon so you will have to change things for it. I believe all of us running anonymous platform will have to change but I'm not sure what will happen for those running stock.
____________


PROUD MEMBER OF Team Starfire World BOINC

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 7 · Next

Message boards : Number crunching : Just wondering if there's any good reason to upgrade anything?

Copyright © 2014 University of California