Welcome to MilkyWay@home

GPU app teaser


Advanced search

Message boards : Application Code Discussion : GPU app teaser
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 11 · 12 · 13 · 14 · 15 · 16 · 17 . . . 19 · Next

AuthorMessage
ProfileKevint
Avatar

Send message
Joined: 22 Nov 07
Posts: 285
Credit: 1,076,786,368
RAC: 0
1 billion credit badge10 year member badge
Message 12855 - Posted: 25 Feb 2009, 14:27:21 UTC
Last modified: 25 Feb 2009, 14:27:53 UTC

I have a team mate that is having trouble.

He running Vista 32. He purchased a new HD 3850, BOINC 6.4.5

We have downloaded the Catalyst 8.12 and installed it. Installed the 0.19 version of the GPU app -

Everything looks fine. I have a couple boxes running so I got myself through the install, but I am not all that familiar with Vista so I don't know if this is a vista problem or something else.

Any Idea's ?

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=45384

28 Feb 2009 6:16:38 UTC
CPU time 0.405603
stderr out

<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
Running Milkyway@home ATI GPU application version 0.19 by Gipsel
CPU: Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz (4 cores/threads) 2.39398 GHz (398ms)
CAL Runtime: 1.3.145
Found 0 CAL devices
No compatible GPU found!

</stderr_txt>
]]>

Validate state Invalid
.
ID: 12855 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 12856 - Posted: 25 Feb 2009, 15:25:13 UTC - in response to Message 12855.  
Last modified: 25 Feb 2009, 15:25:41 UTC

I have a team mate that is having trouble.

He running Vista 32.

Any Idea's ?
Found 0 CAL devices
No compatible GPU found!

Protected mode installation of BOINC is not going to work with GPU apps under Vista (same as CUDA). Or are you accessing the machine over a remote desktop connection?
ID: 12856 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileKevint
Avatar

Send message
Joined: 22 Nov 07
Posts: 285
Credit: 1,076,786,368
RAC: 0
1 billion credit badge10 year member badge
Message 12858 - Posted: 25 Feb 2009, 15:34:34 UTC - in response to Message 12856.  

I have a team mate that is having trouble.

He running Vista 32.

Any Idea's ?
Found 0 CAL devices
No compatible GPU found!

Protected mode installation of BOINC is not going to work with GPU apps under Vista (same as CUDA). Or are you accessing the machine over a remote desktop connection?


We did install BOINC in protected mode -

Let me give him a call and see if that fixes it.



.
ID: 12858 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge10 year member badge
Message 12902 - Posted: 25 Feb 2009, 23:54:35 UTC

@Cluster Physik

Just installed 19b, and have a couple comments ...

One, shouldn't the file names be set on zslip to reflect the content, ie: win64-19b? or something like that ...

B) the application name in the err out is still saying 19 ... I can tell it is different because of the new content ... but, should we also not be somehow indicating the incremented versions, if nothing else for troubleshooting purposes ...

4. Have you sent the detection code to UCB? We are at least two versions from when I sent them notes on OpenCL and "Brook" (and generic GPU identification using windows API), but it looks like someone has to do the dirty work ... and you already have ... (PM sent also about this)
ID: 12902 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 12912 - Posted: 26 Feb 2009, 0:19:16 UTC - in response to Message 12902.  

@Cluster Physik

Just installed 19b, and have a couple comments ...

One, shouldn't the file names be set on zslip to reflect the content, ie: win64-19b? or something like that ...

I think that's a good comment and I can sort that out on zslip.


ID: 12912 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge10 year member badge
Message 12915 - Posted: 26 Feb 2009, 0:35:04 UTC - in response to Message 12912.  

@Cluster Physik

Just installed 19b, and have a couple comments ...

One, shouldn't the file names be set on zslip to reflect the content, ie: win64-19b? or something like that ...

I think that's a good comment and I can sort that out on zslip.


Cool, I usually D/L and hang onto the versions till it is very clear that the new is the keeper. Just incase I have to back-level the application ... I do the same thing with BOINC Manager (in case they start to hide the old list, even paranoids have enemies) ... :)
ID: 12915 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 12921 - Posted: 26 Feb 2009, 1:33:13 UTC - in response to Message 12902.  

@Cluster Physik

Just installed 19b, and have a couple comments ...

B) the application name in the err out is still saying 19 ... I can tell it is different because of the new content ... but, should we also not be somehow indicating the incremented versions, if nothing else for troubleshooting purposes ...

I thought the same and 0.19c and 0.19d (and future versions) identify themselves as such. And before someone is looking in the P3DNow! forum and is trying one of the newer versions, I can't recommend that. The multi GPU scheduler is not working correctly in the moment and trashes most WUs. So one should stay with 0.19b for the time being.

4. Have you sent the detection code to UCB? We are at least two versions from when I sent them notes on OpenCL and "Brook" (and generic GPU identification using windows API), but it looks like someone has to do the dirty work ... and you already have ... (PM sent also about this)

No, I didn't send it. But I've heard they got already some detection code. If all goes well it will be integrated in version 6.7 in the next month.
ID: 12921 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 12922 - Posted: 26 Feb 2009, 1:47:55 UTC - in response to Message 12915.  

@Cluster Physik

Just installed 19b, and have a couple comments ...

One, shouldn't the file names be set on zslip to reflect the content, ie: win64-19b? or something like that ...

I think that's a good comment and I can sort that out on zslip.


Cool, I usually D/L and hang onto the versions till it is very clear that the new is the keeper. Just incase I have to back-level the application ... I do the same thing with BOINC Manager (in case they start to hide the old list, even paranoids have enemies) ... :)

OK, zslip is updated so that the filenames also include OS, whether ATI, 32 or 64 bit, and version number.


ID: 12922 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfilePaul D. Buck

Send message
Joined: 12 Apr 08
Posts: 621
Credit: 161,934,067
RAC: 0
100 million credit badge10 year member badge
Message 12932 - Posted: 26 Feb 2009, 5:04:26 UTC - in response to Message 12921.  

No, I didn't send it. But I've heard they got already some detection code. If all goes well it will be integrated in version 6.7 in the next month.


Maybe yours is better ... :)

It could not hurt ...
ID: 12932 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Honza

Send message
Joined: 28 Aug 07
Posts: 31
Credit: 86,152,236
RAC: 0
50 million credit badge10 year member badge
Message 12933 - Posted: 26 Feb 2009, 7:24:35 UTC - in response to Message 12242.  

1% CPU load is not possible with the current split of the work between GPU and CPU. I don't plan to change anything on that because the effort for doing the remaining 0.1% of the CPU calculations on the GPU appears to be too much. If all the more urgent issues are solved, maybe one can think about it again. But I doubt the conclusion will be much different.
One needs the CPU about half a second in the beginning and slightly more (about a second) at the end of a WU (scales with CPU speed of course). In between a CPU load of about 10% of a core or maybe even less should be doable.

Does it makes sense to compile ATI GPU version with CPU SSSE3, SSE4.1 and/or SSE4.2 instructions? Or at least try to see how it would behave...
BOINC Project specifications and hardware requirements
ID: 12933 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 12936 - Posted: 26 Feb 2009, 8:10:40 UTC - in response to Message 12933.  

Does it makes sense to compile ATI GPU version with CPU SSSE3, SSE4.1 and/or SSE4.2 instructions? Or at least try to see how it would behave...

No sorry, the ATI would never fit into my Core 2 SFF, there is not enough room inside and I couldn't get the lid back on, no way ;)


ID: 12936 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 13441 - Posted: 1 Mar 2009, 1:12:37 UTC - in response to Message 12933.  
Last modified: 1 Mar 2009, 1:13:20 UTC

1% CPU load is not possible with the current split of the work between GPU and CPU. I don't plan to change anything on that because the effort for doing the remaining 0.1% of the CPU calculations on the GPU appears to be too much. If all the more urgent issues are solved, maybe one can think about it again. But I doubt the conclusion will be much different.
One needs the CPU about half a second in the beginning and slightly more (about a second) at the end of a WU (scales with CPU speed of course). In between a CPU load of about 10% of a core or maybe even less should be doable.

Does it makes sense to compile ATI GPU version with CPU SSSE3, SSE4.1 and/or SSE4.2 instructions? Or at least try to see how it would behave...

The really needed CPU time per WU is about one second or so. The remaining one is just for polling the GPU, which is not going to be faster with SSEx. As the speed increases with anything above SSE2 (used by the GPU app) are minor ones, one could maybe shave off a tenth of a second. Hardly worth it.
ID: 13441 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 13560 - Posted: 1 Mar 2009, 22:57:39 UTC - in response to Message 12800.  

But could you please test that it runs at all?

Here's an update on the double-GPU HD4870X2 vs the single-GPU HD4870/1GB.

HD4870/1GB runs at its highest configured clock speeds when under load from the MW app. Partly OC'ed to 800/900 MHz (gpu/mem) this means 5-8 seconds per WU.

HD4870X2 runds at its idle clocks (500 MHz gpu) even with one GPU under load from MW, which means 8-12 seconds per WU. It seems that giving half the card something to do isn't enough to "wake it up" from idle/2D clocks. But yes, the GPU apps do work on this card as well, albeit somewhat slower than on its lesser sibling.


Brickhead, what your saying is that your only running on one side of the x2 Video Card, but how many WU's are you running @ once with it by it's self or running another Project with it ???
ID: 13560 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 13571 - Posted: 1 Mar 2009, 23:36:17 UTC
Last modified: 1 Mar 2009, 23:37:21 UTC

Oh shock horror, hooray :D I've got Catalyst 9.2 working.

[edit] with a HD4870 / 0.19

ID: 13571 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 13577 - Posted: 2 Mar 2009, 0:28:15 UTC - in response to Message 13571.  
Last modified: 2 Mar 2009, 0:28:30 UTC

Oh shock horror, hooray :D I've got Catalyst 9.2 working.

[edit] with a HD4870 / 0.19

A bit of a strange pattern of crunching at first and then when I tried to run a CPDN WU the 4 MW just froze. I then suspended one of them and the other 3 set off working. Now it's settled down to a pattern of 2 MW WU's crunching every 40ish seconds and a CPDN WU crunching away every 436 hours.

I think that's slower that using using version 9.1 (not referring to CPDN), but then I'm not sure how long these new WUs take now.

ID: 13577 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileGalaxyIce
Avatar

Send message
Joined: 6 Apr 08
Posts: 2018
Credit: 100,142,856
RAC: 0
100 million credit badge10 year member badge
Message 13608 - Posted: 2 Mar 2009, 7:28:03 UTC - in response to Message 13577.  
Last modified: 2 Mar 2009, 7:32:38 UTC

Oh shock horror, hooray :D I've got Catalyst 9.2 working.

[edit] with a HD4870 / 0.19

A bit of a strange pattern of crunching at first and then when I tried to run a CPDN WU the 4 MW just froze. I then suspended one of them and the other 3 set off working. Now it's settled down to a pattern of 2 MW WU's crunching every 40ish seconds and a CPDN WU crunching away every 436 hours.

I think that's slower that using using version 9.1 (not referring to CPDN), but then I'm not sure how long these new WUs take now.

I took it back to 9.1 and the 8.12 and with both it seemed slower than 9.2. I think it's these longer work units which I haven't baselined for myself yet.

I've taken it back to 9.2 and it's been running overnight just fine with what seems to be good performance. It just seemed to take a little while to settle downn - a CPDN WU is crunching alongside 3 MW WUs nicely, all running with BOINC 6.4.6 (XP pro).

ID: 13608 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[Russia] michs

Send message
Joined: 16 Oct 08
Posts: 18
Credit: 164,409,593
RAC: 0
100 million credit badge10 year member badge
Message 13620 - Posted: 2 Mar 2009, 10:57:03 UTC
Last modified: 2 Mar 2009, 11:26:12 UTC

Maybe i found a bug in gpu application 19d.
I have a host with 2 active project: abc and gpu milkyway.
set [ncpus] to 6 in cc_config.xml
in app_info.xml milkyway set n4 w1.1
If all right we have 4 abc workunit running, and 4 milkiway.
if i push button suspend to abc, i have 4 running mw application, and 4 mw application wait.
if next push resume to abc, i have 4 running abc application and all mw application in wait state. 4 mw app running but no progress.
ID: 13620 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
50 million credit badge10 year member badgeextraordinary contributions badge
Message 13653 - Posted: 2 Mar 2009, 16:21:24 UTC - in response to Message 13620.  
Last modified: 2 Mar 2009, 16:28:28 UTC

Maybe i found a bug in gpu application 19d.
I have a host with 2 active project: abc and gpu milkyway.
set [ncpus] to 6 in cc_config.xml
in app_info.xml milkyway set n4 w1.1
If all right we have 4 abc workunit running, and 4 milkiway.
if i push button suspend to abc, i have 4 running mw application, and 4 mw application wait.
if next push resume to abc, i have 4 running abc application and all mw application in wait state. 4 mw app running but no progress.

I see. Basically that is a problem with some interference of the BOINC scheduler and the one build into the GPU app. If the WU, which owns the GPU in the moment gets evicted by the BOINC scheduler, the other WUs still "think" it is running, using the GPU and do not start. This behaviour should go away, if you disable the "leave aplications in memory" option. That way, the paused WUs are really stopped and not just suspended.
But maybe easier, just don't fiddle around with suspending projects ;)

PS:
I still don't get why people raise the number of concurrent WUs. Two are really enough for maximum efficiency. So even taking the start/finishing of WUs into account, you will never need more than 3 WUs (default value) for maximum efficiency of the GPU. It would be better for the throughput, if you lower the wait factor closer to 1. Or are you just trying to avoid the eviction of WUs by the boinc manager?
ID: 13653 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[Russia] michs

Send message
Joined: 16 Oct 08
Posts: 18
Credit: 164,409,593
RAC: 0
100 million credit badge10 year member badge
Message 13662 - Posted: 2 Mar 2009, 16:49:01 UTC - in response to Message 13653.  
Last modified: 2 Mar 2009, 17:04:06 UTC

w1.1 use to down gpu usage and sluggish behaviour of the user interface This is my home computer, and some time i am watching movie and doing something like that.

n4 using only while playing this parameter to try down sluggish behaviour.
And am am using n4 to calculate the time of crunching 1 wu. If i understand right while mw application wait, timer af "wall clock time" working.

Can i use second ati card not supported by your application to connect monitor and dont have sluggish behaviour of the user interface?
ID: 13662 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
STE\/E

Send message
Joined: 29 Aug 07
Posts: 486
Credit: 573,620,762
RAC: 0
500 million credit badge10 year member badge
Message 13664 - Posted: 2 Mar 2009, 17:00:36 UTC

I still don't get why people raise the number of concurrent WUs. Two are really enough for maximum efficiency. So even taking the start/finishing of WUs into account, you will never need more than 3 WUs (default value) for maximum efficiency of the GPU. It would be better for the throughput, if you lower the wait factor closer to 1. Or are you just trying to avoid the eviction of WUs by the boinc manager?


I agree, more in this case isn't neccessaraly better I have found right from the git go ... I'm running 3&3 3 MWay's & 3 Regular Wu's and getting 85,000-90,000 Per Day from 1 Card (4870 512mb with no Overclocking) ...
ID: 13664 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 11 · 12 · 13 · 14 · 15 · 16 · 17 . . . 19 · Next

Message boards : Application Code Discussion : GPU app teaser

©2019 Astroinformatics Group