Advanced search

Message boards : Graphics cards (GPUs) : The new application is out (acemd2)

Author Message
Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15122 - Posted: 10 Feb 2010 | 16:54:14 UTC

We have put the first WUs in the new acemd2 application.

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15128 - Posted: 10 Feb 2010 | 19:42:19 UTC - in response to Message 15122.
Last modified: 10 Feb 2010 | 19:44:41 UTC

Are these indicated by the application ...
ACEMD - GPU molecular dynamics 6.01 (cuda)?
____________
Thanks - Steve

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15129 - Posted: 10 Feb 2010 | 19:46:14 UTC - in response to Message 15128.

I get:

Wed 10 Feb 2010 05:56:59 PM CET GPUGRID Starting task 594-IBUCH_1000smd_pYEEI_100202-3-10-RND9182_0 using acemd2 version 602

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15130 - Posted: 10 Feb 2010 | 19:52:50 UTC - in response to Message 15129.

Are you expecting the 60% increase with this application/task set combo?

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15131 - Posted: 10 Feb 2010 | 20:33:53 UTC - in response to Message 15130.

Exactly as the beta.

gdf

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15132 - Posted: 10 Feb 2010 | 20:39:58 UTC - in response to Message 15129.
Last modified: 10 Feb 2010 | 20:40:31 UTC

I get:

Wed 10 Feb 2010 05:56:59 PM CET GPUGRID Starting task 594-IBUCH_1000smd_pYEEI_100202-3-10-RND9182_0 using acemd2 version 602


OK, I think I follow you, I see the same when I find the relevant entry in the Messages tab in BOINC Manager. If you look at the Tasks tab, the Application column says it is ACEMD - GPU molecular dynamics 6.01 (cuda).

Can you please align whatever is necessary so at least the app name and version are the same?
____________
Thanks - Steve

Profile Zydor
Send message
Joined: 8 Feb 09
Posts: 252
Credit: 1,309,451
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 15135 - Posted: 11 Feb 2010 | 1:21:58 UTC - in response to Message 15132.

I got two 6.01's

11/02/2010 01:09:08 GPUGRID Starting task 387-IBUCH_1000smd_pYEEI_100202-3-10-RND6076_1 using acemd2 version 601

Shows as 6.01 in BAM tasks

Shows as acemd2_6.01_windows_intelx86__cuda in Win Task Manager

Looks as though there are still 6.01's in the queue. I have one running, and one queued this end.

Is there any expected run time difference (as such) between 6.01 & 6.02?

Regards
Zy

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1451
Credit: 3,575,929,351
RAC: 310,964
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15138 - Posted: 11 Feb 2010 | 8:10:57 UTC - in response to Message 15135.

Is there any expected run time difference (as such) between 6.01 & 6.02?

Applications: 6.01 is Windows, 6.02 is Linux.

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15139 - Posted: 11 Feb 2010 | 8:22:10 UTC - in response to Message 15138.

The "application" column in boinc manager is not relevant yet. What matters is the "messages" string. For now there are only a few WUs for the new app, more to come.

Profile Zydor
Send message
Joined: 8 Feb 09
Posts: 252
Credit: 1,309,451
RAC: 0
Level
Ala
Scientific publications
watwatwatwat
Message 15141 - Posted: 11 Feb 2010 | 9:31:58 UTC - in response to Message 15138.

Ding Dong :)

Thanks Richard

Regards
Zy

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15146 - Posted: 11 Feb 2010 | 12:08:11 UTC - in response to Message 15141.

Old application,
Picked up these two tasks today,
5-GIANNI_BIND_751-81-100-RND2022
52-GIANNI_BIND_2-79-100-RND8818
Both use v6.71 (cuda23).
Although they say that they need 15% of one CPU each, they are actually only using about 1% between them. Tasks running on two GT240's on an i7-920 W7Pro X64, 7Threads enabled for CPU use, total CPU usage 88%. I freed up one thread to expedite the GPU tasks (the other 7 are crunching WCG tasks).
Are there many new application tasks compared to the older tasks?

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15150 - Posted: 11 Feb 2010 | 19:19:09 UTC - in response to Message 15146.

It might be some consolation to others that my cards are getting the older, slower and wobblier tasks!

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15156 - Posted: 11 Feb 2010 | 20:37:57 UTC - in response to Message 15150.

In project preferences you should be able to select the application.

gdf

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15162 - Posted: 12 Feb 2010 | 1:12:27 UTC - in response to Message 15156.

OK, I know you mean:
Goto Your account, select GPUGrid Preferences, Select Edit GPUGrid Prefernces, and Deselect ACEMD - leaving ACEMD 2.

Thanks,

http://www.gpugrid.net/prefs.php?subset=project

zpm
Avatar
Send message
Joined: 2 Mar 09
Posts: 159
Credit: 13,639,818
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 15164 - Posted: 12 Feb 2010 | 5:40:28 UTC - in response to Message 15162.

6 hrs and 25 minutes on my gtx 260 216 cores. i would say that it is faster.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15169 - Posted: 12 Feb 2010 | 10:29:50 UTC - in response to Message 15164.
Last modified: 12 Feb 2010 | 10:31:23 UTC

I am also getting big performance improvements from my GTX260sp216:


    21,804.53 3,627.96 4,531.91 6,118.08 ACEMD - GPU molecular dynamics v6.01 (cuda)
    34,410.33 6,055.99 4,531.91 6,118.08 Full-atom molecular dynamics v6.71 (cuda23)
    41,126.95 11,995.04 4,531.91 6,118.08 Full-atom molecular dynamics v6.71 (cuda23)

    19,842.14 5,306.09 3,977.21 5,369.23 ACEMD - GPU molecular dynamics v6.01 (cuda)
    26,610.25 5,641.57 3,977.21 5,369.23 Full-atom molecular dynamics v6.71 (cuda23)
    26,838.32 5,967.80 3,977.21 5,369.23 Full-atom molecular dynamics v6.71 (cuda23)
    42,542.30 12,353.83 4,531.91 4,531.91 Full-atom molecular dynamics v6.71 (cuda23)
    40,984.92 12,257.39 4,531.91 4,531.91 Full-atom molecular dynamics v6.71 (cuda23)
    33,375.15 7,349.85 2,882.39 3,891.23 Full-atom molecular dynamics v6.71 (cuda23)

Profile [AF>Libristes] Dudumomo
Send message
Joined: 30 Jan 09
Posts: 45
Credit: 425,620,748
RAC: 0
Level
Gln
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15171 - Posted: 12 Feb 2010 | 12:09:50 UTC - in response to Message 15169.
Last modified: 12 Feb 2010 | 12:10:49 UTC

Wouaaaahh !!
Awesome !!

I had a 6.70 and right after a 6.02 for the same kind of Wu on my GTX275
6.70 : GPU Time: 18,351.38 CPU Time: 9,129.17
6.02 : GPU Time: 12,894.11 CPU Time: 1,503.43

GPU Time: -30% CPU Time: -80%

Well done !!

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15173 - Posted: 12 Feb 2010 | 12:29:51 UTC - in response to Message 15171.

Agreed, Well Done!
Some serious amount of improvement was achieved by refining the application.

Now seeing similar improvement on my GT240 cards:
1841813 1158304, 45,402.39 2,734.11 4,531.91 6,118.08 ACEMD - GPU molecular dynamics v6.01 (cuda)
1840137 1157154, 66,144.96 3,361.48 4,531.91 6,118.08 Full-atom molecular dynamics v6.71 (cuda23)

45,402.39s rather than 66,144.96s is a 45% GPU improvement.
CPU time also dropped by 23% on these CC1.2 cards.

canardo
Send message
Joined: 11 Feb 09
Posts: 4
Credit: 8,675,472
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 15176 - Posted: 12 Feb 2010 | 14:24:57 UTC - in response to Message 15122.

Looks like a nice improvement between 30% and 40% on

# Device 0: "GeForce GTS 250"
# Clock rate: 1.84 GHz
# Total amount of global memory: 1073414144 bytes
# Number of multiprocessors: 16
# Number of cores:

Run time CPU time Claimed cred. Granted cred. Application

58933,27 3531,89 4531,91 6118,08 F-a mol. dyn. v6.71 (cuda23)
42016,99 2342,64 4531,91 6118,08 ACEMD - GPU mol dyn v6.01(cuda)

71% 66% 100% 100% ratio

well done

Jaak

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15182 - Posted: 12 Feb 2010 | 16:15:08 UTC - in response to Message 15173.
Last modified: 12 Feb 2010 | 16:22:59 UTC

Agreed, Well Done!
Some serious amount of improvement was achieved by refining the application.

Now seeing similar improvement on my GT240 cards:

45,402.39s rather than 66,144.96s is a 45% GPU improvement.
CPU time also dropped by 23% on these CC1.2 cards.

My GT 240 512MB card on the same GIANNI_BIND WUs: an average of 36,200s for v6.01 and 56,344s for v6.71. as you can see quite a bit faster. Two differences: it seems the GDDR5 version is considerably faster in GPUGRID than the GDDR3 version.

The other difference is that my machine is running WinXP and yours Win7. Win7 is still showing a significant slowdown with the new app. While I don't have GIANNI_BIND WUs to compare as my Win7 GT 240 box hasn't gotten any yet, here's a comparison of IBUCH_1000smd_pYEEI WUs: the WinXP box runs an average of 30,350s/WU and the Win7 box is running an average of 34,400s/WU. This is even though the card in the Win box is running at 625MHz while the card in WinXP is at 600MHz. Both are GT 240 512MB GDDR5.

I guess the moral of this story for GPUGRID and the GT 240 is: for fastest performance run WinXP (either 32bit or 64bit is faster for any of the NVidia cards that I've checked), and buy the GDDR5 version rather than GDDR3.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15198 - Posted: 12 Feb 2010 | 21:34:30 UTC - in response to Message 15182.

I agree, and thanks for pointing that out.
The DDR5 cards performance increased by about 55% with the new application, whereas the DDR3 based cards only increased by about 45%.
When you do all the maths the DDR5 cards work out at being about 15% faster than the DDR3 cards, everything else being equal.

Of course the purchase costs can be different too, so it is worth shopping around within your budget for the correct card. It is also worth noting that some GT220 cards actually cost as much as the lesser GT240 cards, but will certainly not perform anywhere near as well. At the present time the GT240s offer the best value for money, by some way, but even within the GT240 range there are some bargains!

There is a lot of variation with these GT 240 cards:

RAM can be DDR3 or DDR5 - note the 15% performance gain from DDR5 and work out if it is worth the extra cost.

RAM can be anywhere from 512MB up to 2GB (but this makes no difference here)! So dont pay any extra for More RAM!

Shader count is generally 96, but not always! There are a few GT240's that have more (112, I think) - they are harder to find and tend to cost more. If you find one with DDR5 and good clocks, it should outperform a GTS250, and cost a fair bit less, both in terms of the initial purchase cost and especially the running costs (about 65W vs 160W +/- 30W depending on the Manufacturers)!

Clocks - well there are many factory flavours! Obviously the higher the better, but if you are into overclocking, what they have already done is what you cant do.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15200 - Posted: 12 Feb 2010 | 22:52:23 UTC - in response to Message 15198.

There is a lot of variation with these GT 240 cards:

Shader count is generally 96, but not always! There are a few GT240's that have more (112, I think) - they are harder to find and tend to cost more. If you find one with DDR5 and good clocks, it should outperform a GTS250, and cost a fair bit less, both in terms of the initial purchase cost and especially the running costs (about 65W vs 160W +/- 30W depending on the Manufacturers)!

I agree that the GT 240 is the best bang for the buck right now. I've seen it as low as $50 AR shipped for the GDDR5 version. The 96 shader GDDR5 version is already faster than the GTS 250 at GPUGRID and as you point out uses less power (the GT 240 GDDR5 uses about 15 watts less running GPUGRID than my 96 shader 9600GSO cards and is considerably faster).

Out of curiosity, where have you seen a 112 shader version? I haven't run across it. Could it be the OEM only GTS 240 that you're thinking of, which is G96 based:

GeForce GTS 240 Reference PCB Schematics Surface
In the weeks to come, NVIDIA will release yet another graphics card based on the G92 graphics processor: the GeForce GTS 240, which bears resemblance with the 8800 GT and 9800 GT. The G92 GPU has near-identical specifications to the said graphics cards, except for that the 55 nm G92b GPU will be used, and with higher reference clock speeds of 675/1674/975 MHz (core/shader/memory). To seat the new GPU, NVIDIA has designed a new reference PCB, the P361 internally called D10P2. The schematic drawings of this has been sourced by VR-Zone.

The P361 PCB covers the basics of accommodating the G92b and eight GDDR3 memory chips connected to the GPU across a 256-bit wide bus. The distinct features of the PCB lie with its VRM area, with a 3+1 phase design. The PCB draws power from a 6-pin PCI-Express power connector. Over to the connectors department, the usual two arrangement of two DVI and one composite is present. There is a single SLI bridge connector for 2-way SLI. With the 55 nm GPU being arguably as cool or cooler than the 65 nm G92 with lower reference clock speeds, one can expect a single slot cooler design to be employed.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15201 - Posted: 12 Feb 2010 | 23:43:07 UTC - in response to Message 15200.
Last modified: 12 Feb 2010 | 23:44:16 UTC

Sorry, it was a few weeks ago that I cam accross it. I thought it was a misprint so I looked into it, but I thought it was genuine after researching it. Just have not managed to find an inexpensive one yet.
The only GTS240 I had, did not work - avoid at all cost!!!

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15204 - Posted: 13 Feb 2010 | 8:59:22 UTC - in response to Message 15201.
Last modified: 13 Feb 2010 | 9:11:22 UTC

You can also safely O.C a GT240 to GPU = 630, Memory (DDR5) 2000. Shader Clock 1580.

For DDR3 the GPU 630, Memory 840, Shader Clock 1580

These figures for clocks are obtained from Nvidias Performance Plugin with ESA and EVGA Precision and readings fron GPU-Z

I did this on Gigabyte cards but can't see why these figure wont work on any 240.

EDIT to add:

Not seeing the percentage increase you're talking mine about 30% but I am running Windows 7 x64 on all my machines.

These aren't massive push them to the edge OC's they are small but stable.
____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline

Profile [AF>Libristes] Dudumomo
Send message
Joined: 30 Jan 09
Posts: 45
Credit: 425,620,748
RAC: 0
Level
Gln
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15206 - Posted: 13 Feb 2010 | 9:40:04 UTC - in response to Message 15204.

In my case, the improvement is better on longer units as the one at 6,118.08 pts. Almost 40% inprovement in such WU (With my GTX275)

Rabinovitch
Avatar
Send message
Joined: 25 Aug 08
Posts: 143
Credit: 64,937,578
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 15216 - Posted: 14 Feb 2010 | 12:21:22 UTC - in response to Message 15138.

Applications: 6.01 is Windows, 6.02 is Linux.


I've interrupted a task for 6.02 app (mysterious Ubuntu 9.10). Too much time for such a too small progress. Is an old application still have a scientific value (my PC can process it, but with a little bit of UI freezes...)?

____________
From Siberia with love!

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15227 - Posted: 15 Feb 2010 | 10:24:38 UTC - in response to Message 15216.

There are no tasks in the ACEMD 2 queue.

GPUGRID Role account
Send message
Joined: 15 Feb 07
Posts: 134
Credit: 1,349,535,983
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 15232 - Posted: 15 Feb 2010 | 11:41:23 UTC - in response to Message 15227.

We're revising the ACEMD 2 application to try to solve the performance problem some Ubuntu, FC Linux users have seen.

MJH

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15271 - Posted: 17 Feb 2010 | 10:49:37 UTC - in response to Message 15232.
Last modified: 17 Feb 2010 | 10:49:55 UTC

New smaller revisions for acemd ver 2 are out. 6.03/6.04.
The Linux application should work better, for Windows we just fixed a minor issue on some input files.

gdf

Rabinovitch
Avatar
Send message
Joined: 25 Aug 08
Posts: 143
Credit: 64,937,578
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 15323 - Posted: 19 Feb 2010 | 6:08:37 UTC - in response to Message 15122.
Last modified: 19 Feb 2010 | 6:09:19 UTC

We have put the first WUs in the new acemd2 application.


I am trying different distributives of GNU/Linux now, and since I tired of all these experiments, temporarily there is Kubuntu 9.10 amd64 installed.

ACEMD2 6.04 is working slow enough on this fresh installation (195.36.03 nVidia driver, as it shown in nvidia X Server Settings) on GTS250, and taking almost full CPU core for it's work. The progress at present is about 2% for 14 minutes.

A 6.70 app working with it's usual speed, but GUI is freezes from time to time anyway. CPU consuptioon is about 30-40% of one core.

Both apps "warming" GPU core equally.

p.s. btw, I could not find a way to force openSUSE 11.2 x64 (KDE, if any) run BOINC and let it use GPU. No one advice from the web was useful and successful, so as me could not find my own solvation.... Does anyone know the remedy?..
____________
From Siberia with love!

Rabinovitch
Avatar
Send message
Joined: 25 Aug 08
Posts: 143
Credit: 64,937,578
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 15325 - Posted: 19 Feb 2010 | 8:34:34 UTC

Oh, by the way, a GUI hangs much rarely and just for a couple seconds while use XRender instaed of OpenGL for desktop effects.
____________
From Siberia with love!

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1451
Credit: 3,575,929,351
RAC: 310,964
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15333 - Posted: 19 Feb 2010 | 18:28:00 UTC

Just had my first task failure (across three 9800GT-class GPUs) since the new app (v6.03) was released. I've mostly been allocated GIANNI_BIND, but this was an OTTO_HERG:

SWAN : FATAL : Failure executing kernel sync [frc_sum_kernel] [700]
Assertion failed: 0, file ../swan/swanlib_nv.cpp, line 203

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.

So here I am ;-)

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15336 - Posted: 19 Feb 2010 | 19:21:11 UTC - in response to Message 15333.

Just had my first task failure (across three 9800GT-class GPUs) since the new app (v6.03) was released. I've mostly been allocated GIANNI_BIND, but this was an OTTO_HERG:

SWAN : FATAL : Failure executing kernel sync [frc_sum_kernel] [700]
Assertion failed: 0, file ../swan/swanlib_nv.cpp, line 203

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.

So here I am ;-)


Usually, this error is overheating or mild overclocking.

gdf

Profile (_KoDAk_)
Avatar
Send message
Joined: 18 Oct 08
Posts: 43
Credit: 6,924,807
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwatwat
Message 15341 - Posted: 20 Feb 2010 | 9:50:17 UTC

how use only
acemdx WU and blok FuLL-Atom ?
____________

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15342 - Posted: 20 Feb 2010 | 9:52:33 UTC - in response to Message 15341.

go to your project preferencs.

gdf

Profile (_KoDAk_)
Avatar
Send message
Joined: 18 Oct 08
Posts: 43
Credit: 6,924,807
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwatwat
Message 15343 - Posted: 20 Feb 2010 | 10:34:42 UTC

NO there only
select
ACEMD
ACEMD ver 2.0
ACEMD beta

where blok FuLL-Atom ?
____________

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15344 - Posted: 20 Feb 2010 | 11:04:03 UTC - in response to Message 15343.
Last modified: 20 Feb 2010 | 11:09:52 UTC

Deselect (remove box tick for) ACEMD


It is confusing that the tasks are called,
Full-atom molecular dynamics v6.71 (cuda23) and ACEMD - GPU molecular dynamics v6.03 (cuda).

While the respective projects are called,
ACEMD and ACEMD ver 2.0.

Full-atom molecular dynamics v6.71 (cuda23) uses ACEMD
ACEMD - GPU molecular dynamics v6.03 (cuda) uses ACEMD ver 2.0

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1947
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 15345 - Posted: 20 Feb 2010 | 12:33:02 UTC - in response to Message 15344.
Last modified: 20 Feb 2010 | 12:33:14 UTC

In practice select only:
ACEMD ver 2.0

The old application will be on until we have finished the old tasks.
gdf

palmss
Send message
Joined: 28 Aug 08
Posts: 7
Credit: 60,897,550
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15346 - Posted: 20 Feb 2010 | 12:47:39 UTC
Last modified: 20 Feb 2010 | 12:58:45 UTC

Hi

I have a problem with the new application, all WU crashing after a few second. But with the old application, it works without problem. Have you an idea about this problem ?

I'm on vista 32 bits with a 8800 GT and 185.85 driver

Thanks you

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1451
Credit: 3,575,929,351
RAC: 310,964
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15347 - Posted: 20 Feb 2010 | 13:46:08 UTC - in response to Message 15336.

Just had my first task failure (across three 9800GT-class GPUs) since the new app (v6.03) was released. I've mostly been allocated GIANNI_BIND, but this was an OTTO_HERG:

SWAN : FATAL : Failure executing kernel sync [frc_sum_kernel] [700]
Assertion failed: 0, file ../swan/swanlib_nv.cpp, line 203

This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.

So here I am ;-)


Usually, this error is overheating or mild overclocking.

gdf

OK, that was on my factory-overclocked Zotac AMP 9800GTX+ - which has nevertheless completed 12 GIANNI_BIND in succession.

So, what are you going to say about this OTTO_HERG - same error message on an absolutely stock Zotac 9800GT, no overclocking at all, and a run of 10 successful completions - mostly, but not exclusively, GIANNI_BIND - to its name?

Please spare me the boilerplate answer this time.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15348 - Posted: 20 Feb 2010 | 14:14:54 UTC - in response to Message 15346.

Palmss, you probably need to update your driver.

MrS
____________
Scanning for our furry friends since Jan 2002

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1451
Credit: 3,575,929,351
RAC: 310,964
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15349 - Posted: 20 Feb 2010 | 14:59:22 UTC - in response to Message 15348.

Palmss, you probably need to update your driver.

MrS

Why are the new applications being sent out under plan_class (cuda), instead of (cuda23), if they need the new drivers?

You're only sending out cufft.dll.2.2 and cudart.dll.2.2: those should be OK with 185.85 driver, shouldn't they?

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15350 - Posted: 20 Feb 2010 | 15:48:13 UTC - in response to Message 15349.

The nomenclature is wanting, as is the application version numbering system, however it will simplify soon; we are presently using up the last of the old tasks. So very shortly we will only be left with the (Up to 60% faster) ACEMD - GPU molecular dynamics tasks (ACEMD ver 2.0), and occasional Beta work.

What is the minimum CUDA/driver version Required to run these New tasks?


PS. A few days ago I did get one error on a GT240 (SWAN: FATAL : swanMalloc failed). It is very rare that I get any errors on these devices; but it was application version 6.01.
I am now using NVidia’s most up to date Beta and overclocking five GT240s with no problems.

Name3-GIANNI_BIND_650-79-100-RND5157_0 Workunit1170404 Created16 Feb 2010 9:20:51 UTC Sent16 Feb 2010 9:20:51 UTC Received16 Feb 2010 18:05:49 UTC Server stateOver OutcomeClient error Client stateCompute error Exit status-40 (0xffffffffffffffd8) Computer ID63357 Report deadline21 Feb 2010 9:20:51 UTC Run time16100.420681 CPU time1031.166 stderr out
<core_client_version>6.10.18</core_client_version>
<![CDATA[
<message>
- exit code -40 (0xffffffd8)
</message>
<stderr_txt>
# There are 2 devices supporting CUDA
# Device 0: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
# Device 1: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
MDIO ERROR: cannot open file "restart.coor"
# There are 2 devices supporting CUDA
# Device 0: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
# Device 1: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
# There are 2 devices supporting CUDA
# Device 0: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
# Device 1: "GeForce GT 240"
# Clock rate: 1.53 GHz
# Total amount of global memory: 536870912 bytes
# Number of multiprocessors: 12
# Number of cores: 96
SWAN: FATAL : swanMalloc failed

Rabinovitch
Avatar
Send message
Joined: 25 Aug 08
Posts: 143
Credit: 64,937,578
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 15354 - Posted: 21 Feb 2010 | 6:46:50 UTC

Seems that I should post it in this thread...

Strange error while WU processing with 6.06 application (Linux). What's wrong?
____________
From Siberia with love!

Tom Philippart
Send message
Joined: 12 Feb 09
Posts: 57
Credit: 23,376,686
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 15370 - Posted: 22 Feb 2010 | 18:47:32 UTC

I don't see a speedup on this computer:
http://www.gpugrid.net/results.php?hostid=50946

what's wrong?

on my 9600GT I clearly see a speed increase of a bit over 50%

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15371 - Posted: 22 Feb 2010 | 19:12:07 UTC - in response to Message 15370.

There is clearly a speedup it's just not so pronounced.


____________
Radio Caroline, the world's most famous offshore pirate radio station.
Great music since April 1964. Support Radio Caroline Team -
Radio Caroline

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15372 - Posted: 22 Feb 2010 | 19:55:53 UTC - in response to Message 15371.

You also are trying to compare two different WU types.
The TONIs use to take quite a bit more than the IBUCHs, take a look at the point differential.
____________
Thanks - Steve

palmss
Send message
Joined: 28 Aug 08
Posts: 7
Credit: 60,897,550
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15397 - Posted: 24 Feb 2010 | 0:30:51 UTC - in response to Message 15348.

Thanks I installed the 196.21 version and it works very well :)

Post to thread

Message boards : Graphics cards (GPUs) : The new application is out (acemd2)

//