GPU Project Status

Message boards : News : GPU Project Status

To post messages, you must log in.

1 · 2 · 3 · 4 · Next

AuthorMessage
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 9682 - Posted: 17 Jun 2014, 13:43:22 UTC

As many of you have noticed, our GPU project has been re-enabled after the update last week.
With the apparent error of our last release being fixed, I'm still not satisfied with the failing rate of GPU tasks, especially on NVIDIA cards.

I think there are currently three main issues:

1. Incompatibility to old AMD GPUs.
We're suspecting our latest version of poemcl won't run properly on cards of the Radeon 4000 series, which only have a limited OpenCL support.
If you are using such a GPU, and it works however, please let me know which OS and driver version you are using.
If there is no working setup, we will exclude the Radeon 4000 series in future releases.

2. Multi-GPU NVIDIA hosts still fail when using more than one graphics card for poemcl tasks.
If you have attached a host with multiple NVIDIA cards to POEM@HOME, please configure your client to use only one of them for our project, see http://boinc.berkeley.edu/wiki/client_configuration headword exclude_gpu.
This is a well known problem which should have been fixed with the last release, but obviously it is a sticky one :(

3. There is a new error, which may be related to incompatible OpenCL library versions.
If your host uses only a single GPU, but still fails with poemcl tasks, please download one of the following files (depending on your OS):
boinc.fzk.de/userdocs/temp/pcl_201_test_linux.tar.gz
boinc.fzk.de/userdocs/temp/pcl_201_test_win.zip
Extract the files to a new folder, run the executable (if it works, this may take some minutes), and post the content of stderr.txt in this thread.

Thanks for your help!
I hope we can improve the application soon with your feedback.
ID: 9682 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Thomas Koch
Project administrator
Project developer
Project scientist

Send message
Joined: 17 Feb 12
Posts: 436
Credit: 37,847
RAC: 0
Message 9684 - Posted: 17 Jun 2014, 13:55:11 UTC

An example for excluding a GPU in the client configuration can be found in this thread.
ID: 9684 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
lolmaster

Send message
Joined: 27 Jun 08
Posts: 1
Credit: 8,198,947
RAC: 0
Message 9685 - Posted: 17 Jun 2014, 14:15:02 UTC

I'm using a ATI 4870, Windows 7 x64, latest driver (13.9). Haven't got finished the WU jet but so far the WU didn't fail.
ID: 9685 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile BlueGhost

Send message
Joined: 19 Aug 11
Posts: 11
Credit: 56,429,329
RAC: 0
Message 9691 - Posted: 17 Jun 2014, 21:03:47 UTC

OpenCL workunit runs without any problems

PC config:

AMD Athlon II X4 610e / 8GB DDR3 1333 RAM / Windows7 64Bit
AMD Radeon 7750 / 1GB GDDR5 RAM / Catalyst Version 13.1

BOINC Version 7.2.42 (64Bit)

GPU Utilization ~ 66% (1 CPUs + 1 ATI GPU)

Data from the GPU OpenCL 2.01 task:

poempp_collective_move_set_1403016202_524560033_0

Runtime 16.218,07 sec / CPU Time 5.901,19 sec
ID: 9691 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Nuadormrac

Send message
Joined: 29 Sep 09
Posts: 95
Credit: 72,775,187
RAC: 0
Message 9692 - Posted: 17 Jun 2014, 21:08:45 UTC
Last modified: 17 Jun 2014, 21:09:24 UTC

Here they're running, but on Windows, or at least Windows 7 there is still the same problem that they take near 6 hours to run. The 6,500 credits per task is not enough for this to grant more then Einstein@home level credits in return for the time spent per task, which is of course at the very low end and substantially below the credit return of most every GPU project out there...
ID: 9692 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jim1348

Send message
Joined: 4 Jul 12
Posts: 104
Credit: 195,960,357
RAC: 0
Message 9693 - Posted: 17 Jun 2014, 21:19:51 UTC

Intel Haswell i7-4771, Win7 64-bit
AMD Radeon HD 7790 (non-overclocked), Catalyst 14.4
BOINC 7.4.0 x64

Name poempp_collective_move_set_1403016200_1568852321_0
Run time 9,335.95
CPU time 3,245.13
Credit 6,500.00
Application version POEM++ OpenCL version v2.01 (opencl_ati_100)

ID: 9693 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
laami Änte

Send message
Joined: 7 Feb 12
Posts: 4
Credit: 251,556
RAC: 0
Message 9696 - Posted: 18 Jun 2014, 8:25:19 UTC

Computer Dell Precision T1799
OS MS Windows 7 Ultimate 64bit, SP1
Graphic Card NVIDIA Quadro K600, 1GB,
Display Driver NVIDIA 9.18.13.3276, 3-432014

BOINC 7.2.33
Task Name poempp_collective_move_set_1403015365_393344081_0

The display driver crashes (-> black screen for some seconds) every time the POEM task is supended. I had to abort the task and exclude POEM from getting new tasks.



ID: 9696 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
SAH

Send message
Joined: 19 Aug 11
Posts: 1
Credit: 42,768,352
RAC: 0
Message 9697 - Posted: 18 Jun 2014, 12:06:56 UTC

My OpenCL workunit finished without any problems

PC config:

Intel Core i5-2500K 4x 3,3 GHz / 8GB DDR3 1333 RAM / Windows7 64Bit
AMD Radeon HD6700 / 1GB GDDR5 RAM / Catalyst Version 13.1

BOINC Version 7.2.42 (64Bit)

Task Name: poempp_collective_move_set_1403015294_210580731_0

Runtime 35.796,58 sec / CPU Time 4.319,04 sec
ID: 9697 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Dirk Broer

Send message
Joined: 30 Dec 11
Posts: 6
Credit: 39,069,777
RAC: 0
Message 9699 - Posted: 18 Jun 2014, 18:57:40 UTC - in response to Message 9682.  

1. Incompatibility to old AMD GPUs.
We're suspecting our latest version of poemcl won't run properly on cards of the Radeon 4000 series, which only have a limited OpenCL support.
If you are using such a GPU, and it works however, please let me know which OS and driver version you are using.
If there is no working setup, we will exclude the Radeon 4000 series in future releases.


I've got my HD 4770 working on this project running the AMD legacy driver under 64-bit Win7
ID: 9699 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile ritterm
Avatar

Send message
Joined: 28 Jan 09
Posts: 31
Credit: 45,540,617
RAC: 0
Message 9701 - Posted: 19 Jun 2014, 1:16:09 UTC
Last modified: 19 Jun 2014, 1:17:05 UTC

So far so good for me with this host:

CPU: AMD FX-8150
GPU: NVIDIA GeForce GTX 550 Ti (1024MB) driver: 335.23 OpenCL: 1.01
OS: MS Win7-64
BOINC: 7.2.42

Running 2 tasks at a time (with 2 CPUs) took about 9.5 hours to complete. I haven't completed one solo. Running two pretty much maxes out the GPU @ 99% load and causes very noticeable (but not terrible) video lag. With the old GPU tasks, any lag was virtually unnoticeable, even running two at a time. However, I don't remember what the GPU load was. Find completed task details here.

I see that the credit/task is still much lower than it used to be on a per hour basis. Not that big a deal, just wondering if that's how it's going to be from now on.

Cheers,

MarkR
ID: 9701 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Alvaro Huerta Martin

Send message
Joined: 13 Jun 13
Posts: 3
Credit: 28,516,669
RAC: 0
Message 9705 - Posted: 19 Jun 2014, 11:26:33 UTC - in response to Message 9682.  
Last modified: 19 Jun 2014, 11:27:10 UTC

Hi, Herr Koch and everybody!

A couple of weeks ago or so, I ran a task which ended with error after a bit more than 5 hours. A new wu was finished yesterday without problems (about 11 hours), but another wu downloaded yesterday, gave me error when I switched on the PC today (20 minutes calculated.) (My machine has a Intel Core 2 Duo Processor @ 2.8 GHz, Win 7 Home Premium 64-bit, and a nVidia GeForce 9600 GT, 1 GB, with the latest nVidia stable drivers: 337.88.) So, I follow your instructions (well, I copied, from the POEM folder inside BOINC, the library "libOpenCL.dll_2.01" to the folder where is the extracted zip, and I renamed it as "libOpenCL.dll" to be able to run the executable, right?), and here you have the complete "stderr.txt" text:


12:48:11 (3144): Can't open init data file - running in standalone mode
--- CommandLine Options ---
Infile: simona_input.xml
Outfile: simona_input_out.xml
XML Snapshot Outfile: simona_input_snap.xml
PDB Outfile: simona_input_snap.pdb
---------------------------
simona_main::parse --- done reading input
simona_main::parse --- no walltime is set
simona_main::parse --- parsing RNG
<SIMONA-RNG> Seeding RNG with seed: 8336380
<SIMONA-RNG> Seeding srand with seed: 8336380
<SIMONA-RNG> Reseeding srand with: 4246192805
simona_main::parse --- parsing moves
simona_main::parse --- parsing forcefields
simona_main::parse --- parsing forcefield 0 Name: nano
Forcefield Evaluation Order ---
PitPotential
LennardJonesCL
------------------------------
simona_main::parse --- parsing PDB to string
simona_main::parse --- parsing configuration
PBC box x = 0
PBC box y = 0
PBC box z = 0
PBC box = 0
Read coordinates, xyz-max =
[ 47.32]
[43.112999]
[40.963001]
xyz-min =
[-43.596001]
[-40.963001]
[-40.963001]
simona_main::parse --- parsing algorithm
RepeatedMove
parsing: RepeatedMove 0
parsing: TransformationSequence 1
parsing: TransformationSequence 2
parsing: BOINCUpdateOutput 3
parsing: DoSnapshot 4
parsing: RecalcEnergies 5
parsing: EnergyOutput 6
parsing: MetadataOutput 7
parsing: ConditionalTransformation 8
MetropolisAcceptanceCriterion using Forcefield 0
parsing: TransformationChoice 9
parsing: SetCollectiveTranslationMove 10
Done parsing weights, number: 1
simona_main::parse --- done parsing
Found 1 OpenCL Platforms
Platform is: 0x3550538
NVIDIA Corporation
FULL_PROFILE
OpenCL 1.1 CUDA 6.0.1
NVIDIA CUDA
cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_d3d9_sharing cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll
Plattform 0 contains:
----- ID 0 ----
Available: 1
CL_DEVICE_GLOBAL_MEM_SIZE 1073741824
--------------
Compiled 2 Kernels.
Kernel Nr. 0
lennard_jones_kernel
Kernel Nr. 1
lennard_jones_kernel_dynamic
Rounding global size to 12224.
X <EH> LennardJonesCL PitPotential Total
BEGIN <E> 130.947 3.29582e+006 3.29595e+006
Recording Snapshot at step Nr. 0
<e> 0 <E> 130.95 3295815.75 3295946.75
<e> 100 <E> 124.59 3295403.75 3295528.25
<e> 200 <E> 121.47 3295403.75 3295525.25
<e> 300 <E> -47.19 3294925.25 3294878.00
<e> 400 <E> -49.31 3294837.00 3294787.75
Recording Snapshot at step Nr. 500
<e> 500 <E> -55.42 3294545.50 3294490.00
<e> 600 <E> -61.16 3294545.25 3294484.00
<e> 700 <E> -64.48 3294526.50 3294462.00
<e> 800 <E> -66.01 3294289.00 3294223.00
<e> 900 <E> -68.64 3294185.75 3294117.00
<SIMONA-RNG> Saving Random Number Generator state
<SIMONA-RNG> Reseeding srand with: 3951144719
END <E> -72.4666 3.29404e+006 3.29397e+006
--- Energysumterm timings ---
PitPotential 18.237
LennardJonesCL 130.184
-----------------------------
Algorithm Destructor Called
Full runtime: 157.094
SIMONA run finished.
12:50:49 (3144): called boinc_finish(0)


Let's hope all this be useful
Cheers,

Alvaro.
ID: 9705 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dskagcommunity
Avatar

Send message
Joined: 26 Oct 11
Posts: 117
Credit: 368,495,687
RAC: 0
Message 9706 - Posted: 19 Jun 2014, 13:01:55 UTC
Last modified: 19 Jun 2014, 13:02:28 UTC

4 units on xp, 285gtx and q6600 seemed to run ok.

Now only more Workunits are needed to test more ;)
DSKAG Austria Research Team: http://www.research.dskag.at



Crunching for my dead Dog who had "good" Braincancer..
ID: 9706 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 30 Dec 09
Posts: 236
Credit: 48,564,069
RAC: 0
Message 9707 - Posted: 19 Jun 2014, 13:05:09 UTC - in response to Message 9692.  

The 6,500 credits per task is not enough for this to grant more then Einstein@home level credits in return for the time spent per task, which is of course at the very low end and substantially below the credit return of most every GPU project out there...


And below the credit return of "old" gpucrystal wus!!
ID: 9707 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
John C MacAlister

Send message
Joined: 14 May 11
Posts: 40
Credit: 26,199,364
RAC: 0
Message 9708 - Posted: 19 Jun 2014, 13:05:39 UTC - in response to Message 9682.  

Thanks, Thomas.

#2 is the issue which causes me most trouble. It it would be just wonderful to be able to use both my GTX-650 Ti GPUs together for POEM WUs.

Priority 1 for a fix??

Thanks, John
ID: 9708 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Nuadormrac

Send message
Joined: 29 Sep 09
Posts: 95
Credit: 72,775,187
RAC: 0
Message 9710 - Posted: 19 Jun 2014, 17:08:52 UTC - in response to Message 9707.  
Last modified: 19 Jun 2014, 17:09:37 UTC

The 6,500 credits per task is not enough for this to grant more then Einstein@home level credits in return for the time spent per task, which is of course at the very low end and substantially below the credit return of most every GPU project out there...


And below the credit return of "old" gpucrystal wus!!


Also below the credit return of PrimeGrid, DistrtGen (when they had GPU work) and GPUgrid as well, basically most everything... Basically it's on par with Einstein and SETI@home, perhaps a tad more then WCG...
ID: 9710 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
JugNut

Send message
Joined: 1 Sep 11
Posts: 15
Credit: 728,763,887
RAC: 0
Message 9711 - Posted: 19 Jun 2014, 18:19:06 UTC - in response to Message 9710.  
Last modified: 19 Jun 2014, 18:20:07 UTC

I agree a credit improvement is in order. My HD 7970 used to crunch 10 of the old POEM work units in 35 minutes now it's 2 every 2hrs 15mins & thats with a relatively fast card. Many other people are taking from 4 to 10+ hours to complete just one work unit.

It's not just about the credit, i'll crunch here anyway because of the value of the work but I still think a fair credit system couldn't hurt. It costs nothing to implement & can only attract more people to the cause. For those that say they don't care about credit well it should mean nothing to them either good or bad but to the rest of us it's an added bonus.

Just my 2 cents.
ID: 9711 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
ExtraTerrestrial Apes
Volunteer moderator
Avatar

Send message
Joined: 18 Jan 09
Posts: 454
Credit: 581,676,488
RAC: 0
Message 9713 - Posted: 19 Jun 2014, 20:10:06 UTC
Last modified: 19 Jun 2014, 20:10:15 UTC

I'm not involved in the POEM development, but I'm pretty sure the new algorithm with extended features can not push as many raw Flops per second as the old one. Hence it gets awarded less credit, although the scientific value of the results may be far higher than in the old runs. Flop counting is considered fair payment, yet has the unpleasant side effect that projects award the more credit per time the simpler their calculations are.

Well, it could also simply be caused by miscounted Flops, as one never really knows how many of them are performed within libraries etc. And there is some wiggle room for special credits adjustments, as evidenced by e.g. the credit bonus given at GPU-Grid for quickly returned results.

MrS
Scanning for our furry friends since Jan 2002
ID: 9713 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mad_Max

Send message
Joined: 24 Jan 12
Posts: 27
Credit: 656,410,733
RAC: 0
Message 9715 - Posted: 19 Jun 2014, 23:36:14 UTC - in response to Message 9682.  


3. There is a new error, which may be related to incompatible OpenCL library versions.
If your host uses only a single GPU, but still fails with poemcl tasks, please download one of the following files (depending on your OS):
boinc.fzk.de/userdocs/temp/pcl_201_test_linux.tar.gz
boinc.fzk.de/userdocs/temp/pcl_201_test_win.zip
Extract the files to a new folder, run the executable (if it works, this may take some minutes), and post the content of stderr.txt in this thread.

Thanks for your help!
I hope we can improve the application soon with your feedback.

On windows this test fails immediately and trow error message about can not find libOpenCL.dll

AMD GPU drivers for Windows not contain such file. (but have OpenCL.dll and amd_OpenCL32.dll/amd_OpenCL64.dll)
AFAIK libOpenCL.dll from linux drivers/SDK.
ID: 9715 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile JumpinJohnny

Send message
Joined: 24 Apr 13
Posts: 1
Credit: 3,199,008
RAC: 0
Message 9716 - Posted: 20 Jun 2014, 14:36:13 UTC - in response to Message 9682.  

1. Incompatibility to old AMD GPUs.
We're suspecting our latest version of poemcl won't run properly on cards of the Radeon 4000 series, which only have a limited OpenCL support.
If you are using such a GPU, and it works however, please let me know which OS and driver version you are using.
If there is no working setup, we will exclude the Radeon 4000 series in future releases.

I'm running the new opencl on a AMD ATI Radeon HD4870 using driver: 1.4.1734 OpenCL: 1.00, on Windoz7Ultimate x64sp1 OS. It seems to be doing fine. I'll let you know if it doesn't finish.
Computer Details
Please keep this support for future releases so I can contribute to POEM with this GPU.
ID: 9716 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mad_Max

Send message
Joined: 24 Jan 12
Posts: 27
Credit: 656,410,733
RAC: 0
Message 9717 - Posted: 21 Jun 2014, 19:29:14 UTC
Last modified: 21 Jun 2014, 20:11:20 UTC

AMD Radeon HD 5750 running on Windows XP with old 12.1 drivers (last with OpenCL support on WinXP)

So far seems everything working well: 2 WUs completed and validated successfully, 3rd in progress.

Compared with the old (1.x) version I see the differences:
- bigger load and heating on the GPU (especially if compare running in one stream without app_info/app_config)
- significantly lower CPU usage
- more lags to user desktop (with previous ver. it almost not affect user experience so i run POEM GPU 24/7, with new i set run GPU only while computer not in use)
- MUCH less credit (single new WU is ~2x more credit, but have MUCH longer runtimes: ~10 hours vs 0.5 hours)
ID: 9717 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
1 · 2 · 3 · 4 · Next

Message boards : News : GPU Project Status


Copyright © 2017 KIT-INT