Welcome to MilkyWay@home

Posts by [AF>HFR>RR] ThierryH

1) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29820)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
It is for technical reasons....

CC 1.3....

P.S.

I confirm all previous results....

MW takes 5m 50s when crunching on GPU alone.....

When with other cpu project it takes.... even 12m 30s.....

So here is great area for coders to improve app...


Nice to see someone else confirm my results :)
2) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29815)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
The GTX 285 that I have takes about 5:24 per WU with no load on the CPU and leaving the computer idle, this is with the WU that begin with gs_s222_3s.


Yes, but I'm talking outside Milkyway load on CPU, like another BOINC project.
3) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29809)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
My GTX295 is doing the WUs in about 7 minutes under XP x64 sp2 w/Boinc 6.10.1 and 190.62(WHQL).


Could you make a test with idle CPU and another with CPU under full load ?

Thanks,
Thierry.
4) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29805)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
After several tests, here's what I see with calculation time:

1/ Q9450 (4 cores) and one GTX260:

- When CPU is idle => 5:45 mn.
- When 4 cores are running => 12:55 mn
- When 3 cores are running => 5:45 mn

2/ I7 920 (4 cores, 8 threads) and two GTX295 (4 GPUs):

- When CPU is idle => 5:38 mn (same time on each GPU)
- When 8 threads are running => 9:30 mn
- When 7 threads are running => 9:30 mn
- When 6 threads are running => 5:38 mn

I hope it could help.

Thierry.
5) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29800)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:

The estimated calculation time? If so I'm not sure how that is calculated, the progress bar is the best estimate for how things are progressing.


On my Q9450/GTX260, my estimation time is 12:55 when it's running full CPU. When I stop the CPU project, calculated time is 5:45 and estimation time start to decrease.
To be sure, i reduce the number of CPU for BOINC from 4 to 3 => calculation time continue to be 5:45 mn.
Another thing, When CPU is idle, calculation time on GTX295 is close to GTX260.

Thierry.
6) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29795)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
Anthony, I found something wrong. When my CPU (i7) is working full with other BOINC projects, calculation time for Milkyway/CUDA is 9.5 minutes. When my CPU is idle, calculation time is 5.5 minutes. Close to two times difference !

Edit: confirmed on my second computer.
7) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29789)
Posted 27 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
I have a Quad Intel processor, NVIDIA GPU running under 190.62 and is CUBA 2.3, I use BOINC 6.6.36. I upgraded to the BOINC 6.6.36 yesterday and since then I have not been able to get new work units.
I receive a message stating my card is not compatible (CUBA 2.3 and not 1.3).

How do I disable the GPU / CUBA function? The only option I can find is USE GPU WHILE COMPUTER IS IN USE. I have not selected this option.

Do I just have to Detach this quit this project? This is dumb if this is the case. I hope I am just having trouble finding the option.

Thanks for any help.


Jim, as I can see on your computers page, you have only series 8000 cards which are not compatible with Milkyway/CUDA application (see fisrt post of this thread).
8) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29763)
Posted 26 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
It should work correctly, I'm not sure how BOINC will react to the multi-gpu setup. If there are compute errors please let us know.


It's working well. I have 2 GTX295 (4 GPUs) on a winXP64 box. No problem.
9) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29761)
Posted 26 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
To be honest the CUDA application is not running at its full potential, the computation is sliced into pieces so that it does not occupy the whole GPU in order to decrease the slowdown of the user interface. There are two ways the application can be executed

1. Using all of the GPU's power, work completes faster but the OS is pretty much unusable because it takes ~2s for the user interface to update, meaning it can only run when the computer is idle
2. Using a partial amount of the GPU's power, work takes longer, updates to the OS user interface are still slow, but hardly noticeable in most situations, meaning it can be run when the computer is in use

I chose to go with #2, but which one makes more sense? I'll look into the feasibility of having this as a project specific preference if there is a 50/50 divide between the two choices.


As you said, #2 still slow down the computer. Best way is perhaps to uncheck "Use GPU while computer is in use" in BoincManager preferences. If users must do that, #1 is better.
For me, GPUs are on crunch boxes with no screen. So #1 solution is the best, but it's just my personal case.
10) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29756)
Posted 26 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
8mn30 on GTX295@640/1400 and 15mn on GTX260.
11) Message boards : Number crunching : CUDA Application for 32 bit Windows (Message 29754)
Posted 26 Aug 2009 by Profile [AF>HFR>RR] ThierryH
Post:
Seems very slow comparing to ATI application :(
But it's working well.
12) Message boards : Number crunching : Compute Errors (Message 24593)
Posted 8 Jun 2009 by Profile [AF>HFR>RR] ThierryH
Post:
Looks like the searches are stopped, we'll not do 3 stream runs until the ATI code is fixed :)


Thanks Travis. It was a very good decision.
Now, Cluster Physik gave us fixed code since 3 days. Perhaps it's time to restart 3 stream runs. This kind of WUs is longer to calculate than others. It could give more work for everyone.

Thank you,
Thierry.
13) Message boards : Application Code Discussion : PPC64 Optimizations (Message 15182)
Posted 13 Mar 2009 by Profile [AF>HFR>RR] ThierryH
Post:

In principle the SPEs could be quite fast for single precision tasks. But last time I checked the SPEs of the Cell version used in the PS3 are not that well suited to double precision, isn't it? It works, but performance isn't great (maybe twice that of the PPE).

One would like to use a PowerXCell 8i for MW. Would be faster than a GTX280. Anyone tried to mod his PS3 with such a beast? Just kidding.


SPEs have 128 bit registers. Each ops can be executed on one quad precision, on two double precission or on four single precision. So, if the code could be vectorized, double precision is only two times slower than single.

Perhaps it could be more quicly optimized if you could send me cpu sources part of your ATI optimization. Thanks in advance.

Thierry.
14) Message boards : Application Code Discussion : PPC64 Optimizations (Message 15180)
Posted 13 Mar 2009 by Profile [AF>HFR>RR] ThierryH
Post:
I plan to work on the client with SPE support whenever I have time.


I'm also trying to do it. Perhaps we can work togather ?

15) Message boards : Application Code Discussion : found a small memory error in the code (Message 13433)
Posted 1 Mar 2009 by Profile [AF>HFR>RR] ThierryH
Post:
Other memory error in evaluation_optimized.c :

Function free_constants, lines :

for (i = 0; i < ap->number_streams; i++) {
free(xyz[i]);
}

Must be replace by :

for (i = 0; i < ap->convolve; i++) {
free(xyz[i]);
}

With this error, each work unit create a memory leak of 1.5 Ko.

Thierry.
16) Message boards : Application Code Discussion : What's needed to compile the app? (Message 12719)
Posted 24 Feb 2009 by Profile [AF>HFR>RR] ThierryH
Post:
It seems my linux was in a bad state. I reformat disk and re-install it. Now compile runs well until the end.
Thank for your help,
Thierry.

17) Message boards : Application Code Discussion : What's needed to compile the app? (Message 12390)
Posted 22 Feb 2009 by Profile [AF>HFR>RR] ThierryH
Post:

You need to install the libstdc++ package.


Thanks. I'll try.
Thierry.

18) Message boards : Application Code Discussion : What's needed to compile the app? (Message 12374)
Posted 22 Feb 2009 by Profile [AF>HFR>RR] ThierryH
Post:
Ok, I fix the trouble by runninf "make install" to compile Boinc. But I have always a trouble: now, I have "can't find -lstdc++" when I link Milkyway.
Any idea ?

Thanks,
Thierry.

19) Message boards : Application Code Discussion : What's needed to compile the app? (Message 12365)
Posted 22 Feb 2009 by Profile [AF>HFR>RR] ThierryH
Post:

The linker obviously does not find your already compiled BOINC libs. Most probably you have to correct the path in the make file so it finds them.


These two files (boinc_api and boinc) are nowere on the disk.

Thierry


I had that problem too...
Go to Message 6819.


Already made it yesterday with no success.

Thierry.

20) Message boards : Application Code Discussion : What's needed to compile the app? (Message 12341)
Posted 22 Feb 2009 by Profile [AF>HFR>RR] ThierryH
Post:

The linker obviously does not find your already compiled BOINC libs. Most probably you have to correct the path in the make file so it finds them.


These two files (boinc_api and boinc) are nowere on the disk.

Thierry


Next 20

©2024 Astroinformatics Group