I haven't thought of Linux or used it in many years. I remember way back when there used to be large differences in Linux and Windows speed with various projects. How is it today generally? I know there are detailed lists but that gets very confusing in trying to get an overview of things. I am specifically asking here about seti. Thanks

---

merle - vote yes for freedom of speech

I haven't heard much about it lately. I think for some specific configurations, like 64-bit CUDA crunching, Linux had a marginal-at-best advantage (less than 5%, so with all the variables involved, there's no real way to say for sure that gain is because of Linux itself).

I think with the Lunatics development, it is pretty well equal across the board using the same builds in different environments (windows, linux, mac). There might be some small variances, but nothing that makes the RAC junkies jump to a different OS.

---

Linux laptop:
record uptime: 1511d 20h 19m (ended due to the power brick giving-up)

I'm thinking about comparing my Ubuntu 12.04 system to the Vista system that's on the same machine. Ubuntu was slightly faster the last time I compared them. With the new oclFFT_plan setting Ubuntu just took a major leap ahead. I'm interested in seeing if Vista will see the same increase. The only difference is Ubuntu is 64 bit while Vista is 32 bit. It really shouldn't matter for GPU tasks though...

So, here are the hosts to compare;
Ubuntu 12.04, http://setiathome.berkeley.edu/results.php?hostid=7258715
Ubuntu at Beta, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72013&offset=60&show_names=0&state=0&appid=
Vista, http://setiathome.berkeley.edu/results.php?hostid=6797524
Vista at Beta, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72451

Those are all the same machine. The Vista times are using the -oclFFT_plan 256 16 256 setting, the Vista at Beta times are Not using the -oclFFT_plan 256 16 256 setting. I used Lunatics 0.43 to install the Apps on the Vista host.
The Ubuntu times at Beta are also Not using the oclFFT_plan settings.

There is a problem trying to convince BOINC that it really should download enough tasks on the Vista host to use all 3 cards. I've noticed that before when using a host that hasn't been used in a while. It's like pulling teeth to get BOINC to download any work. In this case APv7 hasn't been used, so, the time estimates are really long adding to the problem. Even with a 10 day cache BOINC wouldn't download enough work to keep all 3 cards working. I had to enter a FLOPS setting to get BOINC to download enough work to keep all the cards working.

So, there was a lot of stopping and starting, and when you restart BOINC it tends to swap the tasks and cards around, so.....Look for Tasks that haven't been Restarted, 'cause some may have run on the ATI 4670 and the 6770 making the run times worthless. It appears the ATI 4670 doesn't like the oclFFT_plan causing long run times, so, disregard the Device 2 times as they are too long. I'll probably remove that card, and update the driver when I find that round tuit.

Startup;

10/25/2014 6:06:00 AM |  | Starting BOINC client version 7.2.33 for windows_intelx86
10/25/2014 6:06:00 AM |  | CAL: ATI GPU 0: AMD Radeon HD 7700 series (Capeverde) (CAL version 1.4.1720, 2048MB, 2008MB available, 2048 GFLOPS peak)
10/25/2014 6:06:00 AM |  | CAL: ATI GPU 1: ATI Radeon HD 5700/6750/6770 series (Juniper) (CAL version 1.4.1720, 1024MB, 991MB available, 2880 GFLOPS peak)
10/25/2014 6:06:00 AM |  | CAL: ATI GPU 2: ATI Radeon HD 4600 series (R730) (CAL version 1.4.1720, 1024MB, 992MB available, 1152 GFLOPS peak)
10/25/2014 6:06:00 AM |  | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 7700 series (Capeverde) (driver version CAL 1.4.1720 (VM), device version OpenCL 1.2 AMD-APP (923.1), 2048MB, 2008MB available, 2048 GFLOPS peak)
10/25/2014 6:06:00 AM |  | OpenCL: AMD/ATI GPU 1: ATI Radeon HD 5700/6750/6770 series (Juniper) (driver version CAL 1.4.1720 (VM), device version OpenCL 1.2 AMD-APP (923.1), 1024MB, 991MB available, 2880 GFLOPS peak)
10/25/2014 6:06:00 AM |  | OpenCL: AMD/ATI GPU 2: ATI Radeon HD 4600 series (R730) (driver version CAL 1.4.1720, device version OpenCL 1.0 AMD-APP (923.1), 1024MB, 992MB available, 1152 GFLOPS peak)
10/25/2014 6:06:00 AM |  | OpenCL CPU: Intel(R) Core(TM)2 Quad CPU    Q9400  @ 2.66GHz (OpenCL driver vendor: Advanced Micro Devices, Inc., driver version 2.0 (sse2), device version OpenCL 1.2 AMD-APP (923.1))
10/25/2014 6:06:00 AM |  | OS: Microsoft Windows Vista: Ultimate x86 Edition, Service Pack 2, (06.00.6002.00)

I haven't thought of Linux or used it in many years. I remember way back when there used to be large differences in Linux and Windows speed with various projects. How is it today generally? I know there are detailed lists but that gets very confusing in trying to get an overview of things. I am specifically asking here about seti. Thanks

Many years ago, the fastest combination was to run the Windows client on a Linux system using the WINE environment. The native Linux client, and running the Windows client natively on Windows, were both slower.

Since then, The Lunatics did burst upon the scene and have perpetrated many improvements and speedups that even the giants of Intel and AMD came to notice. Ni! :-)

With the great Lunatics work, you shouldn't notice any significant difference between any of the OSes for 'normal' systems. The overhead from the OS should be negligible. A big plus for Linux (and Apple) is that you have no need to run any antivirus that can interfere with normal running.

As already noted however, you can get some rather different speed/utilisation results for multi-GPU behemoths... ;-)


My experience has been that Linux has much more to offer for consistency and for being able to tweak things the way that you want. There are certainly fewer unwanted surprises from "unexpected updates" interfering with the normal running than that seen for Windows! (And I'm very sure various Windows users will 'very strongly' disagree...)

Another plus for Linux is that you can go 'minimal' and have a lean clean system that is dedicated to Boinc and that need not be disturbed by any updates or anti-virus.


Go for what you find the most interesting!

Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

Thanks Martin,
I used Linux for several years. Debian unstable as a matter of fact. I went back to windows because it was easier to administer for a CS novice.

I haven't thought of Linux or used it in many years. I remember way back when there used to be large differences in Linux and Windows speed with various projects. How is it today generally? I know there are detailed lists but that gets very confusing in trying to get an overview of things. I am specifically asking here about seti. Thanks

Many years ago, the fastest combination was to run the Windows client on a Linux system using the WINE environment. The native Linux client, and running the Windows client natively on Windows, were both slower.

Since then, The Lunatics did burst upon the scene and have perpetrated many improvements and speedups that even the giants of Intel and AMD came to notice. Ni! :-)

With the great Lunatics work, you shouldn't notice any significant difference between any of the OSes for 'normal' systems. The overhead from the OS should be negligible. A big plus for Linux (and Apple) is that you have no need to run any antivirus that can interfere with normal running.

As already noted however, you can get some rather different speed/utilisation results for multi-GPU behemoths... ;-)


My experience has been that Linux has much more to offer for consistency and for being able to tweak things the way that you want. There are certainly fewer unwanted surprises from "unexpected updates" interfering with the normal running than that seen for Windows! (And I'm very sure various Windows users will 'very strongly' disagree...)

Another plus for Linux is that you can go 'minimal' and have a lean clean system that is dedicated to Boinc and that need not be disturbed by any updates or anti-virus.


Go for what you find the most interesting!

Happy fast crunchin',
Martin

---

merle - vote yes for freedom of speech

The last time I tried Linux (about 12 months ago) I found that on MB CUDA that the CPU usage was much higher and there was a lot of congestion on the PCIE bus. This was on a machine running 3 x GTX580's crunching on the GPU's only.

Whether this was a problem with the way Linux handles the PCIE bus, the BOINC client or the Linux X41G CUDA app I don't know, but the problems were bad enough to drive me back to XP.

Under XP, the machine was a lot happier and I could run CPU tasks as well.

I haven't written Linux off, I'll try it again one day. I like it as an OS but not as an SAH cruncher.

T.A.

I have been utilizing the spare cycles on a Linux Slackware mail/time/file/web server (no GUI) for SETI a good number of years and have been very happy with the results. This is a CPU only cruncher but all 4 cores are at 100% all the time with no adverse effect on the rest of the server functions when they are needed. The nice thing, as ML1 suggested, is that I can do updates on all the other software without bothering SETI. In most cases I do not even have to kill and restart SETI to when I do updates.


Perhaps a Windows flavor would be a better option but since SETI is a secondary application, Linux it is.

The last time I tried Linux (about 12 months ago) I found that on MB CUDA that the CPU usage was much higher and there was a lot of congestion on the PCIE bus. This was on a machine running 3 x GTX580's crunching on the GPU's only.

Whether this was a problem with the way Linux handles the PCIE bus, the BOINC client or the Linux X41G CUDA app I don't know, but the problems were bad enough to drive me back to XP.

Under XP, the machine was a lot happier and I could run CPU tasks as well.

I haven't written Linux off, I'll try it again one day. I like it as an OS but not as an SAH cruncher.

T.A.

Maybe Eric can offer some advice, he seems to have a similar machine running Linux, http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=72274 Looks as though he's still running APv7s for now.
Or maybe our favorite Team-mate who just brought his machine back to work, http://setiathome.berkeley.edu/show_host_detail.php?hostid=5643864 He has Linux on nVidia.
I have mostly ATIs and I'm getting ready to try the oclFFT_plan on my 3 6800s, http://setiathome.berkeley.edu/show_host_detail.php?hostid=6906726 Hopefully I've conquered the Linux on an EFI Mac problem, I expect the times to be better than in OSX if it doesn't start complaining about disk errors again...

There may be an interesting case in the top hosts: http://setiathome.berkeley.edu/top_hosts.php

One running stock on Windows, one running lunatics on linux. Both have i7 and 4 x GTX780.

Only time will tell ...

But in the mean time You may peek at the daily scores from the links to the Free DC or Boinc stats.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

The last time I tried Linux (about 12 months ago) I found that on MB CUDA that the CPU usage was much higher and there was a lot of congestion on the PCIE bus.

I noticed that MB CUDA under Linux slows to a crawl when the CPU is in use, which may be similar to this observation. This is with both x41g and x41zc that I compiled myself.

...but the problems were bad enough to drive me back to XP.

Under XP, the machine was a lot happier and I could run CPU tasks as well.

My understanding is that WinXP, as an older kernel, doesn't have the high CPU requirements to drive GPU tasks. I would use WinXP as well, if it wasn't limited to 32-bit (yes, I know there's an x64 version but it's not really 'mainstream') and was still 'supported'. Back on WinXP I didn't have to 'free' CPU cores for anything other than AstroPulse on NV hogging 100% CPU (which is now alleviated with the -use_sleep switch).

That said, I still think Linux is a reasonable alternative as an OS to drive BOINC work, although individual application support is still for Windows as a priority unless you're willing/able to compile the applications yourself.

---

Soli Deo Gloria

So, here are the hosts to compare;
Ubuntu 12.04, http://setiathome.berkeley.edu/results.php?hostid=7258715
Ubuntu at Beta, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72013&offset=60&show_names=0&state=0&appid=
Vista, http://setiathome.berkeley.edu/results.php?hostid=6797524
Vista at Beta, http://setiweb.ssl.berkeley.edu/beta/results.php?hostid=72451

Those are all the same machine. The Vista times are using the -oclFFT_plan 256 16 256 setting, the Vista at Beta times are Not using the -oclFFT_plan 256 16 256 setting. I used Lunatics 0.43 to install the Apps on the Vista host.
The Ubuntu times at Beta are also Not using the oclFFT_plan settings...

Well, the ATI 4670 finally finished one complete task, http://setiathome.berkeley.edu/result.php?resultid=3801318360
Run time: 10 hours 59 min 33 sec
CPU time: 15 min 44 sec
Validate state: Valid
Ouch. Well, at least it was valid even though it took much longer than it did in Beta without the oclFFT_plan.
I was hoping for a 20% speedup, not a 90% slowdown. My old NV 8800 GT doesn't like the oclFFT_plan either, and it's results are invalid. So, the ATI 4670 is gone. I went to Update the Vista driver to the same as Ubuntu BUT, the last driver AMD offers for Vista is 13.12. So, we now have Vista with the same cards as Ubuntu and the latest driver offered by AMD.
The results so far are clouded by the Noisy data, but, it would appear Ubuntu is ahead due to the Unblanked ATI 6770 times. Ubuntu is running them in around 54 minutes whereas Vista is taking right at an hour.

Let the Games continue!
Startup;

10/25/2014 7:06:31 PM |  | Starting BOINC client version 7.2.33 for windows_intelx86
10/25/2014 7:06:31 PM |  | CAL: ATI GPU 0: AMD Radeon HD 7700 series (Capeverde) (CAL version 1.4.1848, 2048MB, 2008MB available, 2048 GFLOPS peak)
10/25/2014 7:06:31 PM |  | CAL: ATI GPU 1: ATI Radeon HD 5700/6750/6770 series (Juniper) (CAL version 1.4.1848, 1024MB, 991MB available, 2880 GFLOPS peak)
10/25/2014 7:06:31 PM |  | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 7700 series (Capeverde) (driver version 1348.5 (VM), device version OpenCL 1.2 AMD-APP (1348.5), 2048MB, 2008MB available, 2048 GFLOPS peak)
10/25/2014 7:06:31 PM |  | OpenCL: AMD/ATI GPU 1: ATI Radeon HD 5700/6750/6770 series (Juniper) (driver version 1348.5 (VM), device version OpenCL 1.2 AMD-APP (1348.5), 1024MB, 991MB available, 2880 GFLOPS peak)
10/25/2014 7:06:31 PM |  | OpenCL CPU: Intel(R) Core(TM)2 Quad CPU    Q9400  @ 2.66GHz (OpenCL driver vendor: Advanced Micro Devices, Inc., driver version 1348.5 (sse2), device version OpenCL 1.2 AMD-APP (1348.5))
10/25/2014 7:06:31 PM |  | OS: Microsoft Windows Vista: Ultimate x86 Edition, Service Pack 2, (06.00.6002.00)

Hey! All very interestingly interesting to compare, thanks.


Happy fast crunchin',
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

All CERN vLHC@home and ATLAS@home are Scientific Linux programs (RedHat plus CERN libraries) which run on other Linux, Windows of every type from Windows XP up, Mac OS X via Virtual Box. In the Budapest BOINC workshop also developers from climateprediction.net said they are aiming in this direction, so to avoid the trouble of maintaining different executables for every different OS. Only GPU crunching on Virtual Box is not available today.
Tullio

---

Passing on a tip from a few years ago.

In one of my CUDA on Linux experiments I found that the "niceness" (priority) of the CUDA app was just too low. slowing down the app so that the GPU was barely faster than the CPU. I wrote a script that ran every 60 seconds (to take into account a change of unit) that "reniced" the CUDA tasks

It can be found here

With the changes in the Linux kernel, drivers and apps over the last 3 years it may be obsolete, but if you find your Linux apps running slower than expected it could be worth a try.

T.A.

Tullio,
Thanks, very interesting.

All CERN vLHC@home and ATLAS@home are Scientific Linux programs (RedHat plus CERN libraries) which run on other Linux, Windows of every type from Windows XP up, Mac OS X via Virtual Box. In the Budapest BOINC workshop also developers from climateprediction.net said they are aiming in this direction, so to avoid the trouble of maintaining different executables for every different OS. Only GPU crunching on Virtual Box is not available today.
Tullio

---

merle - vote yes for freedom of speech

Well fans, it appears Vista has closed the Gap, All tasks for Computer 6797524
I think it's safe to say the 6770 is running around 33xx seconds while the 7750 is around 59xx seconds. This is very close to the times in Ubuntu, both of which are much better than the cards performance without the oclFFT_plan.
It's impressive such gains can be had from a simple setting...

As for the priority setting in Linux, back when I was running a CUDA card I found this setting worked to keep the card near it's peak;

<cc_config>
<log_flags>
</log_flags>
<options>
<no_priority_change>1</no_priority_change>
</options>
</cc_config>

I still use it with my ATI cards in Linux and OSX.