Looking for Linux x64 optimized apps

Message boards : Number crunching : Looking for Linux x64 optimized apps
Message board moderation

To post messages, you must log in.

AuthorMessage
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1852112 - Posted: 1 Mar 2017, 10:23:19 UTC

Yes I have looked and found the ones over at Lunatics, however they're dated Jan 2016. Looking at the AVX one its r3345 by Urs. The latest one on the list is a SoG beta app dated Feb 2017 (r3602). Is there anything more recent available for CPU or iGPU.

Machine is an i7 with Intel iGPU. It was a Windows machine 2 days ago. I'm trialling Linux on it. I finally got Beignet to recognise the iGPU and its done a bunch of multi-beam tasks using the stock apps. The iGPU is around the same speed as the prior Windows app but the CPU ones are somewhat slower.
BOINC blog
ID: 1852112 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1852113 - Posted: 1 Mar 2017, 10:40:56 UTC - in response to Message 1852112.  

Machine is an i7 with Intel iGPU. It was a Windows machine 2 days ago. I'm trialling Linux on it. I finally got Beignet to recognise the iGPU and its done a bunch of multi-beam tasks using the stock apps. The iGPU is around the same speed as the prior Windows app but the CPU ones are somewhat slower.

Running the iGPU generally results in poorer performance all round.
The shared memory/caches tend to result in a bottleneck, but even more so is the shared power & thermal limits. Running an AVX application with HyperThreading on will generally give better performance than running the CPU and the iGPU.
Grant
Darwin NT
ID: 1852113 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 1852141 - Posted: 1 Mar 2017, 14:43:03 UTC - in response to Message 1852112.  

Ask for Petri software. It's rocket science !
ID: 1852141 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1852156 - Posted: 1 Mar 2017, 16:58:23 UTC - in response to Message 1852141.  

Ask for Petri software. It's rocket science !


:) Thanks.

His Linux machine does not have an NVIDIA GPU and my optimisations are not in the OpenCL version.

Petri
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
ID: 1852156 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 1852186 - Posted: 1 Mar 2017, 20:50:31 UTC - in response to Message 1852156.  

Ask for Petri software. It's rocket science !


:) Thanks.

His Linux machine does not have an NVIDIA GPU and my optimisations are not in the OpenCL version.

Petri


yes...all you need to do now ... is to buy a GTX 1080 Ti ;)
ID: 1852186 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1852273 - Posted: 2 Mar 2017, 10:52:44 UTC - in response to Message 1852156.  
Last modified: 2 Mar 2017, 10:58:06 UTC

Ask for Petri software. It's rocket science !


:) Thanks.

His Linux machine does not have an NVIDIA GPU and my optimisations are not in the OpenCL version.

Petri

I do have a couple of GTX1060's under Windows. Looking forward to when the Petri special is ready for public consumption.
BOINC blog
ID: 1852273 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1852274 - Posted: 2 Mar 2017, 10:57:06 UTC - in response to Message 1852112.  
Last modified: 2 Mar 2017, 10:57:22 UTC

Looking at the AVX one its r3345 by Urs.

The stock app is r3584 so is more recent than the AVX app.
BOINC blog
ID: 1852274 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 1852280 - Posted: 2 Mar 2017, 13:34:02 UTC - in response to Message 1852274.  

Looking at the AVX one its r3345 by Urs.

The stock app is r3584 so is more recent than the AVX app.


And how do they compare ?
Does someone have used the benchmark tool ?
Any hope to see an updated AVX package ?
ID: 1852280 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1852403 - Posted: 2 Mar 2017, 23:24:23 UTC - in response to Message 1852274.  

Looking at the AVX one its r3345 by Urs.

The stock app is r3584 so is more recent than the AVX app.


For CPU stock and opt have different code bases so uncomparable just by rev number.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1852403 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1852520 - Posted: 3 Mar 2017, 11:23:56 UTC - in response to Message 1852280.  
Last modified: 3 Mar 2017, 11:24:36 UTC

Looking at the AVX one its r3345 by Urs.

The stock app is r3584 so is more recent than the AVX app.


And how do they compare ?
Does someone have used the benchmark tool ?
Any hope to see an updated AVX package ?

Can't tell at the moment because when running stock about all I was getting were vlar work. Since then opt app has picked up most normal work so it's going to take a while. Also I was running OpenCL iGPU app with stock but don't have it in app_info at the moment so that might effect results.

Question for Raistmer or Urs: Does the stock Linux iGPU app have the accuracy options that the Windows version recently received or is it likely to produce invalids.
BOINC blog
ID: 1852520 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1852637 - Posted: 3 Mar 2017, 20:47:14 UTC - in response to Message 1852273.  
Last modified: 3 Mar 2017, 20:48:29 UTC

Ask for Petri software. It's rocket science !


:) Thanks.

His Linux machine does not have an NVIDIA GPU and my optimisations are not in the OpenCL version.

Petri

I do have a couple of GTX1060's under Windows. Looking forward to when the Petri special is ready for public consumption.


Hi,
Raistmers approach is more suitable for the project (A big Thank You R.).

My version(s) of the CUDA app are here and around just to show what is possible with current hardware and extreme tweaking on a narrow HW base (latest CUDA only). And regarding the autocorrelations: the original CUDA implementation was 4x too slow(memory constrained(repack) and 2x too big(real vs. complex with afterprocessing)) . Using some mathematics any mathematician can prove my implementation is valid. Remember the shift from v6 to v7 when the autocorr was introduced? The RAC collapsed...

Since I do not have any current AMD/Radeon/ATI hardware I'm not going to put my efforts to that direction, although I can tell about the ideas behind the changes. The ideas in and beneath of the current CUDA optimizations and the software is free to use. Some of you have compiled your own code. Some based on the other and older piece of code and some of the later or even the latest.

What I'd like to see in the future is that a mind with the interest of "How does this really work?" would carry on the development of the software. Could we find mentors all over the time zones and language barriers to give 24/7 support?

It does not take much to become a new expert on CUDA or OpenCL. They are both relatively new techniques. The other is more open and the other is more "close to the hardware". Keep in mind: The more portable your code is the faster it will run in the future. Hardware evolves. (I'm a dinosaur in a couple of years if do not evolve too.) -- That will be the echo in your mind in the future.

Petri
p.s. I'm not going away. I'll just rest whenever I choose to.
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
ID: 1852637 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1853353 - Posted: 6 Mar 2017, 1:02:08 UTC - in response to Message 1852186.  

Ask for Petri software. It's rocket science !


:) Thanks.

His Linux machine does not have an NVIDIA GPU and my optimisations are not in the OpenCL version.

Petri


yes...all you need to do now ... is to buy a GTX 1080 Ti ;)


. . Or even a nice little GTX1050ti when topped with Special sauce.

. . MSY had a special on yesterday ... GTX750ti-2GB (ASUS Strix)for $99 AUD . two of those with Special Sauce would have really crunched some numbers :)

Stephen
ID: 1853353 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 1859180 - Posted: 2 Apr 2017, 12:24:34 UTC

An update on this. I have swapped all my iGPU machines over to Linux. While I have beignet installed I haven't defined an OpenCL app in my app_info as it slows the rest of the CPU tasks down. For the moment they're running as CPU only rigs. I have defined the astropulse app as well but AP work units are as rare as hens teeth :-)
BOINC blog
ID: 1859180 · Report as offensive

Message boards : Number crunching : Looking for Linux x64 optimized apps


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.