Catalyst 12.8 vs 13.1

Message boards : Number crunching : Catalyst 12.8 vs 13.1

To post messages, you must log in.

AuthorMessage
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1335839 - Posted: 8 Feb 2013, 16:21:06 UTC
Last modified: 8 Feb 2013, 16:22:23 UTC

I did some benchmarking of 12.8 vs 13.1.

MB r390, AP 1761

HD 6670, Phenom 945, ASRock N68C-GS FX, 8G GSkill CAS7 10666

Chart:



TLDR: MB delta runtimes 12.8->13.1 are -5% to +60%, mostly slightly slower. For the +60% wu CPU usage went up by a factor of 4.

AP is slightly slower on 13.1 and cpu usage is down.

Edit: typo
ID: 1335839 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1341159 - Posted: 27 Feb 2013, 4:53:45 UTC

Ran the same benchmarks, same equipment, for Catalyst 13.2 Beta 6:

>>>See important caveat after the graphic<<<



Most importantly when we compare dark blue to green we see improvements in runtimes for all the test wu's 12.8 -> 13.2b6. Also CPU utilization is fixed so whatever was going wrong with 13.1 has...stopped going wrong.

IMPORTANT: Catalyst 13.2b6 failed to compile the AP 1761 kernel. The results I show are for the kernel compiled by 13.1 and run with the 13.2b6 driver.
ID: 1341159 · Report as offensive
Wedge009
Volunteer tester
Avatar

Send message
Joined: 3 Apr 99
Posts: 448
Credit: 241,909,793
RAC: 90,607
Australia
Message 1341239 - Posted: 27 Feb 2013, 12:14:36 UTC - in response to Message 1341159.  

The results I show are for the kernel compiled by 13.1 and run with the 13.2b6 driver.

Are you sure it was 13.1? I'm pretty sure kernel compilation is still broken with that driver release, and that 12.8 is the most recent release that can still compile the kernel correctly.
Soli Deo Gloria
ID: 1341239 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1341252 - Posted: 27 Feb 2013, 12:52:41 UTC - in response to Message 1341239.  

The results I show are for the kernel compiled by 13.1 and run with the 13.2b6 driver.

Are you sure it was 13.1? I'm pretty sure kernel compilation is still broken with that driver release, and that 12.8 is the most recent release that can still compile the kernel correctly.

My own testing and what I have read from others says that AP 1761 works under CCC 13.1, but MB 1760 & 1761 experience driver restarts.
ID: 1341252 · Report as offensive
Wedge009
Volunteer tester
Avatar

Send message
Joined: 3 Apr 99
Posts: 448
Credit: 241,909,793
RAC: 90,607
Australia
Message 1341258 - Posted: 27 Feb 2013, 13:00:53 UTC - in response to Message 1341252.  

Ah, sorry yes, it was MB. I am one of those 'others' who've experienced that. So AP also breaks with the current beta release. This isn't a good trend.
Soli Deo Gloria
ID: 1341258 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1341261 - Posted: 27 Feb 2013, 13:11:15 UTC - in response to Message 1341258.  

Well it's good and bad, for things that run they run better with 13.2. Right now I'm using 13.2 with MB 390 and AP 1761 with the precompiled 13.1 kernel. Works for me but definitely not ready for prime time.

Vast credit to the developers, but it is possible that the problem with AP 1761 is a syntax error that snuck by previous drivers and gets caught by the new driver. But even if that's the case, language stability is important. Quirks of syntax shouldn't change especially when we remain on the same language version (1.2).

One thing I haven't yet checked is if MB 1760 works with 13.2b6. I'll do that tonight.

ID: 1341261 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 5430
Credit: 66,403,595
RAC: 12,411
Russia
Message 1341284 - Posted: 27 Feb 2013, 16:12:02 UTC

Syntax error should not kill compiler. Instead, compiler should report syntax error. Do you see such error message in stderr log or what?

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1341284 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1341288 - Posted: 27 Feb 2013, 16:26:26 UTC - in response to Message 1341284.  

At work now so I can't refer to the stderr. I got a generic message to the effect "unable to compile" and an error code, maybe -11.
ID: 1341288 · Report as offensive
Profile cov_route
Avatar

Send message
Joined: 13 Sep 12
Posts: 342
Credit: 10,267,805
RAC: 758
Canada
Message 1341416 - Posted: 27 Feb 2013, 22:30:04 UTC - in response to Message 1341288.  

Both AP 1761 and MB 1761 fail the kernel compile with CCC 13.2b6. Relevant section from stderr is:

INFO: can't open binary kernel file: .\\MultiBeam_Kernels_r1761.cl_Turks.bin_V6, continue with recompile...
Error : Building Program (source, clBuildProgram):main kernels: not OK code -11
Internal error: Compilation failed.
ID: 1341416 · Report as offensive

Message boards : Number crunching : Catalyst 12.8 vs 13.1


 
©2016 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.