Bitcoin Forum
December 07, 2016, 08:27:48 PM *
News: Latest stable version of Bitcoin Core: 0.13.1  [Torrent].
 
   Home   Help Search Donate Login Register  
Pages: [1]
  Print  
Author Topic: Catalyst 11.12 Mh/s decrease documentation thread  (Read 2419 times)
MacT
Member
**
Offline Offline

Activity: 87


View Profile
December 19, 2011, 12:49:54 AM
 #1

 This thread is for reporting hard numbers of how many mh/s the newest catalyst is taking away from your mining. Please post your card, software, etc, with  as many numbers as you have.

 11.11 and 11.2 Catalyst both steal mining speed and may or may not fix the 100% cpu bug. Here is what I found across three cards. These numbers were obtained under win7 64-bit with guiminer and phoenix using phatk2

                                pre 11.11       immediate 11.12      rebooted 11.12     restored to pre 11.11
xfx 6950                     363                    294                      352                  380
asus 5850 cu               372                    266                      316                  373
asus 5850 cu top          364                    283                      284                  365
                                 =1109 mh/s         =843 mh/s             =952 mh/s         =1118 mh/s
                                 @ 425w               @ 405w                                        @ 436w

 column 1 is the rig before installing 11.11 or 11.12.  column 2 is after results, before a reboot, which wiped away all the  overclocks but not volt mods.
column 3 is a fresh boot with 11.12 installed and overclocks set to match the original system. column 4 is the result after copying back the 6 files needed to revert- which somehow jumped the 6950 by a extra 17 mh/s but at a heavy cost of 11 extra watts.

 Temperatures stayed within 1 degree for all tests

 The 100% cpu bug only went away IF aggression was set to 10 or lower. At aggression 13 the three instances of phoenix battled each other for 25% to 42% of cpu each. At aggression 10 cpu use fell to around 12% each instance.
1481142468
Hero Member
*
Offline Offline

Posts: 1481142468

View Profile Personal Message (Offline)

Ignore
1481142468
Reply with quote  #2

1481142468
Report to moderator
Advertised sites are not endorsed by the Bitcoin Forum. They may be unsafe, untrustworthy, or illegal in your jurisdiction. Advertise here.
1481142468
Hero Member
*
Offline Offline

Posts: 1481142468

View Profile Personal Message (Offline)

Ignore
1481142468
Reply with quote  #2

1481142468
Report to moderator
1481142468
Hero Member
*
Offline Offline

Posts: 1481142468

View Profile Personal Message (Offline)

Ignore
1481142468
Reply with quote  #2

1481142468
Report to moderator
ssateneth
Legendary
*
Offline Offline

Activity: 1288



View Profile
December 19, 2011, 11:31:11 AM
 #2

I'll confirm from past experience that anything past 10 aggression causes high CPU, but only on single card setups. Multicard setups have cpu bug no matter what for me. People say cpu bug is gone on the very newest AMD APP 2.6 + OpenCL 1.2 + Catalyst 11.12/12.1, but I can't justify the huge hash speed loss over maybe 10-15 watts since I'm already hugely underclocking my cpu from 2.4ghz to 800mhz.

Gaming PC - Sapphire 5830 Xtreme - 950/1125 24/7 - Mine off and on depending on gaming activities, about 295 mhash when not using PC @ aggression 5. No CPU bug
Spare PC - Sapphire 5830 xtreme - 980/330 - Sits outside in the winter cold, 316 mhash @ aggression 10. No CPU bug
Living room - Sapphire 5830 xtreme - 925/330 - Limited airflow/dusty. 298 mhash @ aggression 10. No CPU bug
Dedicated - Sap. 5830 @ 975/330 + Asus 5870 @ 975-1085/330 (depending on severe cold weather, heatpipes freezing) + MSI (?) 5870 @ 1085/330 - Sits outside in the winter cold. Currently ~315/~480/501mhash @ aggression 14. CPU bug persist across all aggressions. Using 11.11 on all except gaming PC which uses 12.1 but AMD APP + OpenCL has been reverted to the version that comes with 11.11.

Mousepotato
Hero Member
*****
Offline Offline

Activity: 896


Seal Cub Clubbing Club


View Profile
December 19, 2011, 07:46:04 PM
 #3

I'm running 11.12 but regressed to SDK 2.4 and I'm running at full speed on my dual GPU 5970.  But with the latest and greatest SDK (whatever came with 11.12) I was down about 50 MH/s per core.  If you want to try going back to an earlier SDK, you might want to try deleting the following files first:

C:\Windows\System32\amdocl64.dll
C:\Windows\System32\amdoclcl64.dll
C:\Windows\System32\OpenCL.dll
C:\Windows\SysWOW64\amdocl.dll
C:\Windows\SysWOW64\amdoclcl.dll
C:\Windows\SysWOW64\OpenCL.dll

For some reason these files aren't removed if you do a complete ATI uninstall (at least for me they weren't).

Mousepotato
deepceleron
Legendary
*
Offline Offline

Activity: 1470



View Profile WWW
December 20, 2011, 05:46:10 AM
 #4

There are several things that may restore performance with the new 2.6 Stream SDK/OpenCL included with these drivers:

1. compile pyopencl using newest SDK (I plan to do this when I am done goofing off...I mean done working),
  - this is precompiled in miner exe using the author's SDK version, and I would assume most miners running from source haven't recompiled their python install.

2. tune and optimize OpenCL miner cores to take advantage of new SDK features (calling the OpenCL gurus back).
  - some of the new GPU features being exposed may be beneficial for mining/SHA256 hashing

The 2.6 SDK release notes were delayed but finally came out:

Key features supported in SDK 2.6 and the Catalyst 11.12 drivers include:
• OpenCL runtime integration into Linux and Windows® Catalyst drivers.
• Inclusion of the Khronos C++ wrapper API.
• Multi-GPU support on Linux platforms.
• PX5 support.
• Preview: Support for AVX instructions on CPUs that support AVX.
• Support for FMA4 instructions in OpenCL built-in function libraries on CPUs that support
FMA4.
• Kernel reflection, query kernel parameters, and enable use of OpenCL kernels in data-driven applications.
• Support for atomic counters on APUs.
• Redesign of OpenCL run-time on CPU, significantly improving performance.
• Support for the cl_amd_media_ops2 extension, exposing hardware capabilities for
accelerating image-related processing.
• Async copies preview (set environment variable GPU_ASYNC_MEM_COPY=2 to enable).
The OpenCLTM 1.2 preview includes the following capabilities (requires 8.93.10 preview drivers):
• Host access flags for memory objects enables more efficient buffer handling.
• Pattern-based GPU buffer and image initialization eliminates need for certain buffer/image transfers.
• Memory objects migration supports early transfer of buffers in preparation for when they are needed.
• New generalized image creation API.
• Enhanced image/buffer map operations.
• OpenCL 1.2 CPU device partition, including partition of a CPU after addition to a context.
• Generalized 1D and 2D images, image arrays, and image<-->buffer interop.
The 8.93.10 preview drivers also enable use of the static C++ kernel language.
gDEBugger version 6.1 is a major improvement in performance and robustness over version 6.0.
It can be downloaded for use with this SDK from http://developer.amd.com/gDEBugger.
• Integrated with Microsoft® Visual Studio®


There is also an OpenCL Static C++ Kernel Language ExtensionThis might allow a C++ rewrite for those who want to start all over.

BusmasterDMA
Member
**
Offline Offline

Activity: 118



View Profile
December 20, 2011, 11:20:29 PM
 #5

Thank you to Mousepotato for providing the list of files to remove before downgrading the OpenCL runtime.  Previous attempts at reverting had failed to recover lost performance.

I've now regained ~20MH/s on a 5850 and ~10MH/s per 5830.

Edit:
I'd forgotten that I'd switched to the phatk kernel since it seemed to perform better with 11.12's OpenCL runtime version.  Switching back to phatk2 with the 2.4 runtime yields +5 MH/s/card.

Bears.  Beets.  Battlestar Galactica.  Bitcoin.
Pages: [1]
  Print  
 
Jump to:  

Sponsored by , a Bitcoin-accepting VPN.
Powered by MySQL Powered by PHP Powered by SMF 1.1.19 | SMF © 2006-2009, Simple Machines Valid XHTML 1.0! Valid CSS!