Cat 11.2 vs Cat 11.9 on HD6950

[Raistmer]

I see insrease in elapsed time for some of real-life tasks running on Win7x64 Cat11.9 Guru3D vs Cat 11.2 Vista x86.
Some tests were done to investigate the reason:

App Name   Task name   AR   CPU time   Elapsed

Cat 11.2, BOINC suspended, -hp switch

 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   36.988   127.258
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   33.462   102.796
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   33.54   98.979
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   25.693   83.948

Cat 11.9 Guru3D , BOINC suspended, -hp switch

 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   34.351   124.02
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   30.108   83.32
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   31.091   80.777
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   23.26   77.205

Summary: Cat 11.9 Guru3D shows better (!) performance (provided OS difference has no influence)
-----------------------------------------------------------------------------------------------------------------------------------------------------

Now BOINC runs with CPU tasks, BOINC GPU suspended
Cat 11.2, -hp  (2 runs to get random error estimation)
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   37.425   135.688
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   34.913   105.933
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   36.364   102.722
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   27.612   88.256
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   38.673   139.792
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   34.695   105.814
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   35.428   101.822
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   26.005   86.514

Summary: Loaded CPU increases elapsed and CPU times for GPU app in some degree (expected result, but worth to mention only few % increase in elapsed time)

Cat 11.2, w/o -hp switch  (2 runs to get random error estimation)

 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   38.891   139.279
 MB7_win_x8>_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   36.114   106.623
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   35.724   106.205
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   28.08   89.055
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   39.437   140.192
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   37.565   111.032
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   36.13   104.521
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   27.144   89.224

Summary: with lower priority hindrance from loaded CPU is bigger (but, again, few %)

Now the same (BOINC running idle-priority CPU tasks) for Windows 7 x64 + Catalyst 11.9 Guru3D version:

-hp switch enabled:

App Name   Task name   AR   CPU time   Elapsed
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   34.476   365.609
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   29.874   444.754
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   29.687   376.782
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   22.402   81.003
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   35.646   177.591
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   32.386   166.936
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   29.531   540.635
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   23.743   131.945

w/o -hp switch:


App Name   Task name   AR   CPU time   Elapsed
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   36.083   127.934
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   33.322   87.273
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   33.384   83.407
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   21.637   364.187
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0009_v7.wu    0.008955   35.521   126.087
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0395_v7.wu    0.394768   27.581   434.969
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG0444_v7.wu    0.444184   31.356   82.137
 MB7_win_x86_SSE3_OpenCL_ATi_r374    PG1327_v7.wu    1.326684   23.463   80.48

Summary:
1) Cat 11.9 inappropriate to use when CPU busy with processing too - elapsed times can increase greatly and erraticlly.
2) app priority increase can't help with erraticly increased elapsed times when CPU busy.

So, while high CPU usage issue is fixed indeed, we still have step back re quite old Catalyst drivers.

[Mike]

Seems to be step in the right direction.

Interesting.

[Raistmer]

Seems to be step in the right direction.

Interesting.

Unfortunately not... I simple didn't post most "interesting" part yet... Look for updatres soon...

[Raistmer]

First post updated with new data.
Unfortunately, Cat 11.9 still inappropriate to use.... 

[Raistmer]

And here is the thread on AMD forum regarding this issue: http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=155039&enterthread=y

[benool]

Thanks for testing and posting the results.

So if I understand this well, in case you don't usally run 100% on your CPU, 11.9 should perform fine even faster?

[skildude]

No it appears that 11.9 is still a little slower than the 11.2

I hope the ATI guys get their act together.  I can't run 11.2 because it causes my computer to lock.  I'm forced to use 11.8 which is ok but like he said it has erratic time increases.  If they act on Raistmers information perhaps we'll see something in a couple months.

[Raistmer]

So if I understand this well, in case you don't usally run 100% on your CPU, 11.9 should perform fine even faster?

To check this assumption one should repeat same test with 1 core (of 4 in case of my quad) freed for GPU feeding, maybe 1 core of many would be not enough... depends of nature of issue - why idle-priority processes have so big influence on driver...

[Mike]

IŽve seen a couple guys keeping 1 or 2 cores free to feed GPUs more efficiently.

At least i see improvement in the right direction.
Maybe weŽll see some changes in official release already.