[Split] PowerSpectrum Unit Test

[Jason G]

Well, to get that 30-50% speedup (1.5-2x) on the small GPU, we went a bit further than what the nVidia documentation specifies for efficient reductions, and the code 'looks nice' (a good sign in engineering)... Still the larger sizes to go, might have to send some notes back to nVidia after we finish this, to update the optimisation manual a bit 

[Jason G]

looks like a issue ?

Not 'our' problem    see what msi afterburner says (for memory),  Maybe they confuse ION & ION2, don't know

[Jason G]

First post updated:

Update: PowerSpectrum(+summax reduction) Test #8 - 'Sanity check'
- Check of all needed reduction sizes
- minimal changes to larger sizes, larger than selected thrds/blk is 'almost' stock (but a bit better)
- Looking for any hardware that could yield [BAD] instead of [OK] on some sizes, particularly around selected thrds/blk
- Don't need full results, just confirmation all [OK] & no Opt1 'worst case' slower than stock
- Intend to integrate FFTs next, so this is a critical sanity check.
- having all sizes it's a longer run, and may require several runs to see if a '[BAD]' will manifest.

Please test repeatedly on all Cuda enabled GPUs... No posting of results please (too large for me to look through, I'll go crosseyed  ), just confirm all Opt1 [OK] & faster at all sizes, And alert if you see and marked [BAD] or too slow, may need to run several times to see if a problem appears or not.

Jason

[glennaxl]

All systems are go except....

gtx 295
core 0 - 1 bad at test 1/5 under 128 size
core 1 - 1 slow at test 2/5 under 128 size

gtx 260 - 1 slow at test 4/5 under 256 size

[Jason G]

All systems are go except....

gtx 295
core 0 - 1 bad at test 1/5 under 128 size
core 1 - 1 slow at test 2/5 under 128 size

gtx 260 - 1 slow at test 4/5 under 256 size

Thanks!  on the 295 is the Video memory OC'd ?  I found here the Opt1 around size #thrds/block(256 on Fermi, 128 on 2xx)  can be unstable if VRAM OC is pushed.  I had to back off my Video memory OC by 80MHz for it to stabilise

GTX260 - Please run that one a few times & see if that's consistently slower than stock at size 256.  Will be checking that code  in the meantime.
[Edit:] I see you did, & got one slow out of 5 ... OK


[Edit2:] Darn 128 still a little unstable here too  , will dial size 128 & 256 back & replace the test shortly (might be pushing a tad hard )
Jason

[Jason G]

@glenaxl: have updated the PowerSpectrumTest8 archive attached to first post, to dial back the borderline kernels a bit (for now, will dig deeper into those later if needed).

Jason

[glennaxl]

@glenaxl: have updated the PowerSpectrumTest8 archive attached to first post, to dial back the borderline kernels a bit (for now, will dig deeper into those later if needed).

Jason

Yah, my gtx 295 vram is oc'd to 1080 from 999


The new test8 are all good now. Perfect! 

[Jason G]

The new test8 are all good now. Perfect! 

Good, good.  will keep those ones dialled in a bit then, allowing some possible fine tuning later.  It seems cramming that much data through we're beginning to find weak spots, so will look at moving onto FFT integration.

[_heinz]

Hi Jason,
Device: GeForce GTX 470, 810 MHz clock, 1248 MB memory.
Stock best result
 PS+SuMx( 32768) [OK]   12.7 GFlops   50.7 GB/s

Opt best result
 PS+SuMx( 32768)   16.4   65.7 121.7 [OK]   27.8  111.4 121.7

all others are ok

[_heinz]

Hi Jason,
excellent performance on the ION
worth to post full result
PowerSpectrumTest8.exe -device 0

Device: ION, 1161 MHz clock, 242 MB memory.
Compute capability 1.1
Compiled with CUDA 3020.
                PowerSpectrum+summax Unit test #8 (Sanity Check)
Stock:
 PS+SuMx(      [OK]    0.4 GFlops    1.6 GB/s
 PS+SuMx(    16) [OK]    0.3 GFlops    1.5 GB/s
 PS+SuMx(    32) [OK]    0.3 GFlops    1.1 GB/s
 PS+SuMx(    64) [OK]    0.4 GFlops    1.8 GB/s
 PS+SuMx(   128) [OK]    0.7 GFlops    2.7 GB/s
 PS+SuMx(   256) [OK]    0.8 GFlops    3.4 GB/s
 PS+SuMx(   512) [OK]    1.1 GFlops    4.3 GB/s
 PS+SuMx(  1024) [OK]    1.1 GFlops    4.4 GB/s
 PS+SuMx(  2048) [OK]    1.2 GFlops    4.9 GB/s
 PS+SuMx(  4096) [OK]    1.2 GFlops    4.8 GB/s
 PS+SuMx(  8192) [OK]    1.3 GFlops    5.2 GB/s
 PS+SuMx( 16384) [OK]    1.3 GFlops    5.1 GB/s
 PS+SuMx( 32768) [OK]    1.3 GFlops    5.4 GB/s
 PS+SuMx( 65536) [OK]    1.4 GFlops    5.4 GB/s
PS+SuMx(131072) [OK]    1.4 GFlops    5.6 GB/s

Opt1: 64 thrds/block
                        worst case              best case
                   GFlps  GB/s ulps         GFlps  GB/s ulps
 PS+SuMx(         0.6    2.5 121.7 [OK]    0.7    2.9 121.7
 PS+SuMx(    16)    0.6    2.4 121.7 [OK]    0.6    2.7 121.7
 PS+SuMx(    32)    0.6    2.3 121.7 [OK]    0.6    2.4 121.7
 PS+SuMx(    64)    0.7    2.8 121.7 [OK]    0.7    3.0 121.7
 PS+SuMx(   128)    0.7    2.7 121.7 [OK]    0.7    3.0 121.7
 PS+SuMx(   256)    0.9    3.5 121.7 [OK]    1.0    3.9 121.7
 PS+SuMx(   512)    1.1    4.5 121.7 [OK]    1.2    5.0 121.7
 PS+SuMx(  1024)    1.2    4.6 121.7 [OK]    1.3    5.1 121.7
 PS+SuMx(  2048)    1.3    5.3 121.7 [OK]    1.5    5.9 121.7
 PS+SuMx(  4096)    1.3    5.0 121.7 [OK]    1.4    5.6 121.7
 PS+SuMx(  8192)    1.4    5.5 121.7 [OK]    1.5    6.1 121.7
 PS+SuMx( 16384)    1.3    5.4 121.7 [OK]    1.5    6.0 121.7
 PS+SuMx( 32768)    1.4    5.7 121.7 [OK]    1.6    6.4 121.7
 PS+SuMx( 65536)    1.4    5.8 121.7 [OK]    1.6    6.5 121.7
PS+SuMx(131072)    1.2    4.8 121.7 [OK]    1.7    6.6 121.7

.
Done

[Jason G]

Yes, size 128k drops off a bit on mine too, not sure why yet.

[Raistmer]

Was able to get results for GSO9600 at last:


Device: GeForce 9600 GSO, 1700 MHz clock, 384 MB memory.
Compute capability 1.1
Compiled with CUDA 3020.
      PowerSpectrum+summax Unit test #8 (Sanity Check)
Stock:
 PS+SuMx(      [OK]    1.2 GFlops    5.4 GB/s
 PS+SuMx(    16) [OK]    1.6 GFlops    6.9 GB/s
 PS+SuMx(    32) [OK]    1.8 GFlops    7.3 GB/s
 PS+SuMx(    64) [OK]    2.9 GFlops   11.8 GB/s
 PS+SuMx(   128) [OK]    4.3 GFlops   17.1 GB/s
 PS+SuMx(   256) [OK]    5.5 GFlops   22.1 GB/s
 PS+SuMx(   512) [OK]    6.7 GFlops   27.0 GB/s
 PS+SuMx(  1024) [OK]    7.0 GFlops   28.1 GB/s
 PS+SuMx(  2048) [OK]    7.7 GFlops   30.8 GB/s
 PS+SuMx(  4096) [OK]    7.6 GFlops   30.4 GB/s
 PS+SuMx(  8192) [OK]    7.9 GFlops   31.6 GB/s
 PS+SuMx( 16384) [OK]    7.7 GFlops   31.0 GB/s
 PS+SuMx( 32768) [OK]    8.1 GFlops   32.5 GB/s
 PS+SuMx( 65536) [OK]    7.8 GFlops   31.3 GB/s
 PS+SuMx(131072) [OK]    8.0 GFlops   32.2 GB/s


Opt1: 64 thrds/block
                        worst case              best case
                   GFlps  GB/s ulps         GFlps  GB/s ulps
 PS+SuMx(         1.5    6.5 121.7 [OK]    4.5   19.6 121.7
 PS+SuMx(    16)    2.3    9.6 121.7 [OK]    4.8   20.0 121.7
 PS+SuMx(    32)    3.0   12.1 121.7 [OK]    4.5   18.5 121.7
 PS+SuMx(    64)    3.1   12.7 121.7 [OK]    5.4   21.7 121.7
 PS+SuMx(   128)    4.5   18.1 121.7 [OK]    5.3   21.3 121.7
 PS+SuMx(   256)    5.8   23.1 121.7 [OK]    6.5   25.9 121.7
 PS+SuMx(   512)    6.9   27.8 121.7 [OK]    7.5   30.0 121.7
 PS+SuMx(  1024)    7.3   29.1 121.7 [OK]    7.8   31.2 121.7
 PS+SuMx(  2048)    7.9   31.5 121.7 [OK]    8.4   33.6 121.7
 PS+SuMx(  4096)    7.8   31.1 121.7 [OK]    8.2   32.6 121.7
 PS+SuMx(  8192)    8.1   32.3 121.7 [OK]    8.5   33.9 121.7
 PS+SuMx( 16384)    7.9   31.5 121.7 [OK]    8.2   32.8 121.7
 PS+SuMx( 32768)    8.1   32.5 121.7 [OK]    8.6   34.6 121.7
 PS+SuMx( 65536)    5.7   22.7 121.7 [OK]    8.3   33.2 121.7
 PS+SuMx(131072)    8.2   32.6 121.7 [OK]    8.5   34.1 121.7

[perryjay]

Okay, here's test 8. Figured it would be better for me to post it rather than try to explain what I don't understand.  :8

Microsoft Windows [Version 6.1.7600]
Copyright (c) 2009 Microsoft Corporation.  All rights reserved.

C:\Users\perry>cd\test

C:\test> powerspectrumtest8.exe

Device: GeForce 9500 GT, 1848 MHz clock, 1006 MB memory.
Compute capability 1.1
Compiled with CUDA 3020.
                PowerSpectrum+summax Unit test #8 (Sanity Check)
Stock:
 PS+SuMx(      [OK]    0.7 GFlops    3.1 GB/s
 PS+SuMx(    16) [OK]    0.8 GFlops    3.2 GB/s
 PS+SuMx(    32) [OK]    0.7 GFlops    3.0 GB/s
 PS+SuMx(    64) [OK]    1.0 GFlops    4.2 GB/s
 PS+SuMx(   128) [OK]    0.8 GFlops    3.4 GB/s
 PS+SuMx(   256) [OK]    1.6 GFlops    6.6 GB/s
 PS+SuMx(   512) [OK]    2.0 GFlops    7.8 GB/s
 PS+SuMx(  1024) [OK]    2.1 GFlops    8.2 GB/s
 PS+SuMx(  2048) [OK]    2.1 GFlops    8.2 GB/s
 PS+SuMx(  4096) [OK]    2.0 GFlops    8.1 GB/s
 PS+SuMx(  8192) [OK]    2.1 GFlops    8.4 GB/s
 PS+SuMx( 16384) [OK]    2.1 GFlops    8.4 GB/s
 PS+SuMx( 32768) [OK]    0.5 GFlops    1.9 GB/s
 PS+SuMx( 65536) [OK]    0.4 GFlops    1.5 GB/s
 PS+SuMx(131072) [OK]    2.1 GFlops    8.5 GB/s


Opt1: 64 thrds/block
                        worst case              best case
                   GFlps  GB/s ulps         GFlps  GB/s ulps
 PS+SuMx(         1.1    4.8 121.7 [OK]    1.5    6.8 121.7
 PS+SuMx(    16)    1.2    5.0 121.7 [OK]    1.7    6.9 121.7
 PS+SuMx(    32)    1.2    5.0 121.7 [OK]    1.5    6.1 121.7
 PS+SuMx(    64)    0.5    1.9 121.7 [OK]    1.7    7.1 121.7
 PS+SuMx(   128)    0.6    2.5 121.7 [OK]    1.8    7.2 121.7
 PS+SuMx(   256)    0.6    2.3 121.7 [OK]    2.1    8.3 121.7
 PS+SuMx(   512)    2.0    8.1 121.7 [OK]    2.5   10.1 121.7
 PS+SuMx(  1024)    1.9    7.8 121.7 [OK]    2.6   10.3 121.7
 PS+SuMx(  2048)    2.1    8.6 121.7 [OK]    2.6   10.3 121.7
 PS+SuMx(  4096)    0.5    2.1 121.7 [OK]    2.5   10.0 121.7
 PS+SuMx(  8192)    2.2    8.7 121.7 [OK]    2.8   11.1 121.7
 PS+SuMx( 16384)    2.1    8.2 121.7 [OK]    2.7   10.9 121.7
 PS+SuMx( 32768)    2.2    8.8 121.7 [OK]    2.8   11.1 121.7
 PS+SuMx( 65536)    2.2    8.9 121.7 [OK]    2.8   11.2 121.7
 PS+SuMx(131072)    2.3    9.2 121.7 [OK]    2.8   11.3 121.7



C:\test>

[Jason G]

Was able to get results for GSO9600 at last:

Ouch, not much headroom between worst & best (fast GDDR3 memory  on 9600GSO IIRC).  I reckon the 64k size is an anomaly worth looking into, as with the 128k drop-off on other cards (like ION).  Thankfully that part (larger sizes) is mostly stock, so there should be plenty of tweaking possibilities.... Even if only for a GFlop here and there.

[Jason G]

Okay, here's test 8. Figured it would be better for me to post it rather than try to explain what I don't understand.  :8

Thanks, A couple of sizes choking there for whatever reason.  I think I'm going to have to improve everything from size 64&128 upward before moving onto the FFTs ... Nice that it's working with all '[OK]'