optimized sources

Forum > Windows

optimized sources

<< < (161/179) > >>

_heinz:

--- Quote from: Claggy on 27 Mar 2012, 05:29:52 am ---_heinz, can you do an updated AP v6 Atom app when you have a chance please, while r555 & r557 are faster than r409 on an Atom, they are nowhere near as fast as your special Atom builds,
See the FFTW 3.3.1 static library development thread for benches on my Atom N450

Claggy

--- End quote ---
Hi Claggy, have seen it... be patient...have a lot around my ears at the moment.

Claggy:

--- Quote from: _heinz on 27 Mar 2012, 06:29:35 am ---
--- Quote from: Claggy on 27 Mar 2012, 05:29:52 am ---_heinz, can you do an updated AP v6 Atom app when you have a chance please, while r555 & r557 are faster than r409 on an Atom, they are nowhere near as fast as your special Atom builds,
See the FFTW 3.3.1 static library development thread for benches on my Atom N450

Claggy

--- End quote ---
Hi Claggy, have seen it... be patient...have a lot around my ears at the moment.

--- End quote ---
O.K, thanks, i didn't know if you had seen it or not,

Claggy

_heinz:
How does this special astropulse v6 build for ATOM CPU's run shows the following quick test of ap_6.01r557_SSE3_ATOM_IXE_MKLS_O3.exe
More results in the Beta test area.

Bench results file R3600-20120328-0439-benchAP.txt
stored in .\Testdatas\ directory.

Quick timetable

WU : AP_single_pass.wu
ap_5.05r409_SSE.exe -verbose :
Elapsed 2118.152 secs
CPU 2110.569 secs
ap_5.05r468_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 1532.965 secs, speedup: 27.63% ratio: 1.38x
CPU 1528.560 secs, speedup: 27.58% ratio: 1.38x
ap_6.01r548_SSE_331_noAVX.exe -verbose :
Elapsed 2155.741 secs, speedup: -1.77% ratio: 0.98x
CPU 2145.170 secs, speedup: -1.64% ratio: 0.98x
ap_6.01r557_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 1530.656 secs, speedup: 27.74% ratio: 1.38x
CPU 1524.067 secs, speedup: 27.79% ratio: 1.38x

WU : Raistmer's_tiny.wu
ap_5.05r409_SSE.exe -verbose :
Elapsed 932.693 secs
CPU 926.724 secs
ap_5.05r468_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 650.598 secs, speedup: 30.25% ratio: 1.43x
CPU 646.219 secs, speedup: 30.27% ratio: 1.43x
ap_6.01r548_SSE_331_noAVX.exe -verbose :
Elapsed 850.933 secs, speedup: 8.77% ratio: 1.10x
CPU 846.961 secs, speedup: 8.61% ratio: 1.09x
ap_6.01r557_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 629.632 secs, speedup: 32.49% ratio: 1.48x
CPU 626.048 secs, speedup: 32.45% ratio: 1.48x

WU : sigind_v5.wu
ap_5.05r409_SSE.exe -verbose :
Elapsed 4540.926 secs
CPU 4508.039 secs
ap_5.05r468_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 3578.609 secs, speedup: 21.19% ratio: 1.27x
CPU 3539.101 secs, speedup: 21.49% ratio: 1.27x
ap_6.01r548_SSE_331_noAVX.exe -verbose :
Elapsed 4875.686 secs, speedup: -7.37% ratio: 0.93x
CPU 4810.275 secs, speedup: -6.70% ratio: 0.94x
ap_6.01r557_SSE3_ATOM_IXE_MKLS_O3.exe -verbose :
Elapsed 3355.997 secs, speedup: 26.09% ratio: 1.35x
CPU 3340.746 secs, speedup: 25.89% ratio: 1.35x

======================================
_heinz ;)

_heinz:

--- Quote from: _heinz on 27 Mar 2012, 05:10:34 am ---
--- Quote from: Mike on 26 Mar 2012, 04:17:59 pm ---
--- End quote ---
what I have seen its is no big difference if i suspend work on all 8 cpus or not.
will make some measurement in the next days to document it.

--- End quote ---
some new measurement for V8-Xeon
~300W if I switch on the machine, selftest, no OS loaded
~286W windows is ready, logonscreen
~830W crunching distributed network on all 3 grapicadapters and run on 8 cpu's primegrids Sophie Germain (LLR)
~743W suspend work on CPU's, still GPU work is done
~370W suspend work on GPU's, wait till all fans are down, allow work on CPU's
means primegrid Sophie Germain (LLR) need still ~70W-90W to run on all 8 CPU's
The crunching 3 GPU's need then 743W -300W = 443W ,because the machine's base load power is ~300W.

summary we can say 2GTX470 + 1GTX570 need 450W for crunching (GPU load 99%)

the relation between the power of 3GPU/ 8CPU is 443W / 70W = 6,33 : 1

Total power of the crunching v8-Xeon is ~840W (820 - 860)

_heinz

Navigation

[0] Message Index

[#] Next page

[*] Previous page

Go to full version