Re: [eigen] Updating the Benchmarks

[Gael Guennebaud <gael.guennebaud@xxxxxxxxx>]

oh actually I was already using image magic to produce both the png
and pdf versions from the postscript given by gnuplot. I've fixed the
conversion commands to get nicer PNGs. So this issue is solved.

gael

On Mon, Jan 31, 2011 at 1:59 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> Since we require LaTeX anyway for the docs, we may as well go through
> PS for the conversion. So we could do pdf2ps and then I'm sure there
> must be a ps2png tool.
>
> Benoit
>
> 2011/1/31 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
>> I agree with Benoit, some of your results look weird. You should
>> install an optimized package or re-build or remove it.
>>
>> Regarding the quality of the plots, yes I know the png files looks
>> very bad. Looks like the png renderer of gnuplot is very bad. Perhaps
>> one option would be to generate the png versions from the pdf files
>> using imagemagick if it is installed. I'll see what I can do.
>>
>>
>> gael
>>
>>
>> On Sun, Jan 30, 2011 at 2:23 AM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>>> Last spam, I promise -- but the dotted/dashed lines seem to
>>> consistently render poorly, so maybe we should only use solid lines?
>>>
>>> And why the gray background? Just use a white background!
>>>
>>> Benoit
>>>
>>> 2011/1/29 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
>>>> Unrelated remark, can we please use a better set of colors for the
>>>> curves? Especially for Goto, the gray dots are looking bad, and let's
>>>> reserve gray/black colors for ourselves (eigen3 and eigen3_novec).
>>>>
>>>> Benoit
>>>>
>>>> 2011/1/29 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
>>>>> 2011/1/29 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
>>>>>> 2011/1/29 Milian Wolff <mail@xxxxxxxxxx>:
>>>>>>> Gael Guennebaud, 26.01.2011:
>>>>>>>> btw, if you plane to bench matrix size up to 9000x9000, make sure you
>>>>>>>> disable the non optimized libraries like GMM, ublas, blitz, etc.
>>>>>>>> because they will take forever. Also make sure you disable the result
>>>>>>>> verifications when running the tests (BTL_CONFIG="--nocheck).
>>>>>>>>
>>>>>>>>
>>>>>>>> BLAS and lapack libs are very very painful regarding
>>>>>>>> configuration/linking Depending on how the libs have been generated
>>>>>>>> you might need very different settings. That's why you have these
>>>>>>>> trouble. That's also a non negligible advantage of Eigen.
>>>>>>>
>>>>>>> http://users.physik.fu-berlin.de/~milianw/eigen/bench-eigen3/i7/
>>>>>>>
>>>>>>> I've finally done the benchmarks on my i7 machine. Should I add that to the
>>>>>>> wiki?
>>>>>>
>>>>>> this is great, thanks!
>>>>>>
>>>>>> but wait... we /beat/ intel MKL???
>>>>>>
>>>>>> http://users.physik.fu-berlin.de/~milianw/eigen/bench-eigen3/i7/matrix_matrix.png
>>>>>>
>>>>>> this is awesome but... we need to triple-check this before we publish.
>>>>>>
>>>>>> MKL and GOTO are normally very close to each other, and close to the
>>>>>> theoretical maximum. In your results, MKL does surprisingly poorly.
>>>>>>
>>>>>> Is there something wrong in your MKL setup?
>>>>>
>>>>> And, same thing about ATLAS. Earlier this year, we were head-to-head
>>>>> with ATLAS. Why is it performing so poorly on your setup? Why is it
>>>>> performing exactly like eigen3_novec? Is your ATLAS built without SSE
>>>>> support?
>>>>>
>>>>> Benoit
>>>>>
>>>>>
>>>>>>
>>>>>> Benoit
>>>>>>
>>>>>>>
>>>>>>> Bye
>>>>>>> --
>>>>>>> Milian Wolff
>>>>>>> mail@xxxxxxxxxx
>>>>>>> http://milianw.de
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>>
>>
>>
>>
>
>
>