Path: csiph.com!news.swapon.de!fu-berlin.de!uni-berlin.de!not-for-mail
From: Terry Reedy <tjreedy@udel.edu>
Newsgroups: comp.lang.python
Subject: Re: Intel Distribution for Python
Date: Tue, 10 May 2016 23:47:49 -0400
Lines: 53
Message-ID: <mailman.586.1462938486.32212.python-list@python.org>
References: <31116921-144f-498e-873f-63510c6b8c63@googlegroups.com> <CAPTjJmrn4NmvdKpo6+D+7LPD0hRaSTSFmMRxxrHgpcJ7iJy7Lw@mail.gmail.com> <ngua18$e9v$1@ger.gmane.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: quoted-printable
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.0
In-Reply-To: <CAPTjJmrn4NmvdKpo6+D+7LPD0hRaSTSFmMRxxrHgpcJ7iJy7Lw@mail.gmail.com>
Precedence: list
Xref: csiph.com comp.lang.python:108515

On 5/10/2016 11:12 PM, Chris Angelico wrote:
> On Wed, May 11, 2016 at 12:00 PM, beliavsky--- via Python-list
> <python-list@python.org> wrote:
>> The Intel Distribution for Python 2017 Beta https://software.intel.com=
/en-us/python-distribution is available for Windows, Linux, and Mac OS fo=
r Python 2.7 and 3.5.
>>
>> "The Beta product adds new Python packages like scikit-learn, mpi4py, =
numba, conda, tbb (Python interfaces to Intel=C2=AE Threading Building Bl=
ocks) and pyDAAL (Python interfaces to Intel=C2=AE Data Analytics Acceler=
ation Library). The Beta also delivers performance improvements for NumPy=
/SciPy through linking with performance libraries like Intel=C2=AE MKL, I=
ntel=C2=AE Message Passing Interface (Intel=C2=AE MPI), Intel=C2=AE TBB a=
nd Intel=C2=AE DAAL."
>>
>> I just installed Intel Python today. Has anyone tried it? Does it run =
your programs faster than the usual CPython?
>>
>
> I haven't used it, but based on a reading of their blurbs, I suspect
> you won't see any significant improvement in base Python code - the
> advantage is the numeric computation work.

All their benchmarks are heavy numeric computation.

> For general Python performance, check out PyPy, although it tends to
> lag behind CPython in versions somewhat. However, PyPy doesn't do
> anything for your numpy performance, and doesn't even guarantee that
> everything works:
>
> http://pypy.org/download.html#installing-numpy
>
> So if the Intel Math Kernel Library lives up to the descriptions, it
> might be the thing to fill in this gap - "use PyPy to speed up your
> Python code, or Intel Python to speed up your numpy code". (Most
> programs won't have performance issues on both of those at once.)
>
> It's worth noting that Intel engineers have been proposing a number of
> performance improvements for backporting into the CPython core, and
> some of them even made it into 2.7. I've no idea whether the page you
> linked to is part of the same project or not; it might be completely
> separate, or it might be the source of all those improvements.

I presume it is all related.  What I find interesting is that Intel=20
thinks it will be more profitable to be involved in numerical=20
computation driven by python than not to.



--=20
Terry Jan Reedy