[TriEmbed] In need of a tiny, cheap, powerful(ish) computer

Carl Nobile carl.nobile at gmail.com
Mon Sep 21 21:12:18 CDT 2015


This is why I use Windows for nothing except updating my Tom-Tom. You are
right if you are using Windoze expect any Python packages that need C/C++
to compile to just not work correctly. For me this is just way too limiting
so I'm a Linux only type of guy. I do this stuff professionally, so I
decline all Windows Python work--I can't have things failing to build
correctly.

So if you get it to work on the RPi then you can recommend it and the numpy
install will always just work.

~Carl


On Mon, Sep 21, 2015 at 10:02 PM, Adam Haile <adammhaile at gmail.com> wrote:

> Yeah... I think I'll look into seeing if I can integrate numpy in any way.
> The problem I've run into with numpy is on Windows. When trying to install
> via pip, it always gives the dreaded vcvarsall.bat error and then you have
> to go an install the microsoft python compiler package to get it to work.
> Another option is to manually download a precompiled WHL file
> <http://www.lfd.uci.edu/~gohlke/pythonlibs/#numpy>, but most users don't
> find that. And I hate dependencies :P But I'm cool with making it an option
> thing for the sake of speed... fortunately, python is flexible enough where
> I could probably make it auto-detect and then just use the correct version
> if numpy is installed.
>
> On Mon, Sep 21, 2015 at 9:58 PM, Carl Nobile <carl.nobile at gmail.com>
> wrote:
>
>> I install everything in a virtual environment then use pip to install
>> everything. This keeps one apps install separate from another's. For
>> instance one app can use python 3.4 and another use 2.7.4.
>>
>> Pip should pull down all of numpy's dependencies. Don't use the machine's
>> install with most Python packages. Now scipy is a pain to install even with
>> pip. There are circular dependencies between numpy and scipy.
>>
>> You may want to see if the RPi can handle the work when using numpy, just
>> to see. This may give you other options.
>>
>> ~Carl
>>
>>
>> On Mon, Sep 21, 2015 at 9:49 PM, Adam Haile <adammhaile at gmail.com> wrote:
>>
>>> I do where I can right now... but numpy is a non-trivial install on some
>>> systems (in my experience). Something not worth it for the majority of
>>> users since most probably don't use more than a couple hundred LEDs, if
>>> that.
>>>
>>> I may need to find a way to come up with an optional numpy version...
>>>
>>> On Mon, Sep 21, 2015 at 9:45 PM, Carl Nobile <carl.nobile at gmail.com>
>>> wrote:
>>>
>>>> Adam,
>>>>
>>>> Doing it in pure python may be the issue. Use the numpy package it does
>>>> matrix math extremly fast. If you're flipping pixels around then you will
>>>> get an unbelievable speed increase with numpy.
>>>>
>>>> ~Carl
>>>>
>>>>
>>>> On Mon, Sep 21, 2015 at 9:39 PM, Adam Haile via TriEmbed <
>>>> triembed at triembed.org> wrote:
>>>>
>>>>> It's all CPU. There's usually less than 100MB RAM usage. And yup, 24/7
>>>>> operation.
>>>>> I'm already using nothing but single byte math (with a tiny few float
>>>>> exceptions) and have optimized the heck out of it with a couple of
>>>>> performance profilers. Pretty much it comes down to doing hundreds of
>>>>> thousands of pixel operations per second all in pure python, with no GPU
>>>>> acceleration... this is because my "display" is large LED matrices. So,
>>>>> admittedly, not the best for performance... but I'm also doing way more
>>>>> than users of my library would generally ever try.
>>>>>
>>>>> On Mon, Sep 21, 2015 at 5:01 PM, John Vaughters <
>>>>> jvaughters04 at yahoo.com> wrote:
>>>>>
>>>>>>
>>>>>>
>>>>>> >Forget I mentioned the video thing... Pi is not fast enough
>>>>>> regardless.
>>>>>>
>>>>>> Right, so the pi lacks memory and processing power.
>>>>>>
>>>>>> I guess the question is what is your application wanting? Are you
>>>>>> swaping memory? or is the Processor pegged? Or both?
>>>>>>
>>>>>> Do you need 24/7 on time?
>>>>>>
>>>>>> Is it your software that you can change to optimize your system? You
>>>>>> mentioned speed over cores; software changes can possibly fix these
>>>>>> problems.
>>>>>>
>>>>>> Finding sub $150 computers in the x86 family is going to be tough
>>>>>> unless you want a used laptop. But even then, laptops running 24/7 are not
>>>>>> the most reliable.
>>>>>>
>>>>>> I guess my suggestion would be to try to figure out what the hold up
>>>>>> is in your application and system if you want to get the cost to it's
>>>>>> lowest. Or just try the used laptop route.
>>>>>>
>>>>>> John Vaughters
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Triangle, NC Embedded Computing mailing list
>>>>> TriEmbed at triembed.org
>>>>> http://mail.triembed.org/mailman/listinfo/triembed_triembed.org
>>>>> TriEmbed web site: http://TriEmbed.org
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>>
>>>> -------------------------------------------------------------------------------
>>>> Carl J. Nobile (Software Engineer)
>>>> carl.nobile at gmail.com
>>>>
>>>> -------------------------------------------------------------------------------
>>>>
>>>
>>>
>>
>>
>> --
>>
>> -------------------------------------------------------------------------------
>> Carl J. Nobile (Software Engineer)
>> carl.nobile at gmail.com
>>
>> -------------------------------------------------------------------------------
>>
>
>


-- 
-------------------------------------------------------------------------------
Carl J. Nobile (Software Engineer)
carl.nobile at gmail.com
-------------------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.triembed.org/pipermail/triembed_triembed.org/attachments/20150921/78b50730/attachment.htm>


More information about the TriEmbed mailing list