[CentOS] Centos laptop:: video cards

[Valeri Galtsev]

On Sun, October 5, 2014 4:57 am, ken wrote:
>
>
> On 10/05/2014 04:58 AM ken wrote:
>> On 10/05/2014 04:02 AM John R Pierce wrote:
>>> On 10/5/2014 12:48 AM, ken wrote:
>>>>
>>>> I sincerely *hope* that it isn't some kind of trend that video cards
>>>> are using shared memory instead of dedicated memory on the card
>>>> itself.  All machines I've bought or built  since the late '90s have
>>>> had video cards with a .5G of dedicated memory.  This is mostly
>>>> because video memory is physically different, using static RAM rather
>>>> than dynamic RAM. The former is something like ten times faster than
>>>> the latter.
>>>
>>> NO video card uses static ram, at least not since the early 1980s.
>>
>> Perhaps you're intimately familiar with each and every video card
>> manufactured since the early '80s except for the ones I bought with my
>> machines, because I've always insisted on video cards with static RAM.
>> Or perhaps your understanding of static RAM is different from what I'm
>> talking about.
>>
>>
>>>
>>> the modern CPUs with integrated graphcis controllers such as the Intel
>>> HD4500 stuff is excellent, at least on MS Windows systems. the main
>>> memory controller on these CPUs has HUGE bandwidth, the video display
>>> overhead is lost in the noise unless maybe you're running dual huge
>>> screens.   a dedicated controller might be 2-3X faster or more at 3D
>>> gaming graphics, but its not usefully faster at normal desktop
>>> graphics.   dedicated controllers use significantly more battery power
>>> than integrated ones, a consideration on a portable laptop.
>>
>> It would be nice to have authoritative sources for these opinions.
>>
>> Also, the speed of a video card is going to depend a lot on the
>> instruction set provided by the particular card and and then also very
>> much on how well the software/drivers make use of that instruction set.
>>   Those factors are going to vary widely, which is why I spoke only to
>> the speed of the *memory*.  So saying "a dedicated controller might be
>> 2-3X faster or more at 3D" is meaningless, like saying 'a car with ABC
>> tires might be faster....'
>>
>> Dynamic RAM actually uses *more* electricity than static RAM.
>
> Here are some sources which support the statement above that dynamic RAM
> uses more electricity than static RAM, making static RAM more suitable
> for use in laptops and other situations where power consumption is an
> important consideration:
>
> <http://computer.howstuffworks.com/question452.htm>
>
> <http://www.wisegeek.org/what-is-the-difference-between-static-ram-and-dynamic-ram.htm#didyouknowout>
>
> <http://en.wikipedia.org/wiki/Static_random-access_memory>
>

To put this into layman's language... No wrong, to give in simple words
what we learned in electrical engineer course ;-)

Static RAM uses CMOS 2 state cell (flip-flop). Therefore it contains at
least 4 MOS FET transistors (2 complimentary pairs). CMOS 2 state cell
only consumes energy in transition between states (and needs energy to
charge/discharge related to gate capacitance). The rest of the time it is
virtually zero energy consumption.

Dynamic RAM uses as memory cell capacitance, and hence the minimum number
of FET transistors per memory cell can be diminished to 1. This reduces
the cost and area on the chip (or increases density). But you have to
refresh charge of these capacitors. Hence, you constantly spend energy on
maintaining memory content.

Also, purely on physical principle: changing state of dynamic RAM cell
requires charging or discharging capacitor (which must be larger than
stray capacitances, as it needs to keep charge for some time, and in
switching process it shouldn't change much because something else changes
which is coupled to it via stray capacitance). This leads to [correct]
conclusion that static memory will switch between states faster, and will
cost less energy spent on switching.

So it all finally boils down to choice of dynamic RAM based on:

Less hardware and higher density on the chip (hence less cost per bit)

at the expense of

Higher energy consumption (including consumption just to maintain the
state without changing content), slower speed, and extra complexity
elsewhere (hardware to maintain state: scan through address rows or
columns, whichever...)

If I were designing small low energy consumption in stand by state and
switching often from stand by to run state device (smart phone?), I would
go with static RAM....

... Going back to my original curiosity: Are there laptops with AMD/ATI
dedicated video memory chips, and how are they compared to similar NVIDIA
ones? Anybody?

Valeri

++++++++++++++++++++++++++++++++++++++++
Valeri Galtsev
Sr System Administrator
Department of Astronomy and Astrophysics
Kavli Institute for Cosmological Physics
University of Chicago
Phone: 773-702-4247
++++++++++++++++++++++++++++++++++++++++