RayLopez99 wrote:
You would have been better off making two (or more) posts.
Your subject line would then be a little shorter
Just like the title says, has anybody tried the fanless (no fans) graphics card,
which are used at the low end (saw one at FRY's) I guess because the power
consumption is not high enough to justify a fan? Are you a FAN of these cards?
I have two fanless cards. One of those cards, was not stable, until a blowing 80mm
case cooling fan was pointed at it. That's what you may need to do.
Even if you need to do that, use a separate fan mounted near the video card,
there are advantages. If that 80mm case fan of mine wears out, it only takes
seconds to replace. By comparison, try and find a replacement for the crappy
little fan that comes on normal video cards. Replacing that is not an easy job
but can be done by replacing the whole cooler.
Second question: why do you even need a graphics card if you have an i7 multi-core uP chip?
You have seven cores, you would think for apps that DON'T automatically consume all
seven cores (that is, for NON-game apps, I think), that the other 'spare' cores would be
marshaled by the uP to do graphics?
An i7 does not have seven cores. No x86 processor has seven cores. It's a poor
numbering scheme Intel came up with. (I.e. Number never matches core count.)
Use ark.intel.com if you want to know more.
Video cards, come with up to 2048 cores, or roughly 200x what the best CPU has.
Video card cores, excel at graphics operations. CPU cores excel at general purpose
computing problems (good branch capabilities). CPU cores are generally all FP capable,
whereas the graphics cards limit how many cores can do FP (floating point). For
example, if you saw in an advert "FP/16", it would mean only 2048/16 = 128 graphics
cores support 64 bit floating point arithmetic.
Or is that not how it works? I know that some chess
games have a switch that essentially says to "maximizes your PC resources" which I take to
mean the program will attempt to use all available RAM and all cores in a PC. So by definition
you won't have any 'spare' cores to do graphics? Perhaps AutoCad is the same way? So that
would explain why you need a PCI-type graphics card and not have to rely on the uP for
graphics support.
Any insight appreciated.
Software that runs on the CPU, can be single threaded or multi-threaded. Multi-threaded
software designed to work on multiple cores, will "maximize" your performance. Not all
software designs can be written that way. Some compute problems are not of the
"divide and conquer" type. To give you an idea of the split, the Adobe Photoshop
program is split into single threaded and multi-threaded filters. Even though
Photoshop deals in image processing, one of the application areas that makes
good usage of multi-threading, only half of the filters written ended up being
multi-threaded. The single-threaded ones still run on one processor core. So
the other cores on your processor are asleep. Depending on the filters you
commonly use on Photoshop, a fancy processor could be largely wasted.
I tried to open a complex AutoCad file and even with an i7 uP had
difficulty doing anything with it, hence this post. It could have been the file however,
since it was designed to stress your PC by being very complex. But I don't see why I could
not open and play with it faster than I could. I did not have a graphics card in that machine,
just relying on the Intel internal graphics co-processor.
RL
Be aware, that CAD software likes to use OpenGL. DirectX and OpenGL may both be
supported by GPU chips. The OpenGL driver on low end hardware, is hobbled on
purpose to not give good performance in CAD applications. The "certified" OpenGL
driver of a FireGL or Quadro card, may run faster. This is a money grab by the
video card manufacturers. If you look at the silicon on the FireGL or Quadro
cards, they're "regular" GPU chips, with a configuration bit set that indicates
they're to support good OpenGL operation. In the same way that a Tesla card, used
for GPGPU computing, has a config bit set to enable more of the FP (floating point)
capabilities. The silicon all starts out equal, and has configuration bits
to control market segment, and what is charged for the chip when so set.
(I.e. One GPU chip, is designed to be used in multiple video card designs.
It can be used in a gamer card, a FireGL card, a GPGPU computing card, it can
have blocks of cores disabled for differentiation, it can use cheaper slower
graphics memory and so on.)
Your Intel internal graphics co-processor, might not even have OpenGL, but if
that were the case, the CAD tool would likely complain right away. In the past,
when things like that happened (no OpenGL support), software emulation was
available with things like "Mesa".
(No - *don't* use this. This is purely an illustration. Don't run off investigating Mesa.)
http://en.wikipedia.org/wiki/Mesa_(computer_graphics)
When I researched CAD problems in the past, some students did some testing,
using their low end (non-CAD) cards, and magically, when around 50 objects
were present in their CAD window, graphics performance started to drop. So
when you open "real" CAD drawings with hundred of thousands of polys in them,
the performance will go into the toilet. If you have a FireGL or Quadro
card, less so.
http://en.wikipedia.org/wiki/Quadro
http://en.wikipedia.org/wiki/ATI_FireGL
http://en.wikipedia.org/wiki/Comparison_of_ATI_graphics_processing_units#FireGL_series
Even saying all of that, when I went to the desk of a mechanical designer
at work, he was demonstrating the CAD job that took *seven hours* to load into his
computer. We paid perhaps $20K for a computer for the poor guy, and even
with the most expensive OpenGL card money could buy, he still had what I
would class as pathetic performance. He was able to give us a demo of
rotation of the 3D object he was building, but I would not call the
result "fluid-like". So if you were thinking that "spending money"
always fixes issues like this, it doesn't. I was quite disappointed when
I saw it, and felt sorry for the guy. He's a good mechie, and it's a waste
to have him using crap like that. He had enough projects on the go though,
that he never lacked for useful work to do.
Part of the problem, in cases like that, is the tool not doing a very
good job of "culling" things that cannot be seen. The software may
use the graphics card to resolve "visibility" for objects, and when
there are hundred of thousands of objects, or imported piece parts,
doing a "dumb" visibility check really slows down the video card. In the view
he was using, lots of internal bits and pieces, could never be seen, and
should have been removed from the display list.
One of the best ways to speed up any computer, is better written software,
not more expensive processors. When an assembler I was using, was sped
up by a factor of one hundred, after being rewritten, what would I have
needed to pay, to get a 100x speedup using only a new processor ? Buying hardware
to fix bad software, is not very cost effective.
Paul