(OSX) hardware survey of Mac, video card, monitor

1
Dear forum,

Curious to know what hardware OSX Psychtoolbox users are using:

1. type of Mac, OS version, processor, RAM
2. type of video card driving your stimuli
3. type of CRT stimulus monitor, VGA or DVI
4. at what pixel resolution and refresh rate you run (and can run) your stimuli

Please feel free to include as much detail as you want about your set up.

A lab I am helping (to whom I have recommended OSX Psychtoolbox and
MATLAB 7.0.4) is most interested in presenting stimuli at pixel
resolution 1024 x 768 at 120 Hz refresh rate. Looking at high-end
items, assuming they would be the best and most capable, my first
choice would be:

1. Dual 2.5GHz PowerPC G5, most recent OSX, 1 GB DDR 400 SDRAM
2. ATI Radeon 9800 XT with 256 MB DDR SDRAM
3. Sony GDM-C520K or possibly Sony GDM-F500

But it is difficult from video card and monitor specs I have seen to
be certain that 1024 x 768 at 120 Hz is specifically achievable. And
if it is, maybe my hardware choices are massively excessive. So
curious to know what hardware has been successful for others. Thank
you for your input.

Daniel Shima
Vanderbilt Vision Research Center
2
Dear Daniel,

Here's some thoughts:
I am sure that any higher-end mac built since the powermac 7500 will run that resolution
and refresh rate just fine (but of course, the above system would run system 9.x and
matlab 5.2.1, PTB 2.5.x, and the special video driver).

Anything low-end mac including the the imac 233 will work as well (part experience, part
extrapolation) and should be able to run os x.

We're running most of our experiments on a powermac G4 @ 867 Mhz and it performs
nicely, but then, our stimuli are generally relatively simple.

If you need to run OS X I and want a new computer, I suspect a mac mini will in principle
do just fine (my powerbook G4 @ 800 Mhz does). The only disadvantage might be that it
will only drive a single monitor.

Note that you will benefit little from a dual processer system, as matlab will not use the
second processor. (You could, I guess, run a second copy of matlab that would run on the
second processor to e.g. calculate images in the background, but whether it's a good idea
for timing and so I can't tell).

It all depends on how much processing your stimuli require whether you will benefit from
spending much more $$$.

To give you more precise advice, I think you'll need to specify a bit the type of stimuli/
experiments that will have to run on the system(s). Since the requirements will determine
whether you should bet your money on faster processors, or better graphics cards with
e.g.. more VRAM.

Gr.
Frans


--- In psychtoolbox@yahoogroups.com, Daniel Shima <danielshima@g...> wrote:
>
> Dear forum,
>
> Curious to know what hardware OSX Psychtoolbox users are using:
>
> 1. type of Mac, OS version, processor, RAM
> 2. type of video card driving your stimuli
> 3. type of CRT stimulus monitor, VGA or DVI
> 4. at what pixel resolution and refresh rate you run (and can run) your stimuli
>
> Please feel free to include as much detail as you want about your set up.
>
> A lab I am helping (to whom I have recommended OSX Psychtoolbox and
> MATLAB 7.0.4) is most interested in presenting stimuli at pixel
> resolution 1024 x 768 at 120 Hz refresh rate. Looking at high-end
> items, assuming they would be the best and most capable, my first
> choice would be:
>
> 1. Dual 2.5GHz PowerPC G5, most recent OSX, 1 GB DDR 400 SDRAM
> 2. ATI Radeon 9800 XT with 256 MB DDR SDRAM
> 3. Sony GDM-C520K or possibly Sony GDM-F500
>
> But it is difficult from video card and monitor specs I have seen to
> be certain that 1024 x 768 at 120 Hz is specifically achievable. And
> if it is, maybe my hardware choices are massively excessive. So
> curious to know what hardware has been successful for others. Thank
> you for your input.
>
> Daniel Shima
> Vanderbilt Vision Research Center
3
Hello Daniel,

i agree with Frans thoughts - it very much depends on the kind of
stimuli and experiment:

Dual-processors don't necessarily help - they could even hurt when
you need very exact
presentation timing: Matlab and PTB don't make use of a 2nd CPU at
the moment. Scripts
that use Rush() or Priority() try to get exclusive access of the
single processor in a single
cpu system, so nothing interferes with PTB's operation. If a second
cpu is present, other
code can run on that cpu in parallel and interfere with PTB's
operation on the first cpu due
to competition for system ressources that are shared between all
processors, e.g.,
memory, bus-bandwidth, gfx-bandwidth, operating system kernel locks...

But there are instructions in the internet for shutting down the 2nd
processor if necessary.

--> For most experiments, a second processors won't help much...

More system memory (RAM) is useful if you want to load and display a
large number of
images per trial, e.g., playing long "movie clips".

A fast processor will help if the Matlab code in your trial-loop is
complex and compute
intense, or if you are generally impatient ;)

Drawing of stimuli, e.g., images via Screen('DrawTexture'), large
number of lines and dots
via 'DrawDots' or 'DrawLines' is mostly done by the graphics card,
so a fast modern
graphics card is more important than a fast processor.

If you want to display relatively short, but high image resolution
movie clips at a high
frame rate or few high-res images with very high precision timing,
you need a graphics
card with enough VRAM - images can then be preloaded before the trial
onto the gfx-card
memory and drawn from there at a higher speed than from system memory.

E.g.: If you want to draw a 1024 x 768 pure luminance image, it will
use approx. 0.8 MB of
memory per image. If you have color images, they'll use s = width *
height * 4 = approx. 3
MB of memory per image. If you have 256 MB of VRAM on your ATI card,
you'll be probably
able to use 240 MB for images == 80 images can be preloaded for very
fast drawing.

Here are some numbers from our machine ( 1.6 Ghz single processor G5,
1 GB RAM,
NVidia GeForceFX 5200 Ultra with 64 MB VRAM, latest OS-X,
"not-yet-released" faster
DrawTexture code for images with a size of 1024 x 768 pixel, tested
at a monitor refresh
rate of 120 Hz:

When playing short sequences repeatedly from VRAM, either pure
Luminance or RGB, we
get zero dropped frames over a couple of hundred samples -- perfect
presentation timing.
When playing longer sequences from system memory, we get about 4%
dropped frames,
e.g., 40 out of 1200 frames miss their deadline.

The gfx-card you proposed is more powerful so it should be able to
handle this.

Drawing many big images is the hardest case for the system, drawing
other stims like
lines, dots, filled polygons (as long as they are convex) is "easier".

When you say 1024 x 768 do you mean monitor resolution or size of
images to be
presented? Size of images (as in the benchmark above) matters much
more than the
display resolution.

So what kind of experiments/stims are they planning to do/use?

best,
-mario

--- In psychtoolbox@yahoogroups.com, Daniel Shima <danielshima@g...>
wrote:
>
> Dear forum,
>
> Curious to know what hardware OSX Psychtoolbox users are using:
>
> 1. type of Mac, OS version, processor, RAM
> 2. type of video card driving your stimuli
> 3. type of CRT stimulus monitor, VGA or DVI
> 4. at what pixel resolution and refresh rate you run (and can run)
your stimuli
>
> Please feel free to include as much detail as you want about your
set up.
>
> A lab I am helping (to whom I have recommended OSX Psychtoolbox and
> MATLAB 7.0.4) is most interested in presenting stimuli at pixel
> resolution 1024 x 768 at 120 Hz refresh rate. Looking at high-end
> items, assuming they would be the best and most capable, my first
> choice would be:
>
> 1. Dual 2.5GHz PowerPC G5, most recent OSX, 1 GB DDR 400 SDRAM
> 2. ATI Radeon 9800 XT with 256 MB DDR SDRAM
> 3. Sony GDM-C520K or possibly Sony GDM-F500
>
> But it is difficult from video card and monitor specs I have seen to
> be certain that 1024 x 768 at 120 Hz is specifically achievable.
And
> if it is, maybe my hardware choices are massively excessive. So
> curious to know what hardware has been successful for others. Thank
> you for your input.
>
> Daniel Shima
> Vanderbilt Vision Research Center