Your benchmarks corelate with vram also. Have you tried 8600M with the
same vram as the ati cards?
Michael
On Feb 9, 2009, at 10:06 AM, Michael Diehr <[email protected]> wrote:
We had the chance to run a few benchmarks on our own QC
compositions. These compositions do a lot of video processing, but
not a lot of vertex processing (e.g. typically just a few billboards
with processing multiple video inputs or quicktime movies).
Summary: It's clear from these #s that the ATI cards are still more
efficient at QC / Core Image, as even a 3 years macbook pro holds it
own against the newest models.
However, it seems like the raw horsepower of the latest MBP model is
enough to overcome the NVIDIA problem.
Machine VCard Benchmarks (fps)
A B C
MBP17CD X1600 14 20 12
MBP15SR X1600 14 20 15
MPB15'06 8600M 8 -- -- (B and C were not run)
MBP15'08 9600M 17 30 20
Key:
MBP17CD 17" macbook pro Core Duo w/ATI X1600 256MB
MBB15SR 15" macbook pro Core 2 Duo w/ATI X1600 256MB
MBP15'06 15" macbook pro Core 2 Duo w/NVIDIA GeForce 8600M GT 128MB
MBP15'08 15" macbook pro Core 2 Duo w/NVIDIA GeForce 9600M GT 512MB
Benchmarks:
A DV video input run through a core image accumulator
B 2 quicktime movies (MJPEG) kaleidoscoped and overlayed
C 2 quicktime movies (MJPEG) given multiple 3d transformations.
If we have a chance, we'll repeat these on the latest iMacs, as we
are very curious how the X2600 vs 8800 perform.
On Feb 7, 2009, at 4:31 PM, Michael Diehr wrote:
Curious if anyone has re-done any QC benchmarks with the latest
9600M GTs in the Mac Book Pros, especially after the 10.5.6 and
latest graphics update(s)?
Also, any rought ideas about how a MBP with 9600M GT compares to an
iMac with either the radeon HD 2600 PRO or NVIDIA 8800 GS ?
(Trying to figure out what is the most bang-for-the-buck QC machine
today) :)
On Nov 7, 2008, at 5:56 PM, vade wrote:
I got some advice from an ATI/AMD developer, who said, only use
that pixel format if you are going from CPU -> OpenGL. If you are
doing OpenGL only processing and using FBOs for RTT and the like
you should use GL_UNSIGNED_BYTE and GL_RGBA iirc.
A friend of mine and I also noticed 10.5.5 had much improved NV
performance on his MBP 2.4Ghz with 8600M, but still lack luster in
QC compared to my older x1600.
On Nov 7, 2008, at 7:04 PM, Michael Diehr wrote:
I was part of that prior discussion, and my take-home message was
a little different -- OpenGL-based games and benchmarking
suggests the NVIDIAs are faster than the ATIs (as one would
expect, otherwise why would Apple have bothered changing?), but
there's something particular to QuartzComposer that suffers on
the NVIDIA chips. So, my conclusion was that it wasn't the
OpenGL drivers, but QC that was the problem.
Of interest, I've been playing with my own OpenGL calls, and
found some odd differences between ATI and NVDIA: namely the
advice to use
GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV
caused terrible performance on ATI, but was fine on NVIDIA.
This advice comes from:
http://developer.apple.com/documentation/GraphicsImaging/Conceptual/OpenGL-MacProgGuide/opengl_performance/chapter_13_section_2.html#//apple_ref/doc/uid/TP40001987-CH213-SW23
So, I wonder if there's some OpenGL code inside QC that is doing
something wrong, such as using the wrong pixel format?
-mike
On Nov 7, 2008, at 5:51 AM, Adrian Ward wrote:
Hi Dom,
This sounds very like the performance issues we had with various
NVIDIA GeForce 8XXXs versus the ATIs, which were particularly
exacerbated by 1) feeding a structure of images into an iterator
and 2) having multiple compositions rendering in the same
process (or so it seemed).
In the latter case, we were trying to use a Composition Loader
patch but the problem was also noticeable in QC editor by just
having two composition windows open at the same time. The frame
rate was sometimes comically low: 6 vs 60 fps! In both cases,
the ATIs clearly out-performed the NVIDIAs, often dramatically,
even when comparing the much older ATI X1900s to the newer
NVIDIA 8800s. The mobile NVIDIA chips were also effected. Our
conclusion was that the NVIDIA OpenGL drivers are not as optimal
as required, despite the hardware being more advanced and that
this problem is NVIDIA's fault and not Apple's.
Luckily, we had the luxury of dictating our host platform so we
abandoned NVIDIA at the last moment and are now rolling out
everything for ATIs.
Best,
Ade.
On 7 Nov 2008, at 13:18, Dominik Wagner tcm wrote:
Hi everyone,
we are in a little bit of a pickle here the sample code i
attached really has major performance issues on the new
NVIDIA GeForce 9600M GT OpenGL Engine (interestingly enough it
is almost as bad as the on board one) but runs really well on
the ATI of the second generation macbook pros and imacs (ATI
Radeon X1600) and we can't really figure out why.
So what we do is essentially the QuartzComposerTV sample code,
and add in on composition that runs perfectly on its own with
it's 60 fps, but totally screws up everything when drawn in the
same context together with other composisions. Which is what
this sample code does: it renders a stack of Composition into
an OpenGL context. To make it more like we really do it i put
the rendering into a extra thread and a pixelbuffer, which is
in turn put into an opengl view by drawing a quad with the
pixelbuffer as a texture. Focus of this example is how many
renderings the background thread can do, not if we display
every rendering.
I also narrowed it down to the zodiac sings composition having
many images as an input which travel down through iterators.
E.g. when instead of sending in 12 different zodiac signs send
in only one then it works nicely at high speed.
What are my questions:
- do we anyhting wrong in setting up the contexts and pixel
buffers that is hurtin on nvidia but not on ati?
- do we maybe need to do some clearing or cleanup between
rendering the compositions?
- has quartz composer performance issues on nvidia that it has
not on the ati cards?
- in our actual application we have a lot of other pixelbuffers
and contexts in place and there it hurts even more as in this
sample. seems like resource run out and textures are swapped/
uploaded every frame or something the likes of it
we are glad for any help we can get!
the sample code can be found at:
http://public.me.com/nikdo - QCTV-nvidia-problem.zip
it performs at 756x512 at ~44fps on the ati macbook pros, and
~30fps on the new macbook pro... here
best,
dom
--
Dominik Wagner Mail: [email protected]
TheCodingMonkeys http://www.codingmonkeys.de/
Blog - DasGenie: !Scrap http://scrap.dasgenie.com/
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected]
)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/adrian%40clayinteractive.co.uk
This email sent to [email protected]
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected]
)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/md03%40xochi.com
This email sent to [email protected]
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected]
)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/doktorp%40mac.com
This email sent to [email protected]
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected]
)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/md03%40xochi.com
This email sent to [email protected]
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected]
)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/lattam%40mac.com
This email sent to [email protected]
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Quartzcomposer-dev mailing list ([email protected])
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com
This email sent to [email protected]
