Quadro Review : 4x AGP vs 8x AGP

[Anonymous]

Hello Group,

I came across this on the net, it is a comparison between a Quadro XGL 900
and Quadro XGL 980. The only significant difference is the AGP 8x ( AGP
3.0 ) on the 980. Nvidia did also redesign the PCB somewhat.

http://www.xbitlabs.com/video/3dsmax5-980xgl/

The results in a nutshell: AGP 8x does not offer any statistically
significant performance increase for your dollars, at this time.

Why not? I would venture a suggestion that the AGP path is not the
bottleneck of the system. Most likely the CPU and memory speeds are to
blame, possibly the GPU is saturated with data. Also AGP 3.0 new revisions
have not been implimented in the current software you are using. Most likely
AGP 8x will offer increased performance as the other bottlenecks of your
Workstations catch up with these new videocards.

----------
Ashley Fulks
Production Manager @ www.nisku.ca
Specialized Supercomputers for Inventor
http://www.nisku.ca/autocomp.htm

[Anonymous]

On Tue, 17 Dec 2002 23:34:12 -0800, "Ashley Fulks"
wrote:

>http://www.xbitlabs.com/video/3dsmax5-980xgl/
>
>The results in a nutshell: AGP 8x does not offer any statistically
>significant performance increase for your dollars, at this time.
>
>Why not? I would venture a suggestion that the AGP path is not the
>bottleneck of the system. Most likely the CPU and memory speeds are to
>blame, possibly the GPU is saturated with data. Also AGP 3.0 new revisions
>have not been implimented in the current software you are using. Most likely
>AGP 8x will offer increased performance as the other bottlenecks of your
>Workstations catch up with these new videocards.

The reason that 8x is pretty much a nonissue has to do more with the
kind of data that really travels over the AGP bus vs. what is taken
care of on the card natively. With today's 3D cards (i.e., any card
that has a GPU on board) the card is doing the grunt work in
calculating triangles and lighting, and because many cards have 128MB
or more, they can also handle on their own a decent amount of texture
information onboard rather than going to system memory via AGP. But
with games using a vast quantity of textures, there is very little if
any hard 3D data going across the AGP bus, it's mostly texture info.

However, today's game software and that used in benchmarks use
compressed textures and low-res textures, because they are highly
tuned for dealing with low amounts of RAM on video cards, historic AGP
bottlenecks and other limiting issues.

I think that for the foreseeable future, AGP is less the bottleneck
simply because of history, rather than any major hardware-related
reason. And for CAD and 3D related work, where texture information is
loads less than that of a decent game, you're simply not going to be
pushing the limits of AGP 4x anytime soon.

Matt
mstachoni@comcast.net
mstachoni@bhhtait.com

[Anonymous]

I've noticed that the CPU appears to be a definite bottleneck in the system
when using CAD software. I'm curious to see how it will perform when 64-bit
systems become the norm. Bus "width" and speed are definitely areas of
concern in performance as well.


"Matt Stachoni" wrote in message
news:3ij30v86r6m35e9emrtb3ivr9434frd15c@4ax.com...
> On Tue, 17 Dec 2002 23:34:12 -0800, "Ashley Fulks"
> wrote:
>
> >http://www.xbitlabs.com/video/3dsmax5-980xgl/
> >
> >The results in a nutshell: AGP 8x does not offer any statistically
> >significant performance increase for your dollars, at this time.
> >
> >Why not? I would venture a suggestion that the AGP path is not the
> >bottleneck of the system. Most likely the CPU and memory speeds are to
> >blame, possibly the GPU is saturated with data. Also AGP 3.0 new
revisions
> >have not been implimented in the current software you are using. Most
likely
> >AGP 8x will offer increased performance as the other bottlenecks of your
> >Workstations catch up with these new videocards.
>
> The reason that 8x is pretty much a nonissue has to do more with the
> kind of data that really travels over the AGP bus vs. what is taken
> care of on the card natively. With today's 3D cards (i.e., any card
> that has a GPU on board) the card is doing the grunt work in
> calculating triangles and lighting, and because many cards have 128MB
> or more, they can also handle on their own a decent amount of texture
> information onboard rather than going to system memory via AGP. But
> with games using a vast quantity of textures, there is very little if
> any hard 3D data going across the AGP bus, it's mostly texture info.
>
> However, today's game software and that used in benchmarks use
> compressed textures and low-res textures, because they are highly
> tuned for dealing with low amounts of RAM on video cards, historic AGP
> bottlenecks and other limiting issues.
>
> I think that for the foreseeable future, AGP is less the bottleneck
> simply because of history, rather than any major hardware-related
> reason. And for CAD and 3D related work, where texture information is
> loads less than that of a decent game, you're simply not going to be
> pushing the limits of AGP 4x anytime soon.
>
> Matt
> mstachoni@comcast.net
> mstachoni@bhhtait.com