PCI graphics card performance in Thinkcentre

[jm7]

I hope someone can explain what is happening on the following config:

I have an IBM Thinkcentre m55 (Core Duo 1.86 with 2Gb RAM) - it has an
onboard video controller. This machine does NOT have a PCI-EXPRESS 16 slot.

I needed to run a 3D application so I got a PCI (old standard) NVIDIA FX5200
card.

The 3D graphics features now work BUT EXTREMELY slowly. Is this because
when doing 3D it is going through the SLOWER PCI bus?

This IBM only has one PCI and one PCI-EXPRESS 1 slot on a riser.

I wonder if it is possible to get a riser card with a PC-EXPRESS 16 slot on
it?

Thanks in advance.

 

[DL]

You can get PCI cards with much better specs than the 5200, which is a low
end card.

 

[Guest]

With the PCI card installed,is the onboard controller still running or
disabled.
It needs to be installed.W/o the pci card installed,open device
mgr,uninstall the
on-board w/o restarting pc,shutdown pc,install PCI,start pc,in
xp,re-uninstall
on-board thru device mgr,do not restart pc,install software/drivers for PCI
card,when thru,restart pc.Back in xp,you'll see both devices operating.

 

[jm7]

Thanks for the reply.

The best I can find now is a FX5500 with 256Mb RAM PCI. Is this marginally
better or....

 

[jm7]

I have the monitor operating out of the PCI card. I don't think you can
disable the onboard one permanently.

Would the fact that both are running slow down the 3D graphics? or is it
because the FX 5200 is a low end card as mentioned above? I wonder if the
FX 5500 with 256 would be significantly better?

 

[DL]

If you have onboard vid & you install a vid card, either the mobo bios
automatically disables the onboard or you have to disable it - probably in
the bios.
I cannot comment on the FX5500 as I dont run any 3d, I simply recall from
reading reports on the 5200 it was described as a low end card unsuitible
for many 3d games.
Does your app not give any specs for the vid.requirement?

If you google for 'PCI Express Riser Card' you will find some in your local
Over here they can be sourced for $20 or less

 

[jm7]

The application only says NVIDIA 256 Mb card - they probably assumed PCI-E.

Is there a significant speed diff between PCI-E and PCI??

I have been looking for a riser card that has a PCI host interface and a
PCI-E slot for the card.

I wonder if such a thing exists and if it works.

 

[DL]

FX5200 PCI cards are supplied as 128mb OR 256mb
card memory is however not the only item to look at for 3d / gameing

As I stated google for PCI-E riser cards, they are available

Personally I dont game, so I cannot relate any experience in speed
differential You are I assume, meaning frame rates
What application, specifically, are you have issues with?

 

[Paul]

I hope someone can explain what is happening on the following config:

I have an IBM Thinkcentre m55 (Core Duo 1.86 with 2Gb RAM) - it has an
onboard video controller. This machine does NOT have a PCI-EXPRESS 16 slot.

I needed to run a 3D application so I got a PCI (old standard) NVIDIA FX5200
card.

The 3D graphics features now work BUT EXTREMELY slowly. Is this because
when doing 3D it is going through the SLOWER PCI bus?

This IBM only has one PCI and one PCI-EXPRESS 1 slot on a riser.

I wonder if it is possible to get a riser card with a PC-EXPRESS 16 slot on
it?

Thanks in advance.

First of all, they aren't going to put a PCI interface on a powerful
card. That is why you see cards like FX5200 and FX5500. On the Newegg
page, the X1550 with PCI, is probably the fastest thing there. (There
are some FireGL cards, but there you are paying for a certified OpenGL
driver.)

If you had a PCI Express x1 slot, that is 250MB/sec bidirectional,
versus the 133MB/sec unidirectional of the PCI bus. That means the
PCI Express x1 slot is more useful than the PCI bus slot, but not by
a lot.

But how demanding is your application ?

People certainly have managed to play games on a PCI FX5200,
but obviously not many games are going to be able to run on
more than the minimal detail settings.

As for PCI Express x1 cards, I see five listed here, and not
all of them are low profile. A couple of them are quads and
expensive. Some don't come with the necessary low profile
faceplate. The HIS one is a dual slot width card, which may
or may not be an issue. (You'd have to check the slot layout
of the computer motherboard, to see whether it would fit, or
bump into some other card.) NVS285 is classed as "professional 2D"
in an HP document. So there is not a lot of real choice here.

http://www.newegg.com/Product/Produ...E&N=2010380048 1069620108&bop=And&Order=PRICE

The ATI 1550 is a rebadged 1300. In the chart here, there
are several X1300 models near the bottom. The FX5200 would
be maybe half or less, of the least performance on the
chart.

http://www23.tomshardware.com/graphics_2007.html?modelx=33&model1=739&model2=740&chart=285

So while the X1550 might be an improvement, it won't be
worth the money to upgrade.

If you want serious horsepower, get a computer with
a PCI Express x16 slot. The single biggest mistake people
make, buying Dell/HP/Gateway etc, is picking the cheapest
machine, the one that has no proper video card expansion
slot. And it is not like the manufacturer is going to be
honest, and say "look, this machine is crippled because we
saved $5 by removing the video card slot, and you'll
soon regret buying this". If you were a relative, I would
take the components out of your computer, buy an adequate
motherboard, a retail copy of Windows, a new computer case,
and just... fix it.

To see the impact that variations in bus bandwidth make, the
authors of this Tomshardware article taped and insulated various
numbers of PCI Express lanes. They ran benchmarks with
x1, x2, x4, x8, and x16 lanes enabled. How much an application
feels the effect of limited bandwidth, really depends on how
much raw data has to be pushed across the bus.

Notice how some games work just fine when starved...

http://www.tomshardware.com/2004/11/22/sli_is_coming/page10.html

SPECviewperf, on the other hand, really feels the effects...

http://www.tomshardware.com/2004/11/22/sli_is_coming/page8.html

Paul

 

[Peter Foldes]

Andrew. GO AWAY. What are you saying to the poster. You are a complete idiot and a troll

 

[jm7]

Hi Paul,

I tried a PCI Radeon x1550 card and it almost works perfectly but certain
features don;t.

I am using DOLPHIN IMAGING's 3D program a medical 3D app which reconstructs
CAT SCANS in 3D.

I am tempted to get one of the PCI-E X1 cards - thanks for the link to it.

I wonder if it would do it. It's all about trial and error here.

John

 

[Paul]

Hi Paul,

I tried a PCI Radeon x1550 card and it almost works perfectly but certain
features don;t.

I am using DOLPHIN IMAGING's 3D program a medical 3D app which reconstructs
CAT SCANS in 3D.

I am tempted to get one of the PCI-E X1 cards - thanks for the link to it.

I wonder if it would do it. It's all about trial and error here.

John

http://www.dolphinimaging.com/new_site/imaging/imaging_requirements.pdf

I wonder what features are breaking down ? The X1300 based chips should
be DirectX 9 in hardware. PS3.0 and VS3.0 is pretty recent, feature-wise.

http://www.techpowerup.com/gpudb/

X1300 DX9.0, PS3.0, VS3.0 (Pixel and Vertex shader capable).

Perhaps you could talk to Dophinimaging tech support, and see
why the thing is not working properly. The requirements doc
above, mentions a Nvidia and an ATI card as examples, so presumably
it is not a card specific thing. Are you certain the drivers
are loaded properly ?

If you go to Start:Run and type "dxdiag" without the quotes,
the DirectX diagnostic will load. Click display. Try the DirectDraw
and Direct3D test buttons.

You can also try some version of 3DMark, a 3D benchmarking program.
There are different versions. I like the older 3DMark2001SE as a
start, since it is only a 40MB download. Some of the later ones
are huge.

http://www.majorgeeks.com/download.php?det=99 (3dmark2001se build 330)

Later versions are listed here:

http://www.majorgeeks.com/downloads4.html

The value returned is not what I'm interested in here - rather, whether
any error messages or other anomalies are apparent. For the benchmark
value to be usable, we'd need to compare it to someone else who has the
card.

Paul

 

[RalfG]

The problem is that although you apparently have one or more bottlenecks in
your system when running that application, we don't actually know where the
remaining bottleneck(s) is/are. The on-board graphic chipset of the
Thinkcenter was already 3D capable, though apparently not fast enough for
your purposes. On paper at least the specs for it seem quite high. At least
it should have worked well enough to display 3D renderings with the shared
RAM set to 256Meg for the video.

AGP and PCI Express cards are faster than PCI cards in their data transfer
rates over the bus but as far as the graphics processors and on-card memory
there is no difference between same model cards of each type. The upshot of
that being that most of the time from the user's perspective there is little
if any noticeable difference in performance between =same= GPU and RAM
configuration cards of each type. If there is a bottleneck in your system
due to the data transfer rate between the graphics card and the CPU then you
could expect a PCI-Express card, or even possibly a PCI card with more
on-board RAM, to show faster performance. If on the other hand your software
relies heavily on the CPU rather than the GPU to process significant amounts
of graphics data then a PCI Express video card wouldn't help much because
the main performance bottleneck is the CPU itself. Similarly if slow system
board RAM or data transfers to and from the harddrive are the main
performance bottlenecks then changing the video card wouldn't solve the
problem either.

Assuming that your software is the Dolphin Imaging 10 suite with 3D, the
specs state "High resolution images requires faster CPU for processing". On
the other hand, the recommended video accelerators are both essentially at
the low end in terms of performance, FX 5200 and Radeon 9550. Nothing
definitive here, but these are clues. If the software can't make use of the
dual core CPU then maybe the 35% deficit in CPU clock speed (from the
recommended 3Ghz P4) is the real bottleneck.

 

[Paul]

The problem is that although you apparently have one or more bottlenecks in
your system when running that application, we don't actually know where the
remaining bottleneck(s) is/are. The on-board graphic chipset of the
Thinkcenter was already 3D capable, though apparently not fast enough for
your purposes. On paper at least the specs for it seem quite high. At least
it should have worked well enough to display 3D renderings with the shared
RAM set to 256Meg for the video.

AGP and PCI Express cards are faster than PCI cards in their data transfer
rates over the bus but as far as the graphics processors and on-card memory
there is no difference between same model cards of each type. The upshot of
that being that most of the time from the user's perspective there is little
if any noticeable difference in performance between =same= GPU and RAM
configuration cards of each type. If there is a bottleneck in your system
due to the data transfer rate between the graphics card and the CPU then you
could expect a PCI-Express card, or even possibly a PCI card with more
on-board RAM, to show faster performance. If on the other hand your software
relies heavily on the CPU rather than the GPU to process significant amounts
of graphics data then a PCI Express video card wouldn't help much because
the main performance bottleneck is the CPU itself. Similarly if slow system
board RAM or data transfers to and from the harddrive are the main
performance bottlenecks then changing the video card wouldn't solve the
problem either.

Assuming that your software is the Dolphin Imaging 10 suite with 3D, the
specs state "High resolution images requires faster CPU for processing". On
the other hand, the recommended video accelerators are both essentially at
the low end in terms of performance, FX 5200 and Radeon 9550. Nothing
definitive here, but these are clues. If the software can't make use of the
dual core CPU then maybe the 35% deficit in CPU clock speed (from the
recommended 3Ghz P4) is the real bottleneck.

Performance = clock_rate * IPC. Core2 IPC is higher than a P4, at least on
integer. That is how a Core2 at 1.86Ghz, can be pretty close to the 3GHz P4,
for comparison.

http://www23.tomshardware.com/cpu.html

For some reason, they never bothered to bench an E6300, and it is a constant
source of annoyance. It means you have to use the E6400 entry, and scale
the numbers (for those benchmarks known to scale linearly). The owners of
E6300 processors, are hard done by these charts.

http://www23.tomshardware.com/cpu.html?modelx=33&model1=433&model2=439&chart=158

Paul

 

[jm7]

Thanks for all the detailed info.

I will have to talk to the Dolphin techs about this. it could be the the
Core Duo , PCI slot or the amount of RAM on the CARD (256 is supposed to be
OK).

I have another machine it runs on. P4 3G; PCI-E x16 with only 128Mb RAM but
it uses some of the system RAM to report up to 512Mb on the Graphics card
Geforce7300LE.

Newegg unfortunately does not ship to Australia

 

[jm7]

Thanks everyone.

in the end the Radeon x1550 did the trick. all the features in Dolphin now
work. The only glitch I fouind was in fact a dolphib software quirk.

The radeon in 3D aspects was 3 x faster than the FX 5200 which is a low end
card.

Problem finally solved.